Bump version to v0.6.7 (#13439)

## Summary

A tiny bit faster and the `red-knot` CLI does the same thing.

.devcontainer/devcontainer.json

@@ -20,7 +20,7 @@
 			"extensions": [
 				"ms-python.python",
 				"rust-lang.rust-analyzer",
-				"serayuzgur.crates",
+				"fill-labs.dependi",
 				"tamasfe.even-better-toml",
 				"Swellaby.vscode-rust-test-adapter",
 				"charliermarsh.ruff"

.github/renovate.json5

@@ -14,12 +14,26 @@
     rangeStrategy: "update-lockfile",
   },
   pep621: {
+    // The default for this package manager is to only search for `pyproject.toml` files
+    // found at the repository root: https://docs.renovatebot.com/modules/manager/pep621/#file-matching
     fileMatch: ["^(python|scripts)/.*pyproject\\.toml$"],
   },
   pip_requirements: {
-    fileMatch: ["^docs/requirements.*\\.txt$"],
+    // The default for this package manager is to run on all requirements.txt files:
+    // https://docs.renovatebot.com/modules/manager/pip_requirements/#file-matching
+    // `fileMatch` doesn't work for excluding files; to exclude `requirements.txt` files
+    // outside the `doc/` directory, we instead have to use `ignorePaths`. Unlike `fileMatch`,
+    // which takes a regex string, `ignorePaths` takes a glob string, so we have to use
+    // a "negative glob pattern".
+    // See:
+    // - https://docs.renovatebot.com/modules/manager/#ignoring-files-that-match-the-default-filematch
+    // - https://docs.renovatebot.com/configuration-options/#ignorepaths
+    // - https://docs.renovatebot.com/string-pattern-matching/#negative-matching
+    ignorePaths: ["!docs/requirements*.txt"]
   },
   npm: {
+    // The default for this package manager is to only search for `package.json` files
+    // found at the repository root: https://docs.renovatebot.com/modules/manager/npm/#file-matching
     fileMatch: ["^playground/.*package\\.json$"],
   },
   "pre-commit": {

.github/workflows/ci.yaml

@@ -37,7 +37,7 @@ jobs:
         with:
           fetch-depth: 0
 
-      - uses: tj-actions/changed-files@v44
+      - uses: tj-actions/changed-files@v45
         id: changed
         with:
           files_yaml: |
@@ -142,6 +142,13 @@ jobs:
 
       # Check for broken links in the documentation.
       - run: cargo doc --all --no-deps
+        env:
+          RUSTDOCFLAGS: "-D warnings"
+      # Use --document-private-items so that all our doc comments are kept in
+      # sync, not just public items. Eventually we should do this for all
+      # crates; for now add crates here as they are warning-clean to prevent
+      # regression.
+      - run: cargo doc --no-deps -p red_knot_python_semantic -p red_knot -p ruff_db --document-private-items
         env:
           # Setting RUSTDOCFLAGS because `cargo doc --check` isn't yet implemented (https://github.com/rust-lang/cargo/issues/10025).
           RUSTDOCFLAGS: "-D warnings"

.github/workflows/publish-docs.yml

@@ -34,10 +34,10 @@ jobs:
       - name: "Set docs version"
         run: |
           version="${{ (inputs.plan != '' && fromJson(inputs.plan).announcement_tag) || inputs.ref }}"
-          # if version is missing, exit with error
-          if [[ -z "$version" ]]; then
-          echo "Can't build docs without a version."
-          exit 1
+          # if version is missing, use 'latest'
+          if [ -z "$version" ]; then
+            echo "Using 'latest' as version"
+            version="latest"
           fi
 
           # Use version as display name for now
@@ -145,6 +145,7 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.ASTRAL_DOCS_PAT }}
         run: |
           branch_name="${{ env.branch_name }}"
+
           # auto-merge the PR if the build was triggered by a release. Manual builds should be reviewed by a human.
           # give the PR a few seconds to be created before trying to auto-merge it
           sleep 10

.github/workflows/release.yml

@@ -1,3 +1,5 @@
+# This file was autogenerated by cargo-dist: https://opensource.axo.dev/cargo-dist/
+#
 # Copyright 2022-2024, axodotdev
 # SPDX-License-Identifier: MIT or Apache-2.0
 #
@@ -64,7 +66,7 @@ jobs:
         # we specify bash to get pipefail; it guards against the `curl` command
         # failing. otherwise `sh` won't catch that `curl` returned non-0
         shell: bash
-        run: "curl --proto '=https' --tlsv1.2 -LsSf https://github.com/axodotdev/cargo-dist/releases/download/v0.18.0/cargo-dist-installer.sh | sh"
+        run: "curl --proto '=https' --tlsv1.2 -LsSf https://github.com/axodotdev/cargo-dist/releases/download/v0.22.1/cargo-dist-installer.sh | sh"
       - name: Cache cargo-dist
         uses: actions/upload-artifact@v4
         with:

.github/workflows/sync_typeshed.yaml

@@ -37,13 +37,13 @@ jobs:
       - name: Sync typeshed
         id: sync
         run: |
-          rm -rf ruff/crates/red_knot_python_semantic/vendor/typeshed
-          mkdir ruff/crates/red_knot_python_semantic/vendor/typeshed
-          cp typeshed/README.md ruff/crates/red_knot_python_semantic/vendor/typeshed
-          cp typeshed/LICENSE ruff/crates/red_knot_python_semantic/vendor/typeshed
-          cp -r typeshed/stdlib ruff/crates/red_knot_python_semantic/vendor/typeshed/stdlib
-          rm -rf ruff/crates/red_knot_python_semantic/vendor/typeshed/stdlib/@tests
-          git -C typeshed rev-parse HEAD > ruff/crates/red_knot_python_semantic/vendor/typeshed/source_commit.txt
+          rm -rf ruff/crates/ruff_vendored/vendor/typeshed
+          mkdir ruff/crates/ruff_vendored/vendor/typeshed
+          cp typeshed/README.md ruff/crates/ruff_vendored/vendor/typeshed
+          cp typeshed/LICENSE ruff/crates/ruff_vendored/vendor/typeshed
+          cp -r typeshed/stdlib ruff/crates/ruff_vendored/vendor/typeshed/stdlib
+          rm -rf ruff/crates/ruff_vendored/vendor/typeshed/stdlib/@tests
+          git -C typeshed rev-parse HEAD > ruff/crates/ruff_vendored/vendor/typeshed/source_commit.txt
       - name: Commit the changes
         id: commit
         if: ${{ steps.sync.outcome == 'success' }}

.markdownlint.yaml

@@ -14,6 +14,9 @@ MD041: false
 # MD013/line-length
 MD013: false
 
+# MD014/commands-show-output
+MD014: false
+
 # MD024/no-duplicate-heading
 MD024:
   # Allow when nested under different parents e.g. CHANGELOG.md

.pre-commit-config.yaml

@@ -2,10 +2,12 @@ fail_fast: true
 
 exclude: |
   (?x)^(
-    crates/red_knot_python_semantic/vendor/.*|
+    crates/ruff_vendored/vendor/.*|
     crates/red_knot_workspace/resources/.*|
     crates/ruff_linter/resources/.*|
     crates/ruff_linter/src/rules/.*/snapshots/.*|
+    crates/ruff_notebook/resources/.*|
+    crates/ruff_server/resources/.*|
     crates/ruff/resources/.*|
     crates/ruff_python_formatter/resources/.*|
     crates/ruff_python_formatter/tests/snapshots/.*|
@@ -15,7 +17,7 @@ exclude: |
 
 repos:
   - repo: https://github.com/abravalheri/validate-pyproject
-    rev: v0.18
+    rev: v0.19
     hooks:
       - id: validate-pyproject
 
@@ -43,7 +45,7 @@ repos:
           )$
 
   - repo: https://github.com/crate-ci/typos
-    rev: v1.23.6
+    rev: v1.24.5
     hooks:
       - id: typos
 
@@ -57,7 +59,7 @@ repos:
         pass_filenames: false # This makes it a lot faster
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.5.6
+    rev: v0.6.5
     hooks:
       - id: ruff-format
       - id: ruff

BREAKING_CHANGES.md

@@ -1,5 +1,43 @@
 # Breaking Changes
 
+## 0.6.0
+
+- Detect imports in `src` layouts by default for `isort` rules ([#12848](https://github.com/astral-sh/ruff/pull/12848))
+
+- The pytest rules `PT001` and `PT023` now default to omitting the decorator parentheses when there are no arguments ([#12838](https://github.com/astral-sh/ruff/pull/12838)).
+
+- Lint and format Jupyter Notebook by default ([#12878](https://github.com/astral-sh/ruff/pull/12878)).
+
+    You can disable specific rules for notebooks using [`per-file-ignores`](https://docs.astral.sh/ruff/settings/#lint_per-file-ignores):
+
+    ```toml
+    [tool.ruff.lint.per-file-ignores]
+    "*.ipynb" = ["E501"] # disable line-too-long in notebooks
+    ```
+
+    If you'd prefer to either only lint or only format Jupyter Notebook files, you can use the
+    section-specific `exclude` option to do so. For example, the following would only lint Jupyter
+    Notebook files and not format them:
+
+    ```toml
+    [tool.ruff.format]
+    exclude = ["*.ipynb"]
+    ```
+
+    And, conversely, the following would only format Jupyter Notebook files and not lint them:
+
+    ```toml
+    [tool.ruff.lint]
+    exclude = ["*.ipynb"]
+    ```
+
+    You can completely disable Jupyter Notebook support by updating the [`extend-exclude`](https://docs.astral.sh/ruff/settings/#extend-exclude) setting:
+
+    ```toml
+    [tool.ruff]
+    extend-exclude = ["*.ipynb"]
+    ```
+
 ## 0.5.0
 
 - Follow the XDG specification to discover user-level configurations on macOS (same as on other Unix platforms)

CHANGELOG.md

@@ -1,5 +1,282 @@
 # Changelog
 
+## 0.6.7
+
+### Preview features
+
+- Add Python version support to ruff analyze CLI ([#13426](https://github.com/astral-sh/ruff/pull/13426))
+- Add `exclude` support to `ruff analyze` ([#13425](https://github.com/astral-sh/ruff/pull/13425))
+- Fix parentheses around return type annotations ([#13381](https://github.com/astral-sh/ruff/pull/13381))
+
+### Rule changes
+
+- \[`pycodestyle`\] Fix: Don't autofix if the first line ends in a question mark? (D400) ([#13399](https://github.com/astral-sh/ruff/pull/13399))
+
+### Bug fixes
+
+- Respect `lint.exclude` in ruff check `--add-noqa` ([#13427](https://github.com/astral-sh/ruff/pull/13427))
+
+### Performance
+
+- Avoid tracking module resolver files in Salsa ([#13437](https://github.com/astral-sh/ruff/pull/13437))
+- Use `forget` for module resolver database ([#13438](https://github.com/astral-sh/ruff/pull/13438))
+
+## 0.6.6
+
+### Preview features
+
+- \[`refurb`\] Skip `slice-to-remove-prefix-or-suffix` (`FURB188`) when non-trivial slice steps are present ([#13405](https://github.com/astral-sh/ruff/pull/13405))
+- Add a subcommand to generate dependency graphs ([#13402](https://github.com/astral-sh/ruff/pull/13402))
+
+### Formatter
+
+- Fix placement of inline parameter comments ([#13379](https://github.com/astral-sh/ruff/pull/13379))
+
+### Server
+
+- Fix off-by one error in the `LineIndex::offset` calculation ([#13407](https://github.com/astral-sh/ruff/pull/13407))
+
+### Bug fixes
+
+- \[`fastapi`\] Respect FastAPI aliases in route definitions ([#13394](https://github.com/astral-sh/ruff/pull/13394))
+- \[`pydocstyle`\] Respect word boundaries when detecting function signature in docs ([#13388](https://github.com/astral-sh/ruff/pull/13388))
+
+### Documentation
+
+- Add backlinks to rule overview linter ([#13368](https://github.com/astral-sh/ruff/pull/13368))
+- Fix documentation for editor vim plugin ALE ([#13348](https://github.com/astral-sh/ruff/pull/13348))
+- Fix rendering of `FURB188` docs ([#13406](https://github.com/astral-sh/ruff/pull/13406))
+
+## 0.6.5
+
+### Preview features
+
+- \[`pydoclint`\] Ignore `DOC201` when function name is "**new**" ([#13300](https://github.com/astral-sh/ruff/pull/13300))
+- \[`refurb`\] Implement `slice-to-remove-prefix-or-suffix` (`FURB188`) ([#13256](https://github.com/astral-sh/ruff/pull/13256))
+
+### Rule changes
+
+- \[`eradicate`\] Ignore script-comments with multiple end-tags (`ERA001`) ([#13283](https://github.com/astral-sh/ruff/pull/13283))
+- \[`pyflakes`\] Improve error message for `UndefinedName` when a builtin was added in a newer version than specified in Ruff config (`F821`) ([#13293](https://github.com/astral-sh/ruff/pull/13293))
+
+### Server
+
+- Add support for extensionless Python files for server ([#13326](https://github.com/astral-sh/ruff/pull/13326))
+- Fix configuration inheritance for configurations specified in the LSP settings ([#13285](https://github.com/astral-sh/ruff/pull/13285))
+
+### Bug fixes
+
+- \[`ruff`\] Handle unary operators in `decimal-from-float-literal` (`RUF032`) ([#13275](https://github.com/astral-sh/ruff/pull/13275))
+
+### CLI
+
+- Only include rules with diagnostics in SARIF metadata ([#13268](https://github.com/astral-sh/ruff/pull/13268))
+
+### Playground
+
+- Add "Copy as pyproject.toml/ruff.toml" and "Paste from TOML" ([#13328](https://github.com/astral-sh/ruff/pull/13328))
+- Fix errors not shown for restored snippet on page load ([#13262](https://github.com/astral-sh/ruff/pull/13262))
+
+## 0.6.4
+
+### Preview features
+
+- \[`flake8-builtins`\] Use dynamic builtins list based on Python version ([#13172](https://github.com/astral-sh/ruff/pull/13172))
+- \[`pydoclint`\] Permit yielding `None` in `DOC402` and `DOC403` ([#13148](https://github.com/astral-sh/ruff/pull/13148))
+- \[`pylint`\] Update diagnostic message for `PLW3201` ([#13194](https://github.com/astral-sh/ruff/pull/13194))
+- \[`ruff`\] Implement `post-init-default` (`RUF033`) ([#13192](https://github.com/astral-sh/ruff/pull/13192))
+- \[`ruff`\] Implement useless if-else (`RUF034`) ([#13218](https://github.com/astral-sh/ruff/pull/13218))
+
+### Rule changes
+
+- \[`flake8-pyi`\] Respect `pep8_naming.classmethod-decorators` settings when determining if a method is a classmethod in `custom-type-var-return-type` (`PYI019`) ([#13162](https://github.com/astral-sh/ruff/pull/13162))
+- \[`flake8-pyi`\] Teach various rules that annotations might be stringized ([#12951](https://github.com/astral-sh/ruff/pull/12951))
+- \[`pylint`\] Avoid `no-self-use` for `attrs`-style validators ([#13166](https://github.com/astral-sh/ruff/pull/13166))
+- \[`pylint`\] Recurse into subscript subexpressions when searching for list/dict lookups (`PLR1733`, `PLR1736`) ([#13186](https://github.com/astral-sh/ruff/pull/13186))
+- \[`pyupgrade`\] Detect `aiofiles.open` calls in `UP015` ([#13173](https://github.com/astral-sh/ruff/pull/13173))
+- \[`pyupgrade`\] Mark `sys.version_info[0] < 3` and similar comparisons as outdated (`UP036`) ([#13175](https://github.com/astral-sh/ruff/pull/13175))
+
+### CLI
+
+- Enrich messages of SARIF results ([#13180](https://github.com/astral-sh/ruff/pull/13180))
+- Handle singular case for incompatible rules warning in `ruff format` output ([#13212](https://github.com/astral-sh/ruff/pull/13212))
+
+### Bug fixes
+
+- \[`pydocstyle`\] Improve heuristics for detecting Google-style docstrings ([#13142](https://github.com/astral-sh/ruff/pull/13142))
+- \[`refurb`\] Treat `sep` arguments with effects as unsafe removals (`FURB105`) ([#13165](https://github.com/astral-sh/ruff/pull/13165))
+
+## 0.6.3
+
+### Preview features
+
+- \[`flake8-simplify`\] Extend `open-file-with-context-handler` to work with `dbm.sqlite3` (`SIM115`) ([#13104](https://github.com/astral-sh/ruff/pull/13104))
+- \[`pycodestyle`\] Disable `E741` in stub files (`.pyi`) ([#13119](https://github.com/astral-sh/ruff/pull/13119))
+- \[`pydoclint`\] Avoid `DOC201` on explicit returns in functions that only return `None` ([#13064](https://github.com/astral-sh/ruff/pull/13064))
+
+### Rule changes
+
+- \[`flake8-async`\] Disable check for `asyncio` before Python 3.11 (`ASYNC109`) ([#13023](https://github.com/astral-sh/ruff/pull/13023))
+
+### Bug fixes
+
+- \[`FastAPI`\] Avoid introducing invalid syntax in fix for `fast-api-non-annotated-dependency` (`FAST002`) ([#13133](https://github.com/astral-sh/ruff/pull/13133))
+- \[`flake8-implicit-str-concat`\] Normalize octals before merging concatenated strings in `single-line-implicit-string-concatenation` (`ISC001`) ([#13118](https://github.com/astral-sh/ruff/pull/13118))
+- \[`flake8-pytest-style`\] Improve help message for `pytest-incorrect-mark-parentheses-style` (`PT023`) ([#13092](https://github.com/astral-sh/ruff/pull/13092))
+- \[`pylint`\] Avoid autofix for calls that aren't `min` or `max` as starred expression (`PLW3301`) ([#13089](https://github.com/astral-sh/ruff/pull/13089))
+- \[`ruff`\] Add `datetime.time`, `datetime.tzinfo`, and `datetime.timezone` as immutable function calls (`RUF009`) ([#13109](https://github.com/astral-sh/ruff/pull/13109))
+- \[`ruff`\] Extend comment deletion for `RUF100` to include trailing text from `noqa` directives while preserving any following comments on the same line, if any ([#13105](https://github.com/astral-sh/ruff/pull/13105))
+- Fix dark theme on initial page load for the Ruff playground ([#13077](https://github.com/astral-sh/ruff/pull/13077))
+
+## 0.6.2
+
+### Preview features
+
+- \[`flake8-simplify`\] Extend `open-file-with-context-handler` to work with other standard-library IO modules (`SIM115`) ([#12959](https://github.com/astral-sh/ruff/pull/12959))
+- \[`ruff`\] Avoid `unused-async` for functions with FastAPI route decorator (`RUF029`) ([#12938](https://github.com/astral-sh/ruff/pull/12938))
+- \[`ruff`\] Ignore `fstring-missing-syntax` (`RUF027`) for `fastAPI` paths ([#12939](https://github.com/astral-sh/ruff/pull/12939))
+- \[`ruff`\] Implement check for Decimal called with a float literal (RUF032) ([#12909](https://github.com/astral-sh/ruff/pull/12909))
+
+### Rule changes
+
+- \[`flake8-bugbear`\] Update diagnostic message when expression is at the end of function (`B015`) ([#12944](https://github.com/astral-sh/ruff/pull/12944))
+- \[`flake8-pyi`\] Skip type annotations in `string-or-bytes-too-long` (`PYI053`) ([#13002](https://github.com/astral-sh/ruff/pull/13002))
+- \[`flake8-type-checking`\] Always recognise relative imports as first-party ([#12994](https://github.com/astral-sh/ruff/pull/12994))
+- \[`flake8-unused-arguments`\] Ignore unused arguments on stub functions (`ARG001`) ([#12966](https://github.com/astral-sh/ruff/pull/12966))
+- \[`pylint`\] Ignore augmented assignment for `self-cls-assignment` (`PLW0642`) ([#12957](https://github.com/astral-sh/ruff/pull/12957))
+
+### Server
+
+- Show full context in error log messages ([#13029](https://github.com/astral-sh/ruff/pull/13029))
+
+### Bug fixes
+
+- \[`pep8-naming`\] Don't flag `from` imports following conventional import names (`N817`) ([#12946](https://github.com/astral-sh/ruff/pull/12946))
+- \[`pylint`\] - Allow `__new__` methods to have `cls` as their first argument even if decorated with `@staticmethod` for `bad-staticmethod-argument` (`PLW0211`) ([#12958](https://github.com/astral-sh/ruff/pull/12958))
+
+### Documentation
+
+- Add `hyperfine` installation instructions; update `hyperfine` code samples ([#13034](https://github.com/astral-sh/ruff/pull/13034))
+- Expand note to use Ruff with other language server in Kate ([#12806](https://github.com/astral-sh/ruff/pull/12806))
+- Update example for `PT001` as per the new default behavior ([#13019](https://github.com/astral-sh/ruff/pull/13019))
+- \[`perflint`\] Improve docs for `try-except-in-loop` (`PERF203`) ([#12947](https://github.com/astral-sh/ruff/pull/12947))
+- \[`pydocstyle`\] Add reference to `lint.pydocstyle.ignore-decorators` setting to rule docs ([#12996](https://github.com/astral-sh/ruff/pull/12996))
+
+## 0.6.1
+
+This is a hotfix release to address an issue with `ruff-pre-commit`. In v0.6,
+Ruff changed its behavior to lint and format Jupyter notebooks by default;
+however, due to an oversight, these files were still excluded by default if
+Ruff was run via pre-commit, leading to inconsistent behavior.
+This has [now been fixed](https://github.com/astral-sh/ruff-pre-commit/pull/96).
+
+### Preview features
+
+- \[`fastapi`\] Implement `fast-api-unused-path-parameter` (`FAST003`) ([#12638](https://github.com/astral-sh/ruff/pull/12638))
+
+### Rule changes
+
+- \[`pylint`\] Rename `too-many-positional` to `too-many-positional-arguments` (`R0917`) ([#12905](https://github.com/astral-sh/ruff/pull/12905))
+
+### Server
+
+- Fix crash when applying "fix-all" code-action to notebook cells ([#12929](https://github.com/astral-sh/ruff/pull/12929))
+
+### Other changes
+
+- \[`flake8-naming`\]: Respect import conventions (`N817`) ([#12922](https://github.com/astral-sh/ruff/pull/12922))
+
+## 0.6.0
+
+Check out the [blog post](https://astral.sh/blog/ruff-v0.6.0) for a migration guide and overview of the changes!
+
+### Breaking changes
+
+See also, the "Remapped rules" section which may result in disabled rules.
+
+- Lint and format Jupyter Notebook by default ([#12878](https://github.com/astral-sh/ruff/pull/12878)).
+- Detect imports in `src` layouts by default for `isort` rules ([#12848](https://github.com/astral-sh/ruff/pull/12848))
+- The pytest rules `PT001` and `PT023` now default to omitting the decorator parentheses when there are no arguments ([#12838](https://github.com/astral-sh/ruff/pull/12838)).
+
+### Deprecations
+
+The following rules are now deprecated:
+
+- [`pytest-missing-fixture-name-underscore`](https://docs.astral.sh/ruff/rules/pytest-missing-fixture-name-underscore/) (`PT004`)
+- [`pytest-incorrect-fixture-name-underscore`](https://docs.astral.sh/ruff/rules/pytest-incorrect-fixture-name-underscore/) (`PT005`)
+- [`unpacked-list-comprehension`](https://docs.astral.sh/ruff/rules/unpacked-list-comprehension/) (`UP027`)
+
+### Remapped rules
+
+The following rules have been remapped to new rule codes:
+
+- [`unnecessary-dict-comprehension-for-iterable`](https://docs.astral.sh/ruff/rules/unnecessary-dict-comprehension-for-iterable/): `RUF025` to `C420`
+
+### Stabilization
+
+The following rules have been stabilized and are no longer in preview:
+
+- [`singledispatch-method`](https://docs.astral.sh/ruff/rules/singledispatch-method/) (`PLE1519`)
+- [`singledispatchmethod-function`](https://docs.astral.sh/ruff/rules/singledispatchmethod-function/) (`PLE1520`)
+- [`bad-staticmethod-argument`](https://docs.astral.sh/ruff/rules/bad-staticmethod-argument/) (`PLW0211`)
+- [`if-stmt-min-max`](https://docs.astral.sh/ruff/rules/if-stmt-min-max/) (`PLR1730`)
+- [`invalid-bytes-return-type`](https://docs.astral.sh/ruff/rules/invalid-bytes-return-type/) (`PLE0308`)
+- [`invalid-hash-return-type`](https://docs.astral.sh/ruff/rules/invalid-hash-return-type/) (`PLE0309`)
+- [`invalid-index-return-type`](https://docs.astral.sh/ruff/rules/invalid-index-return-type/) (`PLE0305`)
+- [`invalid-length-return-type`](https://docs.astral.sh/ruff/rules/invalid-length-return-type/) (`PLEE303`)
+- [`self-or-cls-assignment`](https://docs.astral.sh/ruff/rules/self-or-cls-assignment/) (`PLW0642`)
+- [`byte-string-usage`](https://docs.astral.sh/ruff/rules/byte-string-usage/) (`PYI057`)
+- [`duplicate-literal-member`](https://docs.astral.sh/ruff/rules/duplicate-literal-member/) (`PYI062`)
+- [`redirected-noqa`](https://docs.astral.sh/ruff/rules/redirected-noqa/) (`RUF101`)
+
+The following behaviors have been stabilized:
+
+- [`cancel-scope-no-checkpoint`](https://docs.astral.sh/ruff/rules/cancel-scope-no-checkpoint/) (`ASYNC100`): Support `asyncio` and `anyio` context mangers.
+- [`async-function-with-timeout`](https://docs.astral.sh/ruff/rules/async-function-with-timeout/) (`ASYNC109`): Support `asyncio` and `anyio` context mangers.
+- [`async-busy-wait`](https://docs.astral.sh/ruff/rules/async-busy-wait/) (`ASYNC110`): Support `asyncio` and `anyio` context mangers.
+- [`async-zero-sleep`](https://docs.astral.sh/ruff/rules/async-zero-sleep/) (`ASYNC115`): Support `anyio` context mangers.
+- [`long-sleep-not-forever`](https://docs.astral.sh/ruff/rules/long-sleep-not-forever/) (`ASYNC116`): Support `anyio` context mangers.
+
+The following fixes have been stabilized:
+
+- [`superfluous-else-return`](https://docs.astral.sh/ruff/rules/superfluous-else-return/) (`RET505`)
+- [`superfluous-else-raise`](https://docs.astral.sh/ruff/rules/superfluous-else-raise/) (`RET506`)
+- [`superfluous-else-continue`](https://docs.astral.sh/ruff/rules/superfluous-else-continue/) (`RET507`)
+- [`superfluous-else-break`](https://docs.astral.sh/ruff/rules/superfluous-else-break/) (`RET508`)
+
+### Preview features
+
+- \[`flake8-simplify`\] Further simplify to binary in preview for (`SIM108`) ([#12796](https://github.com/astral-sh/ruff/pull/12796))
+- \[`pyupgrade`\] Show violations without auto-fix (`UP031`) ([#11229](https://github.com/astral-sh/ruff/pull/11229))
+
+### Rule changes
+
+- \[`flake8-import-conventions`\] Add `xml.etree.ElementTree` to default conventions ([#12455](https://github.com/astral-sh/ruff/pull/12455))
+- \[`flake8-pytest-style`\] Add a space after comma in CSV output (`PT006`) ([#12853](https://github.com/astral-sh/ruff/pull/12853))
+
+### Server
+
+- Show a message for incorrect settings ([#12781](https://github.com/astral-sh/ruff/pull/12781))
+
+### Bug fixes
+
+- \[`flake8-async`\] Do not lint yield in context manager (`ASYNC100`) ([#12896](https://github.com/astral-sh/ruff/pull/12896))
+- \[`flake8-comprehensions`\] Do not lint `async for` comprehensions (`C419`) ([#12895](https://github.com/astral-sh/ruff/pull/12895))
+- \[`flake8-return`\] Only add return `None` at end of a function (`RET503`) ([#11074](https://github.com/astral-sh/ruff/pull/11074))
+- \[`flake8-type-checking`\] Avoid treating `dataclasses.KW_ONLY` as typing-only (`TCH003`) ([#12863](https://github.com/astral-sh/ruff/pull/12863))
+- \[`pep8-naming`\] Treat `type(Protocol)` et al as metaclass base (`N805`) ([#12770](https://github.com/astral-sh/ruff/pull/12770))
+- \[`pydoclint`\] Don't enforce returns and yields in abstract methods (`DOC201`, `DOC202`) ([#12771](https://github.com/astral-sh/ruff/pull/12771))
+- \[`ruff`\] Skip tuples with slice expressions in (`RUF031`) ([#12768](https://github.com/astral-sh/ruff/pull/12768))
+- \[`ruff`\] Ignore unparenthesized tuples in subscripts when the subscript is a type annotation or type alias (`RUF031`) ([#12762](https://github.com/astral-sh/ruff/pull/12762))
+- \[`ruff`\] Ignore template strings passed to logging and `builtins._()` calls (`RUF027`) ([#12889](https://github.com/astral-sh/ruff/pull/12889))
+- \[`ruff`\] Do not remove parens for tuples with starred expressions in Python \<=3.10 (`RUF031`) ([#12784](https://github.com/astral-sh/ruff/pull/12784))
+- Evaluate default parameter values for a function in that function's enclosing scope ([#12852](https://github.com/astral-sh/ruff/pull/12852))
+
+### Other changes
+
+- Respect VS Code cell metadata when detecting the language of Jupyter Notebook cells ([#12864](https://github.com/astral-sh/ruff/pull/12864))
+- Respect `kernelspec` notebook metadata when detecting the preferred language for a Jupyter Notebook ([#12875](https://github.com/astral-sh/ruff/pull/12875))
+
 ## 0.5.7
 
 ### Preview features

CONTRIBUTING.md

@@ -2,35 +2,6 @@
 
 Welcome! We're happy to have you here. Thank you in advance for your contribution to Ruff.
 
-- [The Basics](#the-basics)
-    - [Prerequisites](#prerequisites)
-    - [Development](#development)
-    - [Project Structure](#project-structure)
-    - [Example: Adding a new lint rule](#example-adding-a-new-lint-rule)
-        - [Rule naming convention](#rule-naming-convention)
-        - [Rule testing: fixtures and snapshots](#rule-testing-fixtures-and-snapshots)
-    - [Example: Adding a new configuration option](#example-adding-a-new-configuration-option)
-- [MkDocs](#mkdocs)
-- [Release Process](#release-process)
-    - [Creating a new release](#creating-a-new-release)
-- [Ecosystem CI](#ecosystem-ci)
-- [Benchmarking and Profiling](#benchmarking-and-profiling)
-    - [CPython Benchmark](#cpython-benchmark)
-    - [Microbenchmarks](#microbenchmarks)
-        - [Benchmark-driven Development](#benchmark-driven-development)
-        - [PR Summary](#pr-summary)
-        - [Tips](#tips)
-    - [Profiling Projects](#profiling-projects)
-        - [Linux](#linux)
-        - [Mac](#mac)
-- [`cargo dev`](#cargo-dev)
-- [Subsystems](#subsystems)
-    - [Compilation Pipeline](#compilation-pipeline)
-    - [Import Categorization](#import-categorization)
-        - [Project root](#project-root)
-        - [Package root](#package-root)
-        - [Import categorization](#import-categorization-1)
-
 ## The Basics
 
 Ruff welcomes contributions in the form of pull requests.
@@ -333,22 +304,34 @@ even patch releases may contain [non-backwards-compatible changes](https://semve
 ### Creating a new release
 
 1. Install `uv`: `curl -LsSf https://astral.sh/uv/install.sh | sh`
+
 1. Run `./scripts/release.sh`; this command will:
+
     - Generate a temporary virtual environment with `rooster`
     - Generate a changelog entry in `CHANGELOG.md`
     - Update versions in `pyproject.toml` and `Cargo.toml`
     - Update references to versions in the `README.md` and documentation
     - Display contributors for the release
+
 1. The changelog should then be editorialized for consistency
+
     - Often labels will be missing from pull requests they will need to be manually organized into the proper section
     - Changes should be edited to be user-facing descriptions, avoiding internal details
+
 1. Highlight any breaking changes in `BREAKING_CHANGES.md`
+
 1. Run `cargo check`. This should update the lock file with new versions.
+
 1. Create a pull request with the changelog and version updates
+
 1. Merge the PR
+
 1. Run the [release workflow](https://github.com/astral-sh/ruff/actions/workflows/release.yml) with:
+
     - The new version number (without starting `v`)
+
 1. The release workflow will do the following:
+
     1. Build all the assets. If this fails (even though we tested in step 4), we haven't tagged or
         uploaded anything, you can restart after pushing a fix. If you just need to rerun the build,
         make sure you're [re-running all the failed
@@ -359,14 +342,25 @@ even patch releases may contain [non-backwards-compatible changes](https://semve
     1. Attach artifacts to draft GitHub release
     1. Trigger downstream repositories. This can fail non-catastrophically, as we can run any
         downstream jobs manually if needed.
+
 1. Verify the GitHub release:
+
     1. The Changelog should match the content of `CHANGELOG.md`
-    1. Append the contributors from the `bump.sh` script
+    1. Append the contributors from the `scripts/release.sh` script
+
 1. If needed, [update the schemastore](https://github.com/astral-sh/ruff/blob/main/scripts/update_schemastore.py).
+
     1. One can determine if an update is needed when
         `git diff old-version-tag new-version-tag -- ruff.schema.json` returns a non-empty diff.
     1. Once run successfully, you should follow the link in the output to create a PR.
-1. If needed, update the `ruff-lsp` and `ruff-vscode` repositories.
+
+1. If needed, update the [`ruff-lsp`](https://github.com/astral-sh/ruff-lsp) and
+    [`ruff-vscode`](https://github.com/astral-sh/ruff-vscode) repositories and follow
+    the release instructions in those repositories. `ruff-lsp` should always be updated
+    before `ruff-vscode`.
+
+    This step is generally not required for a patch release, but should always be done
+    for a minor release.
 
 ## Ecosystem CI
 
@@ -389,7 +383,7 @@ We have several ways of benchmarking and profiling Ruff:
 - Microbenchmarks which run the linter or the formatter on individual files. These run on pull requests.
 - Profiling the linter on either the microbenchmarks or entire projects
 
-> \[!NOTE\]
+> **Note**
 > When running benchmarks, ensure that your CPU is otherwise idle (e.g., close any background
 > applications, like web browsers). You may also want to switch your CPU to a "performance"
 > mode, if it exists, especially when benchmarking short-lived processes.
@@ -403,12 +397,18 @@ which makes it a good target for benchmarking.
 git clone --branch 3.10 https://github.com/python/cpython.git crates/ruff_linter/resources/test/cpython
 ```
 
+Install `hyperfine`:
+
+```shell
+cargo install hyperfine
+```
+
 To benchmark the release build:
 
 ```shell
 cargo build --release && hyperfine --warmup 10 \
-  "./target/release/ruff ./crates/ruff_linter/resources/test/cpython/ --no-cache -e" \
-  "./target/release/ruff ./crates/ruff_linter/resources/test/cpython/ -e"
+  "./target/release/ruff check ./crates/ruff_linter/resources/test/cpython/ --no-cache -e" \
+  "./target/release/ruff check ./crates/ruff_linter/resources/test/cpython/ -e"
 
 Benchmark 1: ./target/release/ruff ./crates/ruff_linter/resources/test/cpython/ --no-cache
   Time (mean ± σ):     293.8 ms ±   3.2 ms    [User: 2384.6 ms, System: 90.3 ms]
@@ -427,7 +427,7 @@ To benchmark against the ecosystem's existing tools:
 
 ```shell
 hyperfine --ignore-failure --warmup 5 \
-  "./target/release/ruff ./crates/ruff_linter/resources/test/cpython/ --no-cache" \
+  "./target/release/ruff check ./crates/ruff_linter/resources/test/cpython/ --no-cache" \
   "pyflakes crates/ruff_linter/resources/test/cpython" \
   "autoflake --recursive --expand-star-imports --remove-all-unused-imports --remove-unused-variables --remove-duplicate-keys resources/test/cpython" \
   "pycodestyle crates/ruff_linter/resources/test/cpython" \
@@ -473,7 +473,7 @@ To benchmark a subset of rules, e.g. `LineTooLong` and `DocLineTooLong`:
 
 ```shell
 cargo build --release && hyperfine --warmup 10 \
-  "./target/release/ruff ./crates/ruff_linter/resources/test/cpython/ --no-cache -e --select W505,E501"
+  "./target/release/ruff check ./crates/ruff_linter/resources/test/cpython/ --no-cache -e --select W505,E501"
 ```
 
 You can run `poetry install` from `./scripts/benchmarks` to create a working environment for the
@@ -530,6 +530,8 @@ You can run the benchmarks with
 cargo benchmark
 ```
 
+`cargo benchmark` is an alias for `cargo bench -p ruff_benchmark --bench linter --bench formatter --`
+
 #### Benchmark-driven Development
 
 Ruff uses [Criterion.rs](https://bheisler.github.io/criterion.rs/book/) for benchmarks. You can use
@@ -568,7 +570,7 @@ cargo install critcmp
 
 #### Tips
 
-- Use `cargo bench -p ruff_benchmark <filter>` to only run specific benchmarks. For example: `cargo benchmark lexer`
+- Use `cargo bench -p ruff_benchmark <filter>` to only run specific benchmarks. For example: `cargo bench -p ruff_benchmark lexer`
     to only run the lexer benchmarks.
 - Use `cargo bench -p ruff_benchmark -- --quiet` for a more cleaned up output (without statistical relevance)
 - Use `cargo bench -p ruff_benchmark -- --quick` to get faster results (more prone to noise)
@@ -911,9 +913,5 @@ There are three ways in which an import can be categorized as "first-party":
     the `src` setting and, for each directory, check for the existence of a subdirectory `foo` or a
     file `foo.py`.
 
-By default, `src` is set to the project root. In the above example, we'd want to set
-`src = ["./src"]` to ensure that we locate `./my_project/src/foo` and thus categorize `import foo`
-as first-party in `baz.py`. In practice, for this limited example, setting `src = ["./src"]` is
-unnecessary, as all imports within `./my_project/src/foo` would be categorized as first-party via
-the same-package heuristic; but if your project contains multiple packages, you'll want to set `src`
-explicitly.
+By default, `src` is set to the project root, along with `"src"` subdirectory in the project root.
+This ensures that Ruff supports both flat and "src" layouts out of the box.

Cargo.toml

@@ -14,9 +14,10 @@ license = "MIT"
 [workspace.dependencies]
 ruff = { path = "crates/ruff" }
 ruff_cache = { path = "crates/ruff_cache" }
-ruff_db = { path = "crates/ruff_db" }
+ruff_db = { path = "crates/ruff_db", default-features = false }
 ruff_diagnostics = { path = "crates/ruff_diagnostics" }
 ruff_formatter = { path = "crates/ruff_formatter" }
+ruff_graph = { path = "crates/ruff_graph" }
 ruff_index = { path = "crates/ruff_index" }
 ruff_linter = { path = "crates/ruff_linter" }
 ruff_macros = { path = "crates/ruff_macros" }
@@ -33,15 +34,17 @@ ruff_python_trivia = { path = "crates/ruff_python_trivia" }
 ruff_server = { path = "crates/ruff_server" }
 ruff_source_file = { path = "crates/ruff_source_file" }
 ruff_text_size = { path = "crates/ruff_text_size" }
+ruff_vendored = { path = "crates/ruff_vendored" }
 ruff_workspace = { path = "crates/ruff_workspace" }
 
 red_knot_python_semantic = { path = "crates/red_knot_python_semantic" }
 red_knot_server = { path = "crates/red_knot_server" }
-red_knot_workspace = { path = "crates/red_knot_workspace" }
+red_knot_workspace = { path = "crates/red_knot_workspace", default-features = false }
 
 aho-corasick = { version = "1.1.3" }
 annotate-snippets = { version = "0.9.2", features = ["color"] }
 anyhow = { version = "1.0.80" }
+assert_fs = { version = "1.1.0" }
 argfile = { version = "0.2.0" }
 bincode = { version = "1.3.3" }
 bitflags = { version = "2.5.0" }
@@ -68,6 +71,7 @@ fern = { version = "0.6.1" }
 filetime = { version = "0.2.23" }
 glob = { version = "0.3.1" }
 globset = { version = "0.4.14" }
+globwalk = { version = "0.9.1" }
 hashbrown = "0.14.3"
 ignore = { version = "0.4.22" }
 imara-diff = { version = "0.1.5" }
@@ -102,13 +106,13 @@ pep440_rs = { version = "0.6.0", features = ["serde"] }
 pretty_assertions = "1.3.0"
 proc-macro2 = { version = "1.0.79" }
 pyproject-toml = { version = "0.9.0" }
-quick-junit = { version = "0.4.0" }
+quick-junit = { version = "0.5.0" }
 quote = { version = "1.0.23" }
 rand = { version = "0.8.5" }
 rayon = { version = "1.10.0" }
 regex = { version = "1.10.2" }
 rustc-hash = { version = "2.0.0" }
-salsa = { git = "https://github.com/MichaReiser/salsa.git", tag = "red-knot-0.0.1" }
+salsa = { git = "https://github.com/salsa-rs/salsa.git", rev = "f608ff8b24f07706492027199f51132244034f29" }
 schemars = { version = "0.8.16" }
 seahash = { version = "4.1.0" }
 serde = { version = "1.0.197", features = ["derive"] }
@@ -230,9 +234,9 @@ inherits = "release"
 # Config for 'cargo dist'
 [workspace.metadata.dist]
 # The preferred cargo-dist version to use in CI (Cargo.toml SemVer syntax)
-cargo-dist-version = "0.18.0"
+cargo-dist-version = "0.22.1"
 # CI backends to support
-ci = ["github"]
+ci = "github"
 # The installers to generate for each app
 installers = ["shell", "powershell"]
 # The archive format to use for windows builds (defaults .zip)
@@ -263,11 +267,11 @@ targets = [
 auto-includes = false
 # Whether cargo-dist should create a GitHub Release or use an existing draft
 create-release = true
-# Publish jobs to run in CI
+# Which actions to run on pull requests
 pr-run-mode = "skip"
 # Whether CI should trigger releases with dispatches instead of tag pushes
 dispatch-releases = true
-# The stage during which the GitHub Release should be created
+# Which phase cargo-dist should use to create the GitHub release
 github-release = "announce"
 # Whether CI should include auto-generated code to build local artifacts
 build-local-artifacts = false
@@ -275,9 +279,11 @@ build-local-artifacts = false
 local-artifacts-jobs = ["./build-binaries", "./build-docker"]
 # Publish jobs to run in CI
 publish-jobs = ["./publish-pypi", "./publish-wasm"]
-# Announcement jobs to run in CI
+# Post-announce jobs to run in CI
 post-announce-jobs = ["./notify-dependents", "./publish-docs", "./publish-playground"]
 # Custom permissions for GitHub Jobs
 github-custom-job-permissions = { "build-docker" = { packages = "write", contents = "read" }, "publish-wasm" = { contents = "read", id-token = "write", packages = "write" } }
 # Whether to install an updater program
 install-updater = false
+# Path that installers should place binaries in
+install-path = "CARGO_HOME"

README.md

@@ -110,7 +110,7 @@ For more, see the [documentation](https://docs.astral.sh/ruff/).
 1. [Who's Using Ruff?](#whos-using-ruff)
 1. [License](#license)
 
-## Getting Started
+## Getting Started<a id="getting-started"></a>
 
 For more, see the [documentation](https://docs.astral.sh/ruff/).
 
@@ -136,8 +136,8 @@ curl -LsSf https://astral.sh/ruff/install.sh | sh
 powershell -c "irm https://astral.sh/ruff/install.ps1 | iex"
 
 # For a specific version.
-curl -LsSf https://astral.sh/ruff/0.5.7/install.sh | sh
-powershell -c "irm https://astral.sh/ruff/0.5.7/install.ps1 | iex"
+curl -LsSf https://astral.sh/ruff/0.6.7/install.sh | sh
+powershell -c "irm https://astral.sh/ruff/0.6.7/install.ps1 | iex"
 ```
 
 You can also install Ruff via [Homebrew](https://formulae.brew.sh/formula/ruff), [Conda](https://anaconda.org/conda-forge/ruff),
@@ -170,7 +170,7 @@ Ruff can also be used as a [pre-commit](https://pre-commit.com/) hook via [`ruff
 ```yaml
 - repo: https://github.com/astral-sh/ruff-pre-commit
   # Ruff version.
-  rev: v0.5.7
+  rev: v0.6.7
   hooks:
     # Run the linter.
     - id: ruff
@@ -195,7 +195,7 @@ jobs:
       - uses: chartboost/ruff-action@v1
 ```
 
-### Configuration
+### Configuration<a id="configuration"></a>
 
 Ruff can be configured through a `pyproject.toml`, `ruff.toml`, or `.ruff.toml` file (see:
 [_Configuration_](https://docs.astral.sh/ruff/configuration/), or [_Settings_](https://docs.astral.sh/ruff/settings/)
@@ -291,7 +291,7 @@ features that may change prior to stabilization.
 See `ruff help` for more on Ruff's top-level commands, or `ruff help check` and `ruff help format`
 for more on the linting and formatting commands, respectively.
 
-## Rules
+## Rules<a id="rules"></a>
 
 <!-- Begin section: Rules -->
 
@@ -367,21 +367,21 @@ quality tools, including:
 
 For a complete enumeration of the supported rules, see [_Rules_](https://docs.astral.sh/ruff/rules/).
 
-## Contributing
+## Contributing<a id="contributing"></a>
 
 Contributions are welcome and highly appreciated. To get started, check out the
 [**contributing guidelines**](https://docs.astral.sh/ruff/contributing/).
 
 You can also join us on [**Discord**](https://discord.com/invite/astral-sh).
 
-## Support
+## Support<a id="support"></a>
 
 Having trouble? Check out the existing issues on [**GitHub**](https://github.com/astral-sh/ruff/issues),
 or feel free to [**open a new one**](https://github.com/astral-sh/ruff/issues/new).
 
 You can also ask for help on [**Discord**](https://discord.com/invite/astral-sh).
 
-## Acknowledgements
+## Acknowledgements<a id="acknowledgements"></a>
 
 Ruff's linter draws on both the APIs and implementation details of many other
 tools in the Python ecosystem, especially [Flake8](https://github.com/PyCQA/flake8), [Pyflakes](https://github.com/PyCQA/pyflakes),
@@ -405,7 +405,7 @@ Ruff is the beneficiary of a large number of [contributors](https://github.com/a
 
 Ruff is released under the MIT license.
 
-## Who's Using Ruff?
+## Who's Using Ruff?<a id="whos-using-ruff"></a>
 
 Ruff is used by a number of major open-source projects and companies, including:
 
@@ -524,7 +524,7 @@ If you're using Ruff, consider adding the Ruff badge to your project's `README.m
 <a href="https://github.com/astral-sh/ruff"><img src="https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json" alt="Ruff" style="max-width:100%;"></a>
 ```
 
-## License
+## License<a id="license"></a>
 
 This repository is licensed under the [MIT License](https://github.com/astral-sh/ruff/blob/main/LICENSE)
 

crates/red_knot/Cargo.toml

@@ -13,9 +13,8 @@ license.workspace = true
 
 [dependencies]
 red_knot_python_semantic = { workspace = true }
-red_knot_workspace = { workspace = true }
+red_knot_workspace = { workspace = true, features = ["zstd"] }
 red_knot_server = { workspace = true }
-
 ruff_db = { workspace = true, features = ["os", "cache"] }
 
 anyhow = { workspace = true }

crates/red_knot/docs/tracing.md

@@ -13,12 +13,17 @@ The CLI supports different verbosity levels.
 - `-vv` activates `debug!` and timestamps: This should be enough information to get to the bottom of bug reports. When you're processing many packages or files, you'll get pages and pages of output, but each line is link to a specific action or state change.
 - `-vvv` activates `trace!` (only in debug builds) and shows tracing-spans: At this level, you're logging everything. Most of this is wasted, it's really slow, we dump e.g. the entire resolution graph. Only useful to developers, and you almost certainly want to use `RED_KNOT_LOG` to filter it down to the area your investigating.
 
-## `RED_KNOT_LOG`
+## Better logging with `RED_KNOT_LOG` and `RAYON_NUM_THREADS`
 
 By default, the CLI shows messages from the `ruff` and `red_knot` crates. Tracing messages from other crates are not shown.
 The `RED_KNOT_LOG` environment variable allows you to customize which messages are shown by specifying one
 or more [filter directives](https://docs.rs/tracing-subscriber/latest/tracing_subscriber/filter/struct.EnvFilter.html#directives).
 
+The `RAYON_NUM_THREADS` environment variable, meanwhile, can be used to control the level of concurrency red-knot uses.
+By default, red-knot will attempt to parallelize its work so that multiple files are checked simultaneously,
+but this can result in a confused logging output where messages from different threads are intertwined.
+To switch off concurrency entirely and have more readable logs, use `RAYON_NUM_THREADS=1`.
+
 ### Examples
 
 #### Show all debug messages

crates/red_knot/src/logging.rs

@@ -5,8 +5,8 @@ use colored::Colorize;
 use std::fmt;
 use std::fs::File;
 use std::io::BufWriter;
-use tracing::log::LevelFilter;
 use tracing::{Event, Subscriber};
+use tracing_subscriber::filter::LevelFilter;
 use tracing_subscriber::fmt::format::Writer;
 use tracing_subscriber::fmt::{FmtContext, FormatEvent, FormatFields};
 use tracing_subscriber::registry::LookupSpan;
@@ -60,10 +60,10 @@ pub(crate) enum VerbosityLevel {
 impl VerbosityLevel {
     const fn level_filter(self) -> LevelFilter {
         match self {
-            VerbosityLevel::Default => LevelFilter::Warn,
-            VerbosityLevel::Verbose => LevelFilter::Info,
-            VerbosityLevel::ExtraVerbose => LevelFilter::Debug,
-            VerbosityLevel::Trace => LevelFilter::Trace,
+            VerbosityLevel::Default => LevelFilter::WARN,
+            VerbosityLevel::Verbose => LevelFilter::INFO,
+            VerbosityLevel::ExtraVerbose => LevelFilter::DEBUG,
+            VerbosityLevel::Trace => LevelFilter::TRACE,
         }
     }
 
@@ -88,7 +88,7 @@ pub(crate) fn setup_tracing(level: VerbosityLevel) -> anyhow::Result<TracingGuar
         match level {
             VerbosityLevel::Default => {
                 // Show warning traces
-                EnvFilter::default().add_directive(tracing::level_filters::LevelFilter::WARN.into())
+                EnvFilter::default().add_directive(LevelFilter::WARN.into())
             }
             level => {
                 let level_filter = level.level_filter();

crates/red_knot/src/main.rs

@@ -7,12 +7,12 @@ use colored::Colorize;
 use crossbeam::channel as crossbeam_channel;
 use salsa::plumbing::ZalsaDatabase;
 
-use red_knot_python_semantic::{ProgramSettings, SearchPathSettings};
+use red_knot_python_semantic::SitePackages;
 use red_knot_server::run_server;
 use red_knot_workspace::db::RootDatabase;
-use red_knot_workspace::site_packages::VirtualEnvironment;
 use red_knot_workspace::watch;
 use red_knot_workspace::watch::WorkspaceWatcher;
+use red_knot_workspace::workspace::settings::Configuration;
 use red_knot_workspace::workspace::WorkspaceMetadata;
 use ruff_db::system::{OsSystem, System, SystemPath, SystemPathBuf};
 use target_version::TargetVersion;
@@ -65,15 +65,14 @@ to resolve type information for the project's third-party dependencies.",
         value_name = "PATH",
         help = "Additional path to use as a module-resolution source (can be passed multiple times)"
     )]
-    extra_search_path: Vec<SystemPathBuf>,
+    extra_search_path: Option<Vec<SystemPathBuf>>,
 
     #[arg(
         long,
         help = "Python version to assume when resolving types",
-        default_value_t = TargetVersion::default(),
-        value_name="VERSION")
-    ]
-    target_version: TargetVersion,
+        value_name = "VERSION"
+    )]
+    target_version: Option<TargetVersion>,
 
     #[clap(flatten)]
     verbosity: Verbosity,
@@ -86,6 +85,36 @@ to resolve type information for the project's third-party dependencies.",
     watch: bool,
 }
 
+impl Args {
+    fn to_configuration(&self, cli_cwd: &SystemPath) -> Configuration {
+        let mut configuration = Configuration::default();
+
+        if let Some(target_version) = self.target_version {
+            configuration.target_version = Some(target_version.into());
+        }
+
+        if let Some(venv_path) = &self.venv_path {
+            configuration.search_paths.site_packages = Some(SitePackages::Derived {
+                venv_path: SystemPath::absolute(venv_path, cli_cwd),
+            });
+        }
+
+        if let Some(custom_typeshed_dir) = &self.custom_typeshed_dir {
+            configuration.search_paths.custom_typeshed =
+                Some(SystemPath::absolute(custom_typeshed_dir, cli_cwd));
+        }
+
+        if let Some(extra_search_paths) = &self.extra_search_path {
+            configuration.search_paths.extra_paths = extra_search_paths
+                .iter()
+                .map(|path| Some(SystemPath::absolute(path, cli_cwd)))
+                .collect();
+        }
+
+        configuration
+    }
+}
+
 #[derive(Debug, clap::Subcommand)]
 pub enum Command {
     /// Start the language server
@@ -115,22 +144,13 @@ pub fn main() -> ExitStatus {
 }
 
 fn run() -> anyhow::Result<ExitStatus> {
-    let Args {
-        command,
-        current_directory,
-        custom_typeshed_dir,
-        extra_search_path: extra_paths,
-        venv_path,
-        target_version,
-        verbosity,
-        watch,
-    } = Args::parse_from(std::env::args().collect::<Vec<_>>());
+    let args = Args::parse_from(std::env::args().collect::<Vec<_>>());
 
-    if matches!(command, Some(Command::Server)) {
+    if matches!(args.command, Some(Command::Server)) {
         return run_server().map(|()| ExitStatus::Success);
     }
 
-    let verbosity = verbosity.level();
+    let verbosity = args.verbosity.level();
     countme::enable(verbosity.is_trace());
     let _guard = setup_tracing(verbosity)?;
 
@@ -146,10 +166,12 @@ fn run() -> anyhow::Result<ExitStatus> {
             })?
     };
 
-    let cwd = current_directory
+    let cwd = args
+        .current_directory
+        .as_ref()
         .map(|cwd| {
             if cwd.as_std_path().is_dir() {
-                Ok(SystemPath::absolute(&cwd, &cli_base_path))
+                Ok(SystemPath::absolute(cwd, &cli_base_path))
             } else {
                 Err(anyhow!(
                     "Provided current-directory path '{cwd}' is not a directory."
@@ -160,33 +182,18 @@ fn run() -> anyhow::Result<ExitStatus> {
         .unwrap_or_else(|| cli_base_path.clone());
 
     let system = OsSystem::new(cwd.clone());
-    let workspace_metadata = WorkspaceMetadata::from_path(system.current_directory(), &system)?;
-
-    // TODO: Verify the remaining search path settings eagerly.
-    let site_packages = venv_path
-        .map(|path| {
-            VirtualEnvironment::new(path, &OsSystem::new(cli_base_path))
-                .and_then(|venv| venv.site_packages_directories(&system))
-        })
-        .transpose()?
-        .unwrap_or_default();
-
-    // TODO: Respect the settings from the workspace metadata. when resolving the program settings.
-    let program_settings = ProgramSettings {
-        target_version: target_version.into(),
-        search_paths: SearchPathSettings {
-            extra_paths,
-            src_root: workspace_metadata.root().to_path_buf(),
-            custom_typeshed: custom_typeshed_dir,
-            site_packages,
-        },
-    };
+    let cli_configuration = args.to_configuration(&cwd);
+    let workspace_metadata = WorkspaceMetadata::from_path(
+        system.current_directory(),
+        &system,
+        Some(cli_configuration.clone()),
+    )?;
 
     // TODO: Use the `program_settings` to compute the key for the database's persistent
     //   cache and load the cache if it exists.
-    let mut db = RootDatabase::new(workspace_metadata, program_settings, system);
+    let mut db = RootDatabase::new(workspace_metadata, system)?;
 
-    let (main_loop, main_loop_cancellation_token) = MainLoop::new();
+    let (main_loop, main_loop_cancellation_token) = MainLoop::new(cli_configuration);
 
     // Listen to Ctrl+C and abort the watch mode.
     let main_loop_cancellation_token = Mutex::new(Some(main_loop_cancellation_token));
@@ -198,7 +205,7 @@ fn run() -> anyhow::Result<ExitStatus> {
         }
     })?;
 
-    let exit_status = if watch {
+    let exit_status = if args.watch {
         main_loop.watch(&mut db)?
     } else {
         main_loop.run(&mut db)
@@ -238,10 +245,12 @@ struct MainLoop {
 
     /// The file system watcher, if running in watch mode.
     watcher: Option<WorkspaceWatcher>,
+
+    cli_configuration: Configuration,
 }
 
 impl MainLoop {
-    fn new() -> (Self, MainLoopCancellationToken) {
+    fn new(cli_configuration: Configuration) -> (Self, MainLoopCancellationToken) {
         let (sender, receiver) = crossbeam_channel::bounded(10);
 
         (
@@ -249,6 +258,7 @@ impl MainLoop {
                 sender: sender.clone(),
                 receiver,
                 watcher: None,
+                cli_configuration,
             },
             MainLoopCancellationToken { sender },
         )
@@ -331,7 +341,7 @@ impl MainLoop {
                 MainLoopMessage::ApplyChanges(changes) => {
                     revision += 1;
                     // Automatically cancels any pending queries and waits for them to complete.
-                    db.apply_changes(changes);
+                    db.apply_changes(changes, Some(&self.cli_configuration));
                     if let Some(watcher) = self.watcher.as_mut() {
                         watcher.update(db);
                     }

crates/red_knot/tests/file_watching.rs

@@ -4,14 +4,12 @@ use std::io::Write;
 use std::time::Duration;
 
 use anyhow::{anyhow, Context};
-use salsa::Setter;
 
-use red_knot_python_semantic::{
-    resolve_module, ModuleName, Program, ProgramSettings, PythonVersion, SearchPathSettings,
-};
+use red_knot_python_semantic::{resolve_module, ModuleName, Program, PythonVersion, SitePackages};
 use red_knot_workspace::db::RootDatabase;
 use red_knot_workspace::watch;
 use red_knot_workspace::watch::{directory_watcher, WorkspaceWatcher};
+use red_knot_workspace::workspace::settings::{Configuration, SearchPathConfiguration};
 use red_knot_workspace::workspace::WorkspaceMetadata;
 use ruff_db::files::{system_path_to_file, File, FileError};
 use ruff_db::source::source_text;
@@ -26,6 +24,7 @@ struct TestCase {
     /// We need to hold on to it in the test case or the temp files get deleted.
     _temp_dir: tempfile::TempDir,
     root_dir: SystemPathBuf,
+    configuration: Configuration,
 }
 
 impl TestCase {
@@ -41,10 +40,6 @@ impl TestCase {
         &self.db
     }
 
-    fn db_mut(&mut self) -> &mut RootDatabase {
-        &mut self.db
-    }
-
     fn stop_watch(&mut self) -> Vec<watch::ChangeEvent> {
         self.try_stop_watch(Duration::from_secs(10))
             .expect("Expected watch changes but observed none.")
@@ -105,27 +100,32 @@ impl TestCase {
         Some(all_events)
     }
 
+    fn apply_changes(&mut self, changes: Vec<watch::ChangeEvent>) {
+        self.db.apply_changes(changes, Some(&self.configuration));
+    }
+
     fn update_search_path_settings(
         &mut self,
-        f: impl FnOnce(&SearchPathSettings) -> SearchPathSettings,
-    ) {
+        configuration: SearchPathConfiguration,
+    ) -> anyhow::Result<()> {
         let program = Program::get(self.db());
-        let search_path_settings = program.search_paths(self.db());
 
-        let new_settings = f(search_path_settings);
+        self.configuration.search_paths = configuration.clone();
+        let new_settings = configuration.into_settings(self.db.workspace().root(&self.db));
 
-        program.set_search_paths(&mut self.db).to(new_settings);
+        program.update_search_paths(&mut self.db, &new_settings)?;
 
         if let Some(watcher) = &mut self.watcher {
             watcher.update(&self.db);
             assert!(!watcher.has_errored_paths());
         }
+
+        Ok(())
     }
 
     fn collect_package_files(&self, path: &SystemPath) -> Vec<File> {
         let package = self.db().workspace().package(self.db(), path).unwrap();
         let files = package.files(self.db());
-        let files = files.read();
         let mut collected: Vec<_> = files.into_iter().collect();
         collected.sort_unstable_by_key(|file| file.path(self.db()).as_system_path().unwrap());
         collected
@@ -178,17 +178,14 @@ fn setup<F>(setup_files: F) -> anyhow::Result<TestCase>
 where
     F: SetupFiles,
 {
-    setup_with_search_paths(setup_files, |_root, workspace_path| SearchPathSettings {
-        extra_paths: vec![],
-        src_root: workspace_path.to_path_buf(),
-        custom_typeshed: None,
-        site_packages: vec![],
+    setup_with_search_paths(setup_files, |_root, _workspace_path| {
+        SearchPathConfiguration::default()
     })
 }
 
 fn setup_with_search_paths<F>(
     setup_files: F,
-    create_search_paths: impl FnOnce(&SystemPath, &SystemPath) -> SearchPathSettings,
+    create_search_paths: impl FnOnce(&SystemPath, &SystemPath) -> SearchPathConfiguration,
 ) -> anyhow::Result<TestCase>
 where
     F: SetupFiles,
@@ -220,25 +217,34 @@ where
 
     let system = OsSystem::new(&workspace_path);
 
-    let workspace = WorkspaceMetadata::from_path(&workspace_path, &system)?;
-    let search_paths = create_search_paths(&root_path, workspace.root());
+    let search_paths = create_search_paths(&root_path, &workspace_path);
 
     for path in search_paths
         .extra_paths
         .iter()
-        .chain(search_paths.site_packages.iter())
+        .flatten()
         .chain(search_paths.custom_typeshed.iter())
+        .chain(search_paths.site_packages.iter().flat_map(|site_packages| {
+            if let SitePackages::Known(path) = site_packages {
+                path.as_slice()
+            } else {
+                &[]
+            }
+        }))
     {
         std::fs::create_dir_all(path.as_std_path())
             .with_context(|| format!("Failed to create search path '{path}'"))?;
     }
 
-    let settings = ProgramSettings {
-        target_version: PythonVersion::default(),
+    let configuration = Configuration {
+        target_version: Some(PythonVersion::PY312),
         search_paths,
     };
 
-    let db = RootDatabase::new(workspace, settings, system);
+    let workspace =
+        WorkspaceMetadata::from_path(&workspace_path, &system, Some(configuration.clone()))?;
+
+    let db = RootDatabase::new(workspace, system)?;
 
     let (sender, receiver) = crossbeam::channel::unbounded();
     let watcher = directory_watcher(move |events| sender.send(events).unwrap())
@@ -253,6 +259,7 @@ where
         watcher: Some(watcher),
         _temp_dir: temp_dir,
         root_dir: root_path,
+        configuration,
     };
 
     // Sometimes the file watcher reports changes for events that happened before the watcher was started.
@@ -305,7 +312,7 @@ fn new_file() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     let foo = case.system_file(&foo_path).expect("foo.py to exist.");
 
@@ -328,7 +335,7 @@ fn new_ignored_file() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(case.system_file(&foo_path).is_ok());
     assert_eq!(&case.collect_package_files(&bar_path), &[bar_file]);
@@ -352,7 +359,7 @@ fn changed_file() -> anyhow::Result<()> {
 
     assert!(!changes.is_empty());
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert_eq!(source_text(case.db(), foo).as_str(), "print('Version 2')");
     assert_eq!(&case.collect_package_files(&foo_path), &[foo]);
@@ -375,7 +382,7 @@ fn deleted_file() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(!foo.exists(case.db()));
     assert_eq!(&case.collect_package_files(&foo_path), &[] as &[File]);
@@ -407,7 +414,7 @@ fn move_file_to_trash() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(!foo.exists(case.db()));
     assert_eq!(&case.collect_package_files(&foo_path), &[] as &[File]);
@@ -439,7 +446,7 @@ fn move_file_to_workspace() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     let foo_in_workspace = case.system_file(&foo_in_workspace_path)?;
 
@@ -467,7 +474,7 @@ fn rename_file() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(!foo.exists(case.db()));
 
@@ -508,7 +515,7 @@ fn directory_moved_to_workspace() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     let init_file = case
         .system_file(sub_new_path.join("__init__.py"))
@@ -559,7 +566,7 @@ fn directory_moved_to_trash() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     // `import sub.a` should no longer resolve
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("sub.a").unwrap()).is_none());
@@ -613,7 +620,7 @@ fn directory_renamed() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     // `import sub.a` should no longer resolve
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("sub.a").unwrap()).is_none());
@@ -678,7 +685,7 @@ fn directory_deleted() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     // `import sub.a` should no longer resolve
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("sub.a").unwrap()).is_none());
@@ -692,15 +699,13 @@ fn directory_deleted() -> anyhow::Result<()> {
 
 #[test]
 fn search_path() -> anyhow::Result<()> {
-    let mut case =
-        setup_with_search_paths([("bar.py", "import sub.a")], |root_path, workspace_path| {
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: workspace_path.to_path_buf(),
-                custom_typeshed: None,
-                site_packages: vec![root_path.join("site_packages")],
-            }
-        })?;
+    let mut case = setup_with_search_paths(
+        [("bar.py", "import sub.a")],
+        |root_path, _workspace_path| SearchPathConfiguration {
+            site_packages: Some(SitePackages::Known(vec![root_path.join("site_packages")])),
+            ..SearchPathConfiguration::default()
+        },
+    )?;
 
     let site_packages = case.root_path().join("site_packages");
 
@@ -713,7 +718,7 @@ fn search_path() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("a").unwrap()).is_some());
     assert_eq!(
@@ -734,16 +739,17 @@ fn add_search_path() -> anyhow::Result<()> {
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("a").unwrap()).is_none());
 
     // Register site-packages as a search path.
-    case.update_search_path_settings(|settings| SearchPathSettings {
-        site_packages: vec![site_packages.clone()],
-        ..settings.clone()
-    });
+    case.update_search_path_settings(SearchPathConfiguration {
+        site_packages: Some(SitePackages::Known(vec![site_packages.clone()])),
+        ..SearchPathConfiguration::default()
+    })
+    .expect("Search path settings to be valid");
 
     std::fs::write(site_packages.join("a.py").as_std_path(), "class A: ...")?;
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert!(resolve_module(case.db().upcast(), ModuleName::new_static("a").unwrap()).is_some());
 
@@ -752,22 +758,21 @@ fn add_search_path() -> anyhow::Result<()> {
 
 #[test]
 fn remove_search_path() -> anyhow::Result<()> {
-    let mut case =
-        setup_with_search_paths([("bar.py", "import sub.a")], |root_path, workspace_path| {
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: workspace_path.to_path_buf(),
-                custom_typeshed: None,
-                site_packages: vec![root_path.join("site_packages")],
-            }
-        })?;
+    let mut case = setup_with_search_paths(
+        [("bar.py", "import sub.a")],
+        |root_path, _workspace_path| SearchPathConfiguration {
+            site_packages: Some(SitePackages::Known(vec![root_path.join("site_packages")])),
+            ..SearchPathConfiguration::default()
+        },
+    )?;
 
     // Remove site packages from the search path settings.
     let site_packages = case.root_path().join("site_packages");
-    case.update_search_path_settings(|settings| SearchPathSettings {
-        site_packages: vec![],
-        ..settings.clone()
-    });
+    case.update_search_path_settings(SearchPathConfiguration {
+        site_packages: None,
+        ..SearchPathConfiguration::default()
+    })
+    .expect("Search path settings to be valid");
 
     std::fs::write(site_packages.join("a.py").as_std_path(), "class A: ...")?;
 
@@ -778,6 +783,48 @@ fn remove_search_path() -> anyhow::Result<()> {
     Ok(())
 }
 
+#[test]
+fn changed_versions_file() -> anyhow::Result<()> {
+    let mut case = setup_with_search_paths(
+        |root_path: &SystemPath, workspace_path: &SystemPath| {
+            std::fs::write(workspace_path.join("bar.py").as_std_path(), "import sub.a")?;
+            std::fs::create_dir_all(root_path.join("typeshed/stdlib").as_std_path())?;
+            std::fs::write(root_path.join("typeshed/stdlib/VERSIONS").as_std_path(), "")?;
+            std::fs::write(
+                root_path.join("typeshed/stdlib/os.pyi").as_std_path(),
+                "# not important",
+            )?;
+
+            Ok(())
+        },
+        |root_path, _workspace_path| SearchPathConfiguration {
+            custom_typeshed: Some(root_path.join("typeshed")),
+            ..SearchPathConfiguration::default()
+        },
+    )?;
+
+    // Unset the custom typeshed directory.
+    assert_eq!(
+        resolve_module(case.db(), ModuleName::new("os").unwrap()),
+        None
+    );
+
+    std::fs::write(
+        case.root_path()
+            .join("typeshed/stdlib/VERSIONS")
+            .as_std_path(),
+        "os: 3.0-",
+    )?;
+
+    let changes = case.stop_watch();
+
+    case.apply_changes(changes);
+
+    assert!(resolve_module(case.db(), ModuleName::new("os").unwrap()).is_some());
+
+    Ok(())
+}
+
 /// Watch a workspace that contains two files where one file is a hardlink to another.
 ///
 /// Setup:
@@ -824,7 +871,7 @@ fn hard_links_in_workspace() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert_eq!(source_text(case.db(), foo).as_str(), "print('Version 2')");
 
@@ -895,7 +942,7 @@ fn hard_links_to_target_outside_workspace() -> anyhow::Result<()> {
 
     let changes = case.stop_watch();
 
-    case.db_mut().apply_changes(changes);
+    case.apply_changes(changes);
 
     assert_eq!(source_text(case.db(), bar).as_str(), "print('Version 2')");
 
@@ -934,7 +981,7 @@ mod unix {
 
         let changes = case.stop_watch();
 
-        case.db_mut().apply_changes(changes);
+        case.apply_changes(changes);
 
         assert_eq!(
             foo.permissions(case.db()),
@@ -1019,7 +1066,7 @@ mod unix {
 
         let changes = case.take_watch_changes();
 
-        case.db_mut().apply_changes(changes);
+        case.apply_changes(changes);
 
         assert_eq!(
             source_text(case.db(), baz.file()).as_str(),
@@ -1032,7 +1079,7 @@ mod unix {
 
         let changes = case.stop_watch();
 
-        case.db_mut().apply_changes(changes);
+        case.apply_changes(changes);
 
         assert_eq!(
             source_text(case.db(), baz.file()).as_str(),
@@ -1103,7 +1150,7 @@ mod unix {
 
         let changes = case.stop_watch();
 
-        case.db_mut().apply_changes(changes);
+        case.apply_changes(changes);
 
         // The file watcher is guaranteed to emit one event for the changed file, but it isn't specified
         // if the event is emitted for the "original" or linked path because both paths are watched.
@@ -1172,11 +1219,11 @@ mod unix {
 
                 Ok(())
             },
-            |_root, workspace| SearchPathSettings {
-                extra_paths: vec![],
-                src_root: workspace.to_path_buf(),
-                custom_typeshed: None,
-                site_packages: vec![workspace.join(".venv/lib/python3.12/site-packages")],
+            |_root, workspace| SearchPathConfiguration {
+                site_packages: Some(SitePackages::Known(vec![
+                    workspace.join(".venv/lib/python3.12/site-packages")
+                ])),
+                ..SearchPathConfiguration::default()
             },
         )?;
 
@@ -1211,7 +1258,7 @@ mod unix {
 
         let changes = case.stop_watch();
 
-        case.db_mut().apply_changes(changes);
+        case.apply_changes(changes);
 
         assert_eq!(
             source_text(case.db(), baz_original_file).as_str(),

crates/red_knot_python_semantic/Cargo.toml

@@ -15,33 +15,33 @@ ruff_db = { workspace = true }
 ruff_index = { workspace = true }
 ruff_python_ast = { workspace = true }
 ruff_python_stdlib = { workspace = true }
+ruff_source_file = { workspace = true }
 ruff_text_size = { workspace = true }
+ruff_python_literal = { workspace = true }
 
+anyhow = { workspace = true }
 bitflags = { workspace = true }
 camino = { workspace = true }
 compact_str = { workspace = true }
 countme = { workspace = true }
-once_cell = { workspace = true }
 ordermap = { workspace = true }
 salsa = { workspace = true }
+thiserror = { workspace = true }
 tracing = { workspace = true }
 rustc-hash = { workspace = true }
 hashbrown = { workspace = true }
-
-[build-dependencies]
-path-slash = { workspace = true }
-walkdir = { workspace = true }
-zip = { workspace = true, features = ["zstd", "deflate"] }
+smallvec = { workspace = true }
+static_assertions = { workspace = true }
+test-case = { workspace = true }
 
 [dev-dependencies]
-ruff_db = { workspace = true, features = ["os", "testing"]}
+ruff_db = { workspace = true, features = ["os", "testing"] }
 ruff_python_parser = { workspace = true }
+ruff_vendored = { workspace = true }
 
 anyhow = { workspace = true }
 insta = { workspace = true }
 tempfile = { workspace = true }
-walkdir = { workspace = true }
-zip = { workspace = true }
 
 [lints]
 workspace = true

crates/red_knot_python_semantic/src/ast_node_ref.rs

@@ -31,10 +31,10 @@ impl<T> AstNodeRef<T> {
     /// which the `AstNodeRef` belongs.
     ///
     /// ## Safety
+    ///
     /// Dereferencing the `node` can result in undefined behavior if `parsed` isn't the
     /// [`ParsedModule`] to which `node` belongs. It's the caller's responsibility to ensure that
     /// the invariant `node belongs to parsed` is upheld.
-
     pub(super) unsafe fn new(parsed: ParsedModule, node: &T) -> Self {
         Self {
             _parsed: parsed,

crates/red_knot_python_semantic/src/builtins.rs

@@ -1,16 +0,0 @@
-use crate::module_name::ModuleName;
-use crate::module_resolver::resolve_module;
-use crate::semantic_index::global_scope;
-use crate::semantic_index::symbol::ScopeId;
-use crate::Db;
-
-/// Salsa query to get the builtins scope.
-///
-/// Can return None if a custom typeshed is used that is missing `builtins.pyi`.
-#[salsa::tracked]
-pub(crate) fn builtins_scope(db: &dyn Db) -> Option<ScopeId<'_>> {
-    let builtins_name =
-        ModuleName::new_static("builtins").expect("Expected 'builtins' to be a valid module name");
-    let builtins_file = resolve_module(db, builtins_name)?.file();
-    Some(global_scope(db, builtins_file))
-}

crates/red_knot_python_semantic/src/db.rs

@@ -1,15 +1,17 @@
+use ruff_db::files::File;
 use ruff_db::{Db as SourceDb, Upcast};
 
 /// Database giving access to semantic information about a Python program.
 #[salsa::db]
-pub trait Db: SourceDb + Upcast<dyn SourceDb> {}
+pub trait Db: SourceDb + Upcast<dyn SourceDb> {
+    fn is_file_open(&self, file: File) -> bool;
+}
 
 #[cfg(test)]
 pub(crate) mod tests {
     use std::sync::Arc;
 
-    use crate::module_resolver::vendored_typeshed_stubs;
-    use ruff_db::files::Files;
+    use ruff_db::files::{File, Files};
     use ruff_db::system::{DbWithTestSystem, System, TestSystem};
     use ruff_db::vendored::VendoredFileSystem;
     use ruff_db::{Db as SourceDb, Upcast};
@@ -30,7 +32,7 @@ pub(crate) mod tests {
             Self {
                 storage: salsa::Storage::default(),
                 system: TestSystem::default(),
-                vendored: vendored_typeshed_stubs().clone(),
+                vendored: ruff_vendored::file_system().clone(),
                 events: std::sync::Arc::default(),
                 files: Files::default(),
             }
@@ -91,7 +93,11 @@ pub(crate) mod tests {
     }
 
     #[salsa::db]
-    impl Db for TestDb {}
+    impl Db for TestDb {
+        fn is_file_open(&self, file: File) -> bool {
+            !file.path(self).is_vendored_path()
+        }
+    }
 
     #[salsa::db]
     impl salsa::Database for TestDb {

crates/red_knot_python_semantic/src/lib.rs

@@ -4,13 +4,12 @@ use rustc_hash::FxHasher;
 
 pub use db::Db;
 pub use module_name::ModuleName;
-pub use module_resolver::{resolve_module, system_module_search_paths, vendored_typeshed_stubs};
-pub use program::{Program, ProgramSettings, SearchPathSettings};
+pub use module_resolver::{resolve_module, system_module_search_paths, Module};
+pub use program::{Program, ProgramSettings, SearchPathSettings, SitePackages};
 pub use python_version::PythonVersion;
 pub use semantic_model::{HasTy, SemanticModel};
 
 pub mod ast_node_ref;
-mod builtins;
 mod db;
 mod module_name;
 mod module_resolver;
@@ -19,6 +18,8 @@ mod program;
 mod python_version;
 pub mod semantic_index;
 mod semantic_model;
+pub(crate) mod site_packages;
+mod stdlib;
 pub mod types;
 
 type FxOrderSet<V> = ordermap::set::OrderSet<V, BuildHasherDefault<FxHasher>>;

crates/red_knot_python_semantic/src/module_name.rs

@@ -168,6 +168,24 @@ impl ModuleName {
         };
         Some(Self(name))
     }
+
+    /// Extend `self` with the components of `other`
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use red_knot_python_semantic::ModuleName;
+    ///
+    /// let mut module_name = ModuleName::new_static("foo").unwrap();
+    /// module_name.extend(&ModuleName::new_static("bar").unwrap());
+    /// assert_eq!(&module_name, "foo.bar");
+    /// module_name.extend(&ModuleName::new_static("baz.eggs.ham").unwrap());
+    /// assert_eq!(&module_name, "foo.bar.baz.eggs.ham");
+    /// ```
+    pub fn extend(&mut self, other: &ModuleName) {
+        self.0.push('.');
+        self.0.push_str(other);
+    }
 }
 
 impl Deref for ModuleName {

crates/red_knot_python_semantic/src/module_resolver/mod.rs

@@ -1,17 +1,17 @@
 use std::iter::FusedIterator;
 
-pub(crate) use module::Module;
+pub use module::Module;
 pub use resolver::resolve_module;
+pub(crate) use resolver::{file_to_module, SearchPaths};
 use ruff_db::system::SystemPath;
-pub use typeshed::vendored_typeshed_stubs;
 
+use crate::module_resolver::resolver::search_paths;
 use crate::Db;
-use resolver::{module_resolution_settings, SearchPathIterator};
+use resolver::SearchPathIterator;
 
 mod module;
 mod path;
 mod resolver;
-mod state;
 mod typeshed;
 
 #[cfg(test)]
@@ -20,7 +20,7 @@ mod testing;
 /// Returns an iterator over all search paths pointing to a system path
 pub fn system_module_search_paths(db: &dyn Db) -> SystemModuleSearchPathsIter {
     SystemModuleSearchPathsIter {
-        inner: module_resolution_settings(db).search_paths(db),
+        inner: search_paths(db),
     }
 }
 

crates/red_knot_python_semantic/src/module_resolver/module.rs

@@ -77,3 +77,9 @@ pub enum ModuleKind {
     /// A python package (`foo/__init__.py` or `foo/__init__.pyi`)
     Package,
 }
+
+impl ModuleKind {
+    pub const fn is_package(self) -> bool {
+        matches!(self, ModuleKind::Package)
+    }
+}

crates/red_knot_python_semantic/src/module_resolver/path.rs

@@ -9,11 +9,11 @@ use ruff_db::files::{system_path_to_file, vendored_path_to_file, File, FileError
 use ruff_db::system::{System, SystemPath, SystemPathBuf};
 use ruff_db::vendored::{VendoredPath, VendoredPathBuf};
 
+use super::typeshed::{typeshed_versions, TypeshedVersionsParseError, TypeshedVersionsQueryResult};
 use crate::db::Db;
 use crate::module_name::ModuleName;
-
-use super::state::ResolverState;
-use super::typeshed::{TypeshedVersionsParseError, TypeshedVersionsQueryResult};
+use crate::module_resolver::resolver::ResolverContext;
+use crate::site_packages::SitePackagesDiscoveryError;
 
 /// A path that points to a Python module.
 ///
@@ -59,8 +59,12 @@ impl ModulePath {
         self.relative_path.push(component);
     }
 
+    pub(crate) fn pop(&mut self) -> bool {
+        self.relative_path.pop()
+    }
+
     #[must_use]
-    pub(crate) fn is_directory(&self, resolver: &ResolverState) -> bool {
+    pub(super) fn is_directory(&self, resolver: &ResolverContext) -> bool {
         let ModulePath {
             search_path,
             relative_path,
@@ -74,7 +78,7 @@ impl ModulePath {
                     == Err(FileError::IsADirectory)
             }
             SearchPathInner::StandardLibraryCustom(stdlib_root) => {
-                match query_stdlib_version(Some(stdlib_root), relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => false,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => {
@@ -84,7 +88,7 @@ impl ModulePath {
                 }
             }
             SearchPathInner::StandardLibraryVendored(stdlib_root) => {
-                match query_stdlib_version(None, relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => false,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => resolver
@@ -96,7 +100,7 @@ impl ModulePath {
     }
 
     #[must_use]
-    pub(crate) fn is_regular_package(&self, resolver: &ResolverState) -> bool {
+    pub(super) fn is_regular_package(&self, resolver: &ResolverContext) -> bool {
         let ModulePath {
             search_path,
             relative_path,
@@ -113,7 +117,7 @@ impl ModulePath {
                         .is_ok()
             }
             SearchPathInner::StandardLibraryCustom(search_path) => {
-                match query_stdlib_version(Some(search_path), relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => false,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => system_path_to_file(
@@ -124,7 +128,7 @@ impl ModulePath {
                 }
             }
             SearchPathInner::StandardLibraryVendored(search_path) => {
-                match query_stdlib_version(None, relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => false,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => resolver
@@ -136,7 +140,7 @@ impl ModulePath {
     }
 
     #[must_use]
-    pub(crate) fn to_file(&self, resolver: &ResolverState) -> Option<File> {
+    pub(super) fn to_file(&self, resolver: &ResolverContext) -> Option<File> {
         let db = resolver.db.upcast();
         let ModulePath {
             search_path,
@@ -150,7 +154,7 @@ impl ModulePath {
                 system_path_to_file(db, search_path.join(relative_path)).ok()
             }
             SearchPathInner::StandardLibraryCustom(stdlib_root) => {
-                match query_stdlib_version(Some(stdlib_root), relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => None,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => {
@@ -159,7 +163,7 @@ impl ModulePath {
                 }
             }
             SearchPathInner::StandardLibraryVendored(stdlib_root) => {
-                match query_stdlib_version(None, relative_path, resolver) {
+                match query_stdlib_version(relative_path, resolver) {
                     TypeshedVersionsQueryResult::DoesNotExist => None,
                     TypeshedVersionsQueryResult::Exists
                     | TypeshedVersionsQueryResult::MaybeExists => {
@@ -273,19 +277,15 @@ fn stdlib_path_to_module_name(relative_path: &Utf8Path) -> Option<ModuleName> {
 
 #[must_use]
 fn query_stdlib_version(
-    custom_stdlib_root: Option<&SystemPath>,
     relative_path: &Utf8Path,
-    resolver: &ResolverState,
+    context: &ResolverContext,
 ) -> TypeshedVersionsQueryResult {
     let Some(module_name) = stdlib_path_to_module_name(relative_path) else {
         return TypeshedVersionsQueryResult::DoesNotExist;
     };
-    let ResolverState {
-        db,
-        typeshed_versions,
-        target_version,
-    } = resolver;
-    typeshed_versions.query_module(*db, &module_name, custom_stdlib_root, *target_version)
+    let ResolverContext { db, target_version } = context;
+
+    typeshed_versions(*db).query_module(&module_name, *target_version)
 }
 
 /// Enumeration describing the various ways in which validation of a search path might fail.
@@ -293,7 +293,7 @@ fn query_stdlib_version(
 /// If validation fails for a search path derived from the user settings,
 /// a message must be displayed to the user,
 /// as type checking cannot be done reliably in these circumstances.
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug)]
 pub(crate) enum SearchPathValidationError {
     /// The path provided by the user was not a directory
     NotADirectory(SystemPathBuf),
@@ -304,18 +304,20 @@ pub(crate) enum SearchPathValidationError {
     NoStdlibSubdirectory(SystemPathBuf),
 
     /// The typeshed path provided by the user is a directory,
-    /// but no `stdlib/VERSIONS` file exists.
+    /// but `stdlib/VERSIONS` could not be read.
     /// (This is only relevant for stdlib search paths.)
-    NoVersionsFile(SystemPathBuf),
-
-    /// `stdlib/VERSIONS` is a directory.
-    /// (This is only relevant for stdlib search paths.)
-    VersionsIsADirectory(SystemPathBuf),
+    FailedToReadVersionsFile {
+        path: SystemPathBuf,
+        error: std::io::Error,
+    },
 
     /// The path provided by the user is a directory,
     /// and a `stdlib/VERSIONS` file exists, but it fails to parse.
     /// (This is only relevant for stdlib search paths.)
     VersionsParseError(TypeshedVersionsParseError),
+
+    /// Failed to discover the site-packages for the configured virtual environment.
+    SitePackagesDiscovery(SitePackagesDiscoveryError),
 }
 
 impl fmt::Display for SearchPathValidationError {
@@ -325,9 +327,16 @@ impl fmt::Display for SearchPathValidationError {
             Self::NoStdlibSubdirectory(path) => {
                 write!(f, "The directory at {path} has no `stdlib/` subdirectory")
             }
-            Self::NoVersionsFile(path) => write!(f, "Expected a file at {path}/stdlib/VERSIONS"),
-            Self::VersionsIsADirectory(path) => write!(f, "{path}/stdlib/VERSIONS is a directory."),
+            Self::FailedToReadVersionsFile { path, error } => {
+                write!(
+                    f,
+                    "Failed to read the custom typeshed versions file '{path}': {error}"
+                )
+            }
             Self::VersionsParseError(underlying_error) => underlying_error.fmt(f),
+            SearchPathValidationError::SitePackagesDiscovery(error) => {
+                write!(f, "Failed to discover the site-packages directory: {error}")
+            }
         }
     }
 }
@@ -342,6 +351,18 @@ impl std::error::Error for SearchPathValidationError {
     }
 }
 
+impl From<TypeshedVersionsParseError> for SearchPathValidationError {
+    fn from(value: TypeshedVersionsParseError) -> Self {
+        Self::VersionsParseError(value)
+    }
+}
+
+impl From<SitePackagesDiscoveryError> for SearchPathValidationError {
+    fn from(value: SitePackagesDiscoveryError) -> Self {
+        Self::SitePackagesDiscovery(value)
+    }
+}
+
 type SearchPathResult<T> = Result<T, SearchPathValidationError>;
 
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
@@ -384,11 +405,10 @@ pub(crate) struct SearchPath(Arc<SearchPathInner>);
 
 impl SearchPath {
     fn directory_path(system: &dyn System, root: SystemPathBuf) -> SearchPathResult<SystemPathBuf> {
-        let canonicalized = system.canonicalize_path(&root).unwrap_or(root);
-        if system.is_directory(&canonicalized) {
-            Ok(canonicalized)
+        if system.is_directory(&root) {
+            Ok(root)
         } else {
-            Err(SearchPathValidationError::NotADirectory(canonicalized))
+            Err(SearchPathValidationError::NotADirectory(root))
         }
     }
 
@@ -407,32 +427,22 @@ impl SearchPath {
     }
 
     /// Create a new standard-library search path pointing to a custom directory on disk
-    pub(crate) fn custom_stdlib(db: &dyn Db, typeshed: SystemPathBuf) -> SearchPathResult<Self> {
+    pub(crate) fn custom_stdlib(db: &dyn Db, typeshed: &SystemPath) -> SearchPathResult<Self> {
         let system = db.system();
-        if !system.is_directory(&typeshed) {
+        if !system.is_directory(typeshed) {
             return Err(SearchPathValidationError::NotADirectory(
                 typeshed.to_path_buf(),
             ));
         }
+
         let stdlib =
             Self::directory_path(system, typeshed.join("stdlib")).map_err(|err| match err {
-                SearchPathValidationError::NotADirectory(path) => {
-                    SearchPathValidationError::NoStdlibSubdirectory(path)
+                SearchPathValidationError::NotADirectory(_) => {
+                    SearchPathValidationError::NoStdlibSubdirectory(typeshed.to_path_buf())
                 }
                 err => err,
             })?;
-        let typeshed_versions =
-            system_path_to_file(db.upcast(), stdlib.join("VERSIONS")).map_err(|err| match err {
-                FileError::NotFound => SearchPathValidationError::NoVersionsFile(typeshed),
-                FileError::IsADirectory => {
-                    SearchPathValidationError::VersionsIsADirectory(typeshed)
-                }
-            })?;
-        super::typeshed::parse_typeshed_versions(db, typeshed_versions)
-            .as_ref()
-            .map_err(|validation_error| {
-                SearchPathValidationError::VersionsParseError(validation_error.clone())
-            })?;
+
         Ok(Self(Arc::new(SearchPathInner::StandardLibraryCustom(
             stdlib,
         ))))
@@ -623,11 +633,11 @@ mod tests {
     use ruff_db::Db;
 
     use crate::db::tests::TestDb;
-
-    use super::*;
     use crate::module_resolver::testing::{FileSpec, MockedTypeshed, TestCase, TestCaseBuilder};
     use crate::python_version::PythonVersion;
 
+    use super::*;
+
     impl ModulePath {
         #[must_use]
         fn join(&self, component: &str) -> ModulePath {
@@ -638,15 +648,6 @@ mod tests {
     }
 
     impl SearchPath {
-        #[must_use]
-        pub(crate) fn is_stdlib_search_path(&self) -> bool {
-            matches!(
-                &*self.0,
-                SearchPathInner::StandardLibraryCustom(_)
-                    | SearchPathInner::StandardLibraryVendored(_)
-            )
-        }
-
         fn join(&self, component: &str) -> ModulePath {
             self.to_module_path().join(component)
         }
@@ -661,7 +662,7 @@ mod tests {
             .build();
 
         assert_eq!(
-            SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf())
+            SearchPath::custom_stdlib(&db, stdlib.parent().unwrap())
                 .unwrap()
                 .to_module_path()
                 .with_py_extension(),
@@ -669,7 +670,7 @@ mod tests {
         );
 
         assert_eq!(
-            &SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf())
+            &SearchPath::custom_stdlib(&db, stdlib.parent().unwrap())
                 .unwrap()
                 .join("foo")
                 .with_pyi_extension(),
@@ -780,7 +781,7 @@ mod tests {
         let TestCase { db, stdlib, .. } = TestCaseBuilder::new()
             .with_custom_typeshed(MockedTypeshed::default())
             .build();
-        SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf())
+        SearchPath::custom_stdlib(&db, stdlib.parent().unwrap())
             .unwrap()
             .to_module_path()
             .push("bar.py");
@@ -792,7 +793,7 @@ mod tests {
         let TestCase { db, stdlib, .. } = TestCaseBuilder::new()
             .with_custom_typeshed(MockedTypeshed::default())
             .build();
-        SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf())
+        SearchPath::custom_stdlib(&db, stdlib.parent().unwrap())
             .unwrap()
             .to_module_path()
             .push("bar.rs");
@@ -824,7 +825,7 @@ mod tests {
             .with_custom_typeshed(MockedTypeshed::default())
             .build();
 
-        let root = SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf()).unwrap();
+        let root = SearchPath::custom_stdlib(&db, stdlib.parent().unwrap()).unwrap();
 
         // Must have a `.pyi` extension or no extension:
         let bad_absolute_path = SystemPath::new("foo/stdlib/x.py");
@@ -872,8 +873,7 @@ mod tests {
             .with_custom_typeshed(typeshed)
             .with_target_version(target_version)
             .build();
-        let stdlib =
-            SearchPath::custom_stdlib(&db, stdlib.parent().unwrap().to_path_buf()).unwrap();
+        let stdlib = SearchPath::custom_stdlib(&db, stdlib.parent().unwrap()).unwrap();
         (db, stdlib)
     }
 
@@ -898,7 +898,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let asyncio_regular_package = stdlib_path.join("asyncio");
         assert!(asyncio_regular_package.is_directory(&resolver));
@@ -926,7 +926,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let xml_namespace_package = stdlib_path.join("xml");
         assert!(xml_namespace_package.is_directory(&resolver));
@@ -948,7 +948,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let functools_module = stdlib_path.join("functools.pyi");
         assert!(functools_module.to_file(&resolver).is_some());
@@ -964,7 +964,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let collections_regular_package = stdlib_path.join("collections");
         assert_eq!(collections_regular_package.to_file(&resolver), None);
@@ -980,7 +980,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let importlib_namespace_package = stdlib_path.join("importlib");
         assert_eq!(importlib_namespace_package.to_file(&resolver), None);
@@ -1001,7 +1001,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py38_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY38);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY38);
 
         let non_existent = stdlib_path.join("doesnt_even_exist");
         assert_eq!(non_existent.to_file(&resolver), None);
@@ -1029,7 +1029,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py39_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY39);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY39);
 
         // Since we've set the target version to Py39,
         // `collections` should now exist as a directory, according to VERSIONS...
@@ -1058,7 +1058,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py39_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY39);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY39);
 
         // The `importlib` directory now also exists
         let importlib_namespace_package = stdlib_path.join("importlib");
@@ -1082,7 +1082,7 @@ mod tests {
         };
 
         let (db, stdlib_path) = py39_typeshed_test_case(TYPESHED);
-        let resolver = ResolverState::new(&db, PythonVersion::PY39);
+        let resolver = ResolverContext::new(&db, PythonVersion::PY39);
 
         // The `xml` package no longer exists on py39:
         let xml_namespace_package = stdlib_path.join("xml");

crates/red_knot_python_semantic/src/module_resolver/resolver.rs

@@ -4,15 +4,17 @@ use std::iter::FusedIterator;
 use rustc_hash::{FxBuildHasher, FxHashSet};
 
 use ruff_db::files::{File, FilePath, FileRootKind};
-use ruff_db::system::{DirectoryEntry, SystemPath, SystemPathBuf};
-use ruff_db::vendored::VendoredPath;
+use ruff_db::system::{DirectoryEntry, System, SystemPath, SystemPathBuf};
+use ruff_db::vendored::{VendoredFileSystem, VendoredPath};
+
+use crate::db::Db;
+use crate::module_name::ModuleName;
+use crate::module_resolver::typeshed::{vendored_typeshed_versions, TypeshedVersions};
+use crate::site_packages::VirtualEnvironment;
+use crate::{Program, PythonVersion, SearchPathSettings, SitePackages};
 
 use super::module::{Module, ModuleKind};
 use super::path::{ModulePath, SearchPath, SearchPathValidationError};
-use super::state::ResolverState;
-use crate::db::Db;
-use crate::module_name::ModuleName;
-use crate::{Program, PythonVersion, SearchPathSettings};
 
 /// Resolves a module name to a module.
 pub fn resolve_module(db: &dyn Db, module_name: ModuleName) -> Option<Module> {
@@ -40,7 +42,7 @@ pub(crate) fn resolve_module_query<'db>(
 
     let module = Module::new(name.clone(), kind, search_path, module_file);
 
-    tracing::debug!(
+    tracing::trace!(
         "Resolved module '{name}' to '{path}'.",
         path = module_file.path(db)
     );
@@ -84,9 +86,7 @@ pub(crate) fn file_to_module(db: &dyn Db, file: File) -> Option<Module> {
         FilePath::SystemVirtual(_) => return None,
     };
 
-    let settings = module_resolution_settings(db);
-
-    let mut search_paths = settings.search_paths(db);
+    let mut search_paths = search_paths(db);
 
     let module_name = loop {
         let candidate = search_paths.next()?;
@@ -119,106 +119,165 @@ pub(crate) fn file_to_module(db: &dyn Db, file: File) -> Option<Module> {
     }
 }
 
-/// Validate and normalize the raw settings given by the user
-/// into settings we can use for module resolution
-///
-/// This method also implements the typing spec's [module resolution order].
-///
-/// [module resolution order]: https://typing.readthedocs.io/en/latest/spec/distributing.html#import-resolution-ordering
-fn try_resolve_module_resolution_settings(
-    db: &dyn Db,
-) -> Result<ModuleResolutionSettings, SearchPathValidationError> {
-    let program = Program::get(db.upcast());
-
-    let SearchPathSettings {
-        extra_paths,
-        src_root,
-        custom_typeshed,
-        site_packages,
-    } = program.search_paths(db.upcast());
-
-    if !extra_paths.is_empty() {
-        tracing::info!("Extra search paths: {extra_paths:?}");
-    }
-
-    if let Some(custom_typeshed) = custom_typeshed {
-        tracing::info!("Custom typeshed directory: {custom_typeshed}");
-    }
-
-    let system = db.system();
-    let files = db.files();
-
-    let mut static_search_paths = vec![];
-
-    for path in extra_paths {
-        let search_path = SearchPath::extra(system, path.clone())?;
-        files.try_add_root(
-            db.upcast(),
-            search_path.as_system_path().unwrap(),
-            FileRootKind::LibrarySearchPath,
-        );
-        static_search_paths.push(search_path);
-    }
-
-    static_search_paths.push(SearchPath::first_party(system, src_root.clone())?);
-
-    static_search_paths.push(if let Some(custom_typeshed) = custom_typeshed.as_ref() {
-        let search_path = SearchPath::custom_stdlib(db, custom_typeshed.clone())?;
-        files.try_add_root(
-            db.upcast(),
-            search_path.as_system_path().unwrap(),
-            FileRootKind::LibrarySearchPath,
-        );
-        search_path
-    } else {
-        SearchPath::vendored_stdlib()
-    });
-
-    let mut site_packages_paths: Vec<_> = Vec::with_capacity(site_packages.len());
-
-    for path in site_packages {
-        let search_path = SearchPath::site_packages(system, path.to_path_buf())?;
-        files.try_add_root(
-            db.upcast(),
-            search_path.as_system_path().unwrap(),
-            FileRootKind::LibrarySearchPath,
-        );
-        site_packages_paths.push(search_path);
-    }
-
-    // TODO vendor typeshed's third-party stubs as well as the stdlib and fallback to them as a final step
-
-    let target_version = program.target_version(db.upcast());
-    tracing::info!("Target version: {target_version}");
-
-    // Filter out module resolution paths that point to the same directory on disk (the same invariant maintained by [`sys.path` at runtime]).
-    // (Paths may, however, *overlap* -- e.g. you could have both `src/` and `src/foo`
-    // as module resolution paths simultaneously.)
-    //
-    // [`sys.path` at runtime]: https://docs.python.org/3/library/site.html#module-site
-    // This code doesn't use an `IndexSet` because the key is the system path and not the search root.
-    let mut seen_paths =
-        FxHashSet::with_capacity_and_hasher(static_search_paths.len(), FxBuildHasher);
-
-    static_search_paths.retain(|path| {
-        if let Some(path) = path.as_system_path() {
-            seen_paths.insert(path.to_path_buf())
-        } else {
-            true
-        }
-    });
-
-    Ok(ModuleResolutionSettings {
-        target_version,
-        static_search_paths,
-        site_packages_paths,
-    })
+pub(crate) fn search_paths(db: &dyn Db) -> SearchPathIterator {
+    Program::get(db).search_paths(db).iter(db)
 }
 
-#[salsa::tracked(return_ref)]
-pub(crate) fn module_resolution_settings(db: &dyn Db) -> ModuleResolutionSettings {
-    // TODO proper error handling if this returns an error:
-    try_resolve_module_resolution_settings(db).unwrap()
+#[derive(Debug, PartialEq, Eq)]
+pub(crate) struct SearchPaths {
+    /// Search paths that have been statically determined purely from reading Ruff's configuration settings.
+    /// These shouldn't ever change unless the config settings themselves change.
+    static_paths: Vec<SearchPath>,
+
+    /// site-packages paths are not included in the above field:
+    /// if there are multiple site-packages paths, editable installations can appear
+    /// *between* the site-packages paths on `sys.path` at runtime.
+    /// That means we can't know where a second or third `site-packages` path should sit
+    /// in terms of module-resolution priority until we've discovered the editable installs
+    /// for the first `site-packages` path
+    site_packages: Vec<SearchPath>,
+
+    typeshed_versions: TypeshedVersions,
+}
+
+impl SearchPaths {
+    /// Validate and normalize the raw settings given by the user
+    /// into settings we can use for module resolution
+    ///
+    /// This method also implements the typing spec's [module resolution order].
+    ///
+    /// [module resolution order]: https://typing.readthedocs.io/en/latest/spec/distributing.html#import-resolution-ordering
+    pub(crate) fn from_settings(
+        db: &dyn Db,
+        settings: &SearchPathSettings,
+    ) -> Result<Self, SearchPathValidationError> {
+        fn canonicalize(path: &SystemPath, system: &dyn System) -> SystemPathBuf {
+            system
+                .canonicalize_path(path)
+                .unwrap_or_else(|_| path.to_path_buf())
+        }
+
+        let SearchPathSettings {
+            extra_paths,
+            src_root,
+            custom_typeshed,
+            site_packages: site_packages_paths,
+        } = settings;
+
+        let system = db.system();
+        let files = db.files();
+
+        let mut static_paths = vec![];
+
+        for path in extra_paths {
+            let path = canonicalize(path, system);
+            files.try_add_root(db.upcast(), &path, FileRootKind::LibrarySearchPath);
+            tracing::debug!("Adding extra search-path '{path}'");
+
+            static_paths.push(SearchPath::extra(system, path)?);
+        }
+
+        tracing::debug!("Adding first-party search path '{src_root}'");
+        static_paths.push(SearchPath::first_party(system, src_root.to_path_buf())?);
+
+        let (typeshed_versions, stdlib_path) = if let Some(custom_typeshed) = custom_typeshed {
+            let custom_typeshed = canonicalize(custom_typeshed, system);
+            tracing::debug!("Adding custom-stdlib search path '{custom_typeshed}'");
+
+            files.try_add_root(
+                db.upcast(),
+                &custom_typeshed,
+                FileRootKind::LibrarySearchPath,
+            );
+
+            let versions_path = custom_typeshed.join("stdlib/VERSIONS");
+
+            let versions_content = system.read_to_string(&versions_path).map_err(|error| {
+                SearchPathValidationError::FailedToReadVersionsFile {
+                    path: versions_path,
+                    error,
+                }
+            })?;
+
+            let parsed: TypeshedVersions = versions_content.parse()?;
+
+            let search_path = SearchPath::custom_stdlib(db, &custom_typeshed)?;
+
+            (parsed, search_path)
+        } else {
+            tracing::debug!("Using vendored stdlib");
+            (
+                vendored_typeshed_versions(db),
+                SearchPath::vendored_stdlib(),
+            )
+        };
+
+        static_paths.push(stdlib_path);
+
+        let site_packages_paths = match site_packages_paths {
+            SitePackages::Derived { venv_path } => VirtualEnvironment::new(venv_path, system)
+                .and_then(|venv| venv.site_packages_directories(system))?,
+            SitePackages::Known(paths) => paths
+                .iter()
+                .map(|path| canonicalize(path, system))
+                .collect(),
+        };
+
+        let mut site_packages: Vec<_> = Vec::with_capacity(site_packages_paths.len());
+
+        for path in site_packages_paths {
+            tracing::debug!("Adding site-packages search path '{path}'");
+            files.try_add_root(db.upcast(), &path, FileRootKind::LibrarySearchPath);
+            site_packages.push(SearchPath::site_packages(system, path)?);
+        }
+
+        // TODO vendor typeshed's third-party stubs as well as the stdlib and fallback to them as a final step
+
+        // Filter out module resolution paths that point to the same directory on disk (the same invariant maintained by [`sys.path` at runtime]).
+        // (Paths may, however, *overlap* -- e.g. you could have both `src/` and `src/foo`
+        // as module resolution paths simultaneously.)
+        //
+        // This code doesn't use an `IndexSet` because the key is the system path and not the search root.
+        //
+        // [`sys.path` at runtime]: https://docs.python.org/3/library/site.html#module-site
+        let mut seen_paths = FxHashSet::with_capacity_and_hasher(static_paths.len(), FxBuildHasher);
+
+        static_paths.retain(|path| {
+            if let Some(path) = path.as_system_path() {
+                seen_paths.insert(path.to_path_buf())
+            } else {
+                true
+            }
+        });
+
+        Ok(SearchPaths {
+            static_paths,
+            site_packages,
+            typeshed_versions,
+        })
+    }
+
+    pub(super) fn iter<'a>(&'a self, db: &'a dyn Db) -> SearchPathIterator<'a> {
+        SearchPathIterator {
+            db,
+            static_paths: self.static_paths.iter(),
+            dynamic_paths: None,
+        }
+    }
+
+    pub(crate) fn custom_stdlib(&self) -> Option<&SystemPath> {
+        self.static_paths.iter().find_map(|search_path| {
+            if search_path.is_standard_library() {
+                search_path.as_system_path()
+            } else {
+                None
+            }
+        })
+    }
+
+    pub(super) fn typeshed_versions(&self) -> &TypeshedVersions {
+        &self.typeshed_versions
+    }
 }
 
 /// Collect all dynamic search paths. For each `site-packages` path:
@@ -231,19 +290,21 @@ pub(crate) fn module_resolution_settings(db: &dyn Db) -> ModuleResolutionSetting
 /// module-resolution priority.
 #[salsa::tracked(return_ref)]
 pub(crate) fn dynamic_resolution_paths(db: &dyn Db) -> Vec<SearchPath> {
-    let ModuleResolutionSettings {
-        target_version: _,
-        static_search_paths,
-        site_packages_paths,
-    } = module_resolution_settings(db);
+    tracing::debug!("Resolving dynamic module resolution paths");
+
+    let SearchPaths {
+        static_paths,
+        site_packages,
+        typeshed_versions: _,
+    } = Program::get(db).search_paths(db);
 
     let mut dynamic_paths = Vec::new();
 
-    if site_packages_paths.is_empty() {
+    if site_packages.is_empty() {
         return dynamic_paths;
     }
 
-    let mut existing_paths: FxHashSet<_> = static_search_paths
+    let mut existing_paths: FxHashSet<_> = static_paths
         .iter()
         .filter_map(|path| path.as_system_path())
         .map(Cow::Borrowed)
@@ -252,7 +313,7 @@ pub(crate) fn dynamic_resolution_paths(db: &dyn Db) -> Vec<SearchPath> {
     let files = db.files();
     let system = db.system();
 
-    for site_packages_search_path in site_packages_paths {
+    for site_packages_search_path in site_packages {
         let site_packages_dir = site_packages_search_path
             .as_system_path()
             .expect("Expected site package path to be a system path");
@@ -299,9 +360,17 @@ pub(crate) fn dynamic_resolution_paths(db: &dyn Db) -> Vec<SearchPath> {
         let installations = all_pth_files.iter().flat_map(PthFile::items);
 
         for installation in installations {
+            let installation = system
+                .canonicalize_path(&installation)
+                .unwrap_or(installation);
+
             if existing_paths.insert(Cow::Owned(installation.clone())) {
-                match SearchPath::editable(system, installation) {
+                match SearchPath::editable(system, installation.clone()) {
                     Ok(search_path) => {
+                        tracing::debug!(
+                            "Adding editable installation to module resolution path {path}",
+                            path = installation
+                        );
                         dynamic_paths.push(search_path);
                     }
 
@@ -448,38 +517,6 @@ impl<'db> Iterator for PthFileIterator<'db> {
     }
 }
 
-/// Validated and normalized module-resolution settings.
-#[derive(Clone, Debug, PartialEq, Eq)]
-pub(crate) struct ModuleResolutionSettings {
-    target_version: PythonVersion,
-
-    /// Search paths that have been statically determined purely from reading Ruff's configuration settings.
-    /// These shouldn't ever change unless the config settings themselves change.
-    static_search_paths: Vec<SearchPath>,
-
-    /// site-packages paths are not included in the above field:
-    /// if there are multiple site-packages paths, editable installations can appear
-    /// *between* the site-packages paths on `sys.path` at runtime.
-    /// That means we can't know where a second or third `site-packages` path should sit
-    /// in terms of module-resolution priority until we've discovered the editable installs
-    /// for the first `site-packages` path
-    site_packages_paths: Vec<SearchPath>,
-}
-
-impl ModuleResolutionSettings {
-    fn target_version(&self) -> PythonVersion {
-        self.target_version
-    }
-
-    pub(crate) fn search_paths<'db>(&'db self, db: &'db dyn Db) -> SearchPathIterator<'db> {
-        SearchPathIterator {
-            db,
-            static_paths: self.static_search_paths.iter(),
-            dynamic_paths: None,
-        }
-    }
-}
-
 /// A thin wrapper around `ModuleName` to make it a Salsa ingredient.
 ///
 /// This is needed because Salsa requires that all query arguments are salsa ingredients.
@@ -492,13 +529,13 @@ struct ModuleNameIngredient<'db> {
 /// Given a module name and a list of search paths in which to lookup modules,
 /// attempt to resolve the module name
 fn resolve_name(db: &dyn Db, name: &ModuleName) -> Option<(SearchPath, File, ModuleKind)> {
-    let resolver_settings = module_resolution_settings(db);
-    let target_version = resolver_settings.target_version();
-    let resolver_state = ResolverState::new(db, target_version);
+    let program = Program::get(db);
+    let target_version = program.target_version(db);
+    let resolver_state = ResolverContext::new(db, target_version);
     let is_builtin_module =
         ruff_python_stdlib::sys::is_builtin_module(target_version.minor, name.as_str());
 
-    for search_path in resolver_settings.search_paths(db) {
+    for search_path in search_paths(db) {
         // When a builtin module is imported, standard module resolution is bypassed:
         // the module name always resolves to the stdlib module,
         // even if there's a module of the same name in the first-party root
@@ -516,24 +553,16 @@ fn resolve_name(db: &dyn Db, name: &ModuleName) -> Option<(SearchPath, File, Mod
 
                 package_path.push(module_name);
 
-                // Must be a `__init__.pyi` or `__init__.py` or it isn't a package.
-                let kind = if package_path.is_directory(&resolver_state) {
-                    package_path.push("__init__");
-                    ModuleKind::Package
-                } else {
-                    ModuleKind::Module
-                };
-
-                // TODO Implement full https://peps.python.org/pep-0561/#type-checker-module-resolution-order resolution
-                if let Some(stub) = package_path.with_pyi_extension().to_file(&resolver_state) {
-                    return Some((search_path.clone(), stub, kind));
+                // Check for a regular package first (highest priority)
+                package_path.push("__init__");
+                if let Some(regular_package) = resolve_file_module(&package_path, &resolver_state) {
+                    return Some((search_path.clone(), regular_package, ModuleKind::Package));
                 }
 
-                if let Some(module) = package_path
-                    .with_py_extension()
-                    .and_then(|path| path.to_file(&resolver_state))
-                {
-                    return Some((search_path.clone(), module, kind));
+                // Check for a file module next
+                package_path.pop();
+                if let Some(file_module) = resolve_file_module(&package_path, &resolver_state) {
+                    return Some((search_path.clone(), file_module, ModuleKind::Module));
                 }
 
                 // For regular packages, don't search the next search path. All files of that
@@ -554,10 +583,27 @@ fn resolve_name(db: &dyn Db, name: &ModuleName) -> Option<(SearchPath, File, Mod
     None
 }
 
+/// If `module` exists on disk with either a `.pyi` or `.py` extension,
+/// return the [`File`] corresponding to that path.
+///
+/// `.pyi` files take priority, as they always have priority when
+/// resolving modules.
+fn resolve_file_module(module: &ModulePath, resolver_state: &ResolverContext) -> Option<File> {
+    // Stubs have precedence over source files
+    module
+        .with_pyi_extension()
+        .to_file(resolver_state)
+        .or_else(|| {
+            module
+                .with_py_extension()
+                .and_then(|path| path.to_file(resolver_state))
+        })
+}
+
 fn resolve_package<'a, 'db, I>(
     module_search_path: &SearchPath,
     components: I,
-    resolver_state: &ResolverState<'db>,
+    resolver_state: &ResolverContext<'db>,
 ) -> Result<ResolvedPackage, PackageKind>
 where
     I: Iterator<Item = &'a str>,
@@ -580,7 +626,10 @@ where
 
         if is_regular_package {
             in_namespace_package = false;
-        } else if package_path.is_directory(resolver_state) {
+        } else if package_path.is_directory(resolver_state)
+            // Pure modules hide namespace packages with the same name
+            && resolve_file_module(&package_path, resolver_state).is_none()
+        {
             // A directory without an `__init__.py` is a namespace package, continue with the next folder.
             in_namespace_package = true;
         } else if in_namespace_package {
@@ -639,6 +688,21 @@ impl PackageKind {
     }
 }
 
+pub(super) struct ResolverContext<'db> {
+    pub(super) db: &'db dyn Db,
+    pub(super) target_version: PythonVersion,
+}
+
+impl<'db> ResolverContext<'db> {
+    pub(super) fn new(db: &'db dyn Db, target_version: PythonVersion) -> Self {
+        Self { db, target_version }
+    }
+
+    pub(super) fn vendored(&self) -> &VendoredFileSystem {
+        self.db.vendored()
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use ruff_db::files::{system_path_to_file, File, FilePath};
@@ -652,6 +716,8 @@ mod tests {
     use crate::module_name::ModuleName;
     use crate::module_resolver::module::ModuleKind;
     use crate::module_resolver::testing::{FileSpec, MockedTypeshed, TestCase, TestCaseBuilder};
+    use crate::ProgramSettings;
+    use crate::PythonVersion;
 
     use super::*;
 
@@ -791,7 +857,7 @@ mod tests {
                 "Search path for {module_name} was unexpectedly {search_path:?}"
             );
             assert!(
-                search_path.is_stdlib_search_path(),
+                search_path.is_standard_library(),
                 "Expected a stdlib search path, but got {search_path:?}"
             );
         }
@@ -887,7 +953,7 @@ mod tests {
                 "Search path for {module_name} was unexpectedly {search_path:?}"
             );
             assert!(
-                search_path.is_stdlib_search_path(),
+                search_path.is_standard_library(),
                 "Expected a stdlib search path, but got {search_path:?}"
             );
         }
@@ -1021,6 +1087,25 @@ mod tests {
         );
     }
 
+    #[test]
+    fn single_file_takes_priority_over_namespace_package() {
+        //const SRC: &[FileSpec] = &[("foo.py", "x = 1")];
+        const SRC: &[FileSpec] = &[("foo.py", "x = 1"), ("foo/bar.py", "x = 2")];
+
+        let TestCase { db, src, .. } = TestCaseBuilder::new().with_src_files(SRC).build();
+
+        let foo_module_name = ModuleName::new_static("foo").unwrap();
+        let foo_bar_module_name = ModuleName::new_static("foo.bar").unwrap();
+
+        // `foo.py` takes priority over the `foo` namespace package
+        let foo_module = resolve_module(&db, foo_module_name.clone()).unwrap();
+        assert_eq!(foo_module.file().path(&db), &src.join("foo.py"));
+
+        // `foo.bar` isn't recognised as a module
+        let foo_bar_module = resolve_module(&db, foo_bar_module_name.clone());
+        assert_eq!(foo_bar_module, None);
+    }
+
     #[test]
     fn typing_stub_over_module() {
         const SRC: &[FileSpec] = &[("foo.py", "print('Hello, world!')"), ("foo.pyi", "x: int")];
@@ -1202,14 +1287,19 @@ mod tests {
         std::fs::write(foo.as_std_path(), "")?;
         std::os::unix::fs::symlink(foo.as_std_path(), bar.as_std_path())?;
 
-        let search_paths = SearchPathSettings {
-            extra_paths: vec![],
-            src_root: src.clone(),
-            custom_typeshed: Some(custom_typeshed.clone()),
-            site_packages: vec![site_packages],
-        };
-
-        Program::new(&db, PythonVersion::PY38, search_paths);
+        Program::from_settings(
+            &db,
+            &ProgramSettings {
+                target_version: PythonVersion::PY38,
+                search_paths: SearchPathSettings {
+                    extra_paths: vec![],
+                    src_root: src.clone(),
+                    custom_typeshed: Some(custom_typeshed.clone()),
+                    site_packages: SitePackages::Known(vec![site_packages]),
+                },
+            },
+        )
+        .context("Invalid program settings")?;
 
         let foo_module = resolve_module(&db, ModuleName::new_static("foo").unwrap()).unwrap();
         let bar_module = resolve_module(&db, ModuleName::new_static("bar").unwrap()).unwrap();
@@ -1673,8 +1763,7 @@ not_a_directory
             .with_site_packages_files(&[("_foo.pth", "/src")])
             .build();
 
-        let search_paths: Vec<&SearchPath> =
-            module_resolution_settings(&db).search_paths(&db).collect();
+        let search_paths: Vec<&SearchPath> = search_paths(&db).collect();
 
         assert!(search_paths.contains(
             &&SearchPath::first_party(db.system(), SystemPathBuf::from("/src")).unwrap()
@@ -1703,16 +1792,22 @@ not_a_directory
         ])
         .unwrap();
 
-        Program::new(
+        Program::from_settings(
             &db,
-            PythonVersion::default(),
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: SystemPathBuf::from("/src"),
-                custom_typeshed: None,
-                site_packages: vec![venv_site_packages, system_site_packages],
+            &ProgramSettings {
+                target_version: PythonVersion::default(),
+                search_paths: SearchPathSettings {
+                    extra_paths: vec![],
+                    src_root: SystemPathBuf::from("/src"),
+                    custom_typeshed: None,
+                    site_packages: SitePackages::Known(vec![
+                        venv_site_packages,
+                        system_site_packages,
+                    ]),
+                },
             },
-        );
+        )
+        .expect("Valid program settings");
 
         // The editable installs discovered from the `.pth` file in the first `site-packages` directory
         // take precedence over the second `site-packages` directory...

crates/red_knot_python_semantic/src/module_resolver/state.rs

@@ -1,25 +0,0 @@
-use ruff_db::vendored::VendoredFileSystem;
-
-use super::typeshed::LazyTypeshedVersions;
-use crate::db::Db;
-use crate::python_version::PythonVersion;
-
-pub(crate) struct ResolverState<'db> {
-    pub(crate) db: &'db dyn Db,
-    pub(crate) typeshed_versions: LazyTypeshedVersions<'db>,
-    pub(crate) target_version: PythonVersion,
-}
-
-impl<'db> ResolverState<'db> {
-    pub(crate) fn new(db: &'db dyn Db, target_version: PythonVersion) -> Self {
-        Self {
-            db,
-            typeshed_versions: LazyTypeshedVersions::new(),
-            target_version,
-        }
-    }
-
-    pub(crate) fn vendored(&self) -> &VendoredFileSystem {
-        self.db.vendored()
-    }
-}

crates/red_knot_python_semantic/src/module_resolver/testing.rs

@@ -4,6 +4,7 @@ use ruff_db::vendored::VendoredPathBuf;
 use crate::db::tests::TestDb;
 use crate::program::{Program, SearchPathSettings};
 use crate::python_version::PythonVersion;
+use crate::{ProgramSettings, SitePackages};
 
 /// A test case for the module resolver.
 ///
@@ -178,6 +179,7 @@ impl TestCaseBuilder<UnspecifiedTypeshed> {
             first_party_files,
             site_packages_files,
         } = self;
+
         TestCaseBuilder {
             typeshed_option: typeshed,
             target_version,
@@ -194,6 +196,7 @@ impl TestCaseBuilder<UnspecifiedTypeshed> {
             site_packages,
             target_version,
         } = self.with_custom_typeshed(MockedTypeshed::default()).build();
+
         TestCase {
             db,
             src,
@@ -220,16 +223,19 @@ impl TestCaseBuilder<MockedTypeshed> {
         let src = Self::write_mock_directory(&mut db, "/src", first_party_files);
         let typeshed = Self::build_typeshed_mock(&mut db, &typeshed_option);
 
-        Program::new(
+        Program::from_settings(
             &db,
-            target_version,
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: src.clone(),
-                custom_typeshed: Some(typeshed.clone()),
-                site_packages: vec![site_packages.clone()],
+            &ProgramSettings {
+                target_version,
+                search_paths: SearchPathSettings {
+                    extra_paths: vec![],
+                    src_root: src.clone(),
+                    custom_typeshed: Some(typeshed.clone()),
+                    site_packages: SitePackages::Known(vec![site_packages.clone()]),
+                },
             },
-        );
+        )
+        .expect("Valid program settings");
 
         TestCase {
             db,
@@ -273,16 +279,17 @@ impl TestCaseBuilder<VendoredTypeshed> {
             Self::write_mock_directory(&mut db, "/site-packages", site_packages_files);
         let src = Self::write_mock_directory(&mut db, "/src", first_party_files);
 
-        Program::new(
+        Program::from_settings(
             &db,
-            target_version,
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: src.clone(),
-                custom_typeshed: None,
-                site_packages: vec![site_packages.clone()],
+            &ProgramSettings {
+                target_version,
+                search_paths: SearchPathSettings {
+                    site_packages: SitePackages::Known(vec![site_packages.clone()]),
+                    ..SearchPathSettings::new(src.clone())
+                },
             },
-        );
+        )
+        .expect("Valid search path settings");
 
         TestCase {
             db,

crates/red_knot_python_semantic/src/module_resolver/typeshed.rs

@@ -1,92 +1,27 @@
-use std::cell::OnceCell;
 use std::collections::BTreeMap;
 use std::fmt;
 use std::num::{NonZeroU16, NonZeroUsize};
 use std::ops::{RangeFrom, RangeInclusive};
 use std::str::FromStr;
 
-use once_cell::sync::Lazy;
-use ruff_db::system::SystemPath;
 use rustc_hash::FxHashMap;
 
-use ruff_db::files::{system_path_to_file, File};
-
-use super::vendored::vendored_typeshed_stubs;
 use crate::db::Db;
 use crate::module_name::ModuleName;
-use crate::python_version::PythonVersion;
+use crate::{Program, PythonVersion};
 
-#[derive(Debug)]
-pub(crate) struct LazyTypeshedVersions<'db>(OnceCell<&'db TypeshedVersions>);
-
-impl<'db> LazyTypeshedVersions<'db> {
-    #[must_use]
-    pub(crate) fn new() -> Self {
-        Self(OnceCell::new())
-    }
-
-    /// Query whether a module exists at runtime in the stdlib on a certain Python version.
-    ///
-    /// Simply probing whether a file exists in typeshed is insufficient for this question,
-    /// as a module in the stdlib may have been added in Python 3.10, but the typeshed stub
-    /// will still be available (either in a custom typeshed dir or in our vendored copy)
-    /// even if the user specified Python 3.8 as the target version.
-    ///
-    /// For top-level modules and packages, the VERSIONS file can always provide an unambiguous answer
-    /// as to whether the module exists on the specified target version. However, VERSIONS does not
-    /// provide comprehensive information on all submodules, meaning that this method sometimes
-    /// returns [`TypeshedVersionsQueryResult::MaybeExists`].
-    /// See [`TypeshedVersionsQueryResult`] for more details.
-    #[must_use]
-    pub(crate) fn query_module(
-        &self,
-        db: &'db dyn Db,
-        module: &ModuleName,
-        stdlib_root: Option<&SystemPath>,
-        target_version: PythonVersion,
-    ) -> TypeshedVersionsQueryResult {
-        let versions = self.0.get_or_init(|| {
-            let versions_path = if let Some(system_path) = stdlib_root {
-                system_path.join("VERSIONS")
-            } else {
-                return &VENDORED_VERSIONS;
-            };
-            let Ok(versions_file) = system_path_to_file(db.upcast(), &versions_path) else {
-                todo!(
-                    "Still need to figure out how to handle VERSIONS files being deleted \
-                    from custom typeshed directories! Expected a file to exist at {versions_path}"
-                )
-            };
-            // TODO(Alex/Micha): If VERSIONS is invalid,
-            // this should invalidate not just the specific module resolution we're currently attempting,
-            // but all type inference that depends on any standard-library types.
-            // Unwrapping here is not correct...
-            parse_typeshed_versions(db, versions_file).as_ref().unwrap()
-        });
-        versions.query_module(module, target_version)
-    }
-}
-
-#[salsa::tracked(return_ref)]
-pub(crate) fn parse_typeshed_versions(
-    db: &dyn Db,
-    versions_file: File,
-) -> Result<TypeshedVersions, TypeshedVersionsParseError> {
-    // TODO: Handle IO errors
-    let file_content = versions_file
-        .read_to_string(db.upcast())
-        .unwrap_or_default();
-    file_content.parse()
-}
-
-static VENDORED_VERSIONS: Lazy<TypeshedVersions> = Lazy::new(|| {
+pub(in crate::module_resolver) fn vendored_typeshed_versions(db: &dyn Db) -> TypeshedVersions {
     TypeshedVersions::from_str(
-        &vendored_typeshed_stubs()
+        &db.vendored()
             .read_to_string("stdlib/VERSIONS")
-            .unwrap(),
+            .expect("The vendored typeshed stubs should contain a VERSIONS file"),
     )
-    .unwrap()
-});
+    .expect("The VERSIONS file in the vendored typeshed stubs should be well-formed")
+}
+
+pub(crate) fn typeshed_versions(db: &dyn Db) -> &TypeshedVersions {
+    Program::get(db).search_paths(db).typeshed_versions()
+}
 
 #[derive(Debug, PartialEq, Eq, Clone)]
 pub(crate) struct TypeshedVersionsParseError {
@@ -174,7 +109,7 @@ impl TypeshedVersions {
     }
 
     #[must_use]
-    fn query_module(
+    pub(in crate::module_resolver) fn query_module(
         &self,
         module: &ModuleName,
         target_version: PythonVersion,
@@ -204,7 +139,7 @@ impl TypeshedVersions {
     }
 }
 
-/// Possible answers [`LazyTypeshedVersions::query_module()`] could give to the question:
+/// Possible answers [`TypeshedVersions::query_module()`] could give to the question:
 /// "Does this module exist in the stdlib at runtime on a certain target version?"
 #[derive(Debug, Copy, PartialEq, Eq, Clone, Hash)]
 pub(crate) enum TypeshedVersionsQueryResult {
@@ -391,6 +326,8 @@ mod tests {
 
     use insta::assert_snapshot;
 
+    use crate::db::tests::TestDb;
+
     use super::*;
 
     const TYPESHED_STDLIB_DIR: &str = "stdlib";
@@ -412,12 +349,9 @@ mod tests {
 
     #[test]
     fn can_parse_vendored_versions_file() {
-        let versions_data = include_str!(concat!(
-            env!("CARGO_MANIFEST_DIR"),
-            "/vendor/typeshed/stdlib/VERSIONS"
-        ));
+        let db = TestDb::new();
 
-        let versions = TypeshedVersions::from_str(versions_data).unwrap();
+        let versions = vendored_typeshed_versions(&db);
         assert!(versions.len() > 100);
         assert!(versions.len() < 1000);
 
@@ -454,9 +388,10 @@ mod tests {
 
     #[test]
     fn typeshed_versions_consistent_with_vendored_stubs() {
-        const VERSIONS_DATA: &str = include_str!("../../../vendor/typeshed/stdlib/VERSIONS");
-        let vendored_typeshed_dir = Path::new("vendor/typeshed").canonicalize().unwrap();
-        let vendored_typeshed_versions = TypeshedVersions::from_str(VERSIONS_DATA).unwrap();
+        let db = TestDb::new();
+        let vendored_typeshed_versions = vendored_typeshed_versions(&db);
+        let vendored_typeshed_dir =
+            Path::new(env!("CARGO_MANIFEST_DIR")).join("../ruff_vendored/vendor/typeshed");
 
         let mut empty_iterator = true;
 

crates/red_knot_python_semantic/src/module_resolver/typeshed/mod.rs

@@ -1,8 +0,0 @@
-pub use self::vendored::vendored_typeshed_stubs;
-pub(super) use self::versions::{
-    parse_typeshed_versions, LazyTypeshedVersions, TypeshedVersionsParseError,
-    TypeshedVersionsQueryResult,
-};
-
-mod vendored;
-mod versions;

crates/red_knot_python_semantic/src/program.rs

@@ -1,32 +1,66 @@
 use crate::python_version::PythonVersion;
-use crate::Db;
-use ruff_db::system::SystemPathBuf;
+use anyhow::Context;
 use salsa::Durability;
+use salsa::Setter;
+
+use ruff_db::system::{SystemPath, SystemPathBuf};
+
+use crate::module_resolver::SearchPaths;
+use crate::Db;
 
 #[salsa::input(singleton)]
 pub struct Program {
     pub target_version: PythonVersion,
 
     #[return_ref]
-    pub search_paths: SearchPathSettings,
+    pub(crate) search_paths: SearchPaths,
 }
 
 impl Program {
-    pub fn from_settings(db: &dyn Db, settings: ProgramSettings) -> Self {
-        Program::builder(settings.target_version, settings.search_paths)
+    pub fn from_settings(db: &dyn Db, settings: &ProgramSettings) -> anyhow::Result<Self> {
+        let ProgramSettings {
+            target_version,
+            search_paths,
+        } = settings;
+
+        tracing::info!("Target version: Python {target_version}");
+
+        let search_paths = SearchPaths::from_settings(db, search_paths)
+            .with_context(|| "Invalid search path settings")?;
+
+        Ok(Program::builder(settings.target_version, search_paths)
             .durability(Durability::HIGH)
-            .new(db)
+            .new(db))
+    }
+
+    pub fn update_search_paths(
+        self,
+        db: &mut dyn Db,
+        search_path_settings: &SearchPathSettings,
+    ) -> anyhow::Result<()> {
+        let search_paths = SearchPaths::from_settings(db, search_path_settings)?;
+
+        if self.search_paths(db) != &search_paths {
+            tracing::debug!("Update search paths");
+            self.set_search_paths(db).to(search_paths);
+        }
+
+        Ok(())
+    }
+
+    pub fn custom_stdlib_search_path(self, db: &dyn Db) -> Option<&SystemPath> {
+        self.search_paths(db).custom_stdlib()
     }
 }
 
-#[derive(Debug, Eq, PartialEq)]
+#[derive(Clone, Debug, Eq, PartialEq)]
 pub struct ProgramSettings {
     pub target_version: PythonVersion,
     pub search_paths: SearchPathSettings,
 }
 
 /// Configures the search paths for module resolution.
-#[derive(Eq, PartialEq, Debug, Clone, Default)]
+#[derive(Eq, PartialEq, Debug, Clone)]
 pub struct SearchPathSettings {
     /// List of user-provided paths that should take first priority in the module resolution.
     /// Examples in other type checkers are mypy's MYPYPATH environment variable,
@@ -42,5 +76,25 @@ pub struct SearchPathSettings {
     pub custom_typeshed: Option<SystemPathBuf>,
 
     /// The path to the user's `site-packages` directory, where third-party packages from ``PyPI`` are installed.
-    pub site_packages: Vec<SystemPathBuf>,
+    pub site_packages: SitePackages,
+}
+
+impl SearchPathSettings {
+    pub fn new(src_root: SystemPathBuf) -> Self {
+        Self {
+            src_root,
+            extra_paths: vec![],
+            custom_typeshed: None,
+            site_packages: SitePackages::Known(vec![]),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Eq, PartialEq)]
+pub enum SitePackages {
+    Derived {
+        venv_path: SystemPathBuf,
+    },
+    /// Resolved site packages paths
+    Known(Vec<SystemPathBuf>),
 }

crates/red_knot_python_semantic/src/python_version.rs

@@ -54,6 +54,13 @@ impl TryFrom<(&str, &str)> for PythonVersion {
     }
 }
 
+impl From<(u8, u8)> for PythonVersion {
+    fn from(value: (u8, u8)) -> Self {
+        let (major, minor) = value;
+        Self { major, minor }
+    }
+}
+
 impl fmt::Display for PythonVersion {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         let PythonVersion { major, minor } = self;

crates/red_knot_python_semantic/src/semantic_index.rs

@@ -16,17 +16,21 @@ use crate::semantic_index::expression::Expression;
 use crate::semantic_index::symbol::{
     FileScopeId, NodeWithScopeKey, NodeWithScopeRef, Scope, ScopeId, ScopedSymbolId, SymbolTable,
 };
+use crate::semantic_index::use_def::UseDefMap;
 use crate::Db;
 
-pub(crate) use self::use_def::UseDefMap;
-
 pub mod ast_ids;
 mod builder;
+pub(crate) mod constraint;
 pub mod definition;
 pub mod expression;
 pub mod symbol;
 mod use_def;
 
+pub(crate) use self::use_def::{
+    BindingWithConstraints, BindingWithConstraintsIterator, DeclarationsIterator,
+};
+
 type SymbolMap = hashbrown::HashMap<ScopedSymbolId, (), ()>;
 
 /// Returns the semantic index for `file`.
@@ -89,8 +93,6 @@ pub(crate) struct SemanticIndex<'db> {
     scopes: IndexVec<FileScopeId, Scope>,
 
     /// Map expressions to their corresponding scope.
-    /// We can't use [`ExpressionId`] here, because the challenge is how to get from
-    /// an [`ast::Expr`] to an [`ExpressionId`] (which requires knowing the scope).
     scopes_by_expression: FxHashMap<ExpressionNodeKey, FileScopeId>,
 
     /// Map from a node creating a definition to its definition.
@@ -113,12 +115,15 @@ pub(crate) struct SemanticIndex<'db> {
     /// Note: We should not depend on this map when analysing other files or
     /// changing a file invalidates all dependents.
     ast_ids: IndexVec<FileScopeId, AstIds>,
+
+    /// Flags about the global scope (code usage impacting inference)
+    has_future_annotations: bool,
 }
 
 impl<'db> SemanticIndex<'db> {
     /// Returns the symbol table for a specific scope.
     ///
-    /// Use the Salsa cached [`symbol_table`] query if you only need the
+    /// Use the Salsa cached [`symbol_table()`] query if you only need the
     /// symbol table for a single scope.
     pub(super) fn symbol_table(&self, scope_id: FileScopeId) -> Arc<SymbolTable> {
         self.symbol_tables[scope_id].clone()
@@ -126,7 +131,7 @@ impl<'db> SemanticIndex<'db> {
 
     /// Returns the use-def map for a specific scope.
     ///
-    /// Use the Salsa cached [`use_def_map`] query if you only need the
+    /// Use the Salsa cached [`use_def_map()`] query if you only need the
     /// use-def map for a single scope.
     pub(super) fn use_def_map(&self, scope_id: FileScopeId) -> Arc<UseDefMap> {
         self.use_def_maps[scope_id].clone()
@@ -155,6 +160,10 @@ impl<'db> SemanticIndex<'db> {
         &self.scopes[id]
     }
 
+    pub(crate) fn scope_ids(&self) -> impl Iterator<Item = ScopeId> {
+        self.scope_ids_by_scope.iter().copied()
+    }
+
     /// Returns the id of the parent scope.
     pub(crate) fn parent_scope_id(&self, scope_id: FileScopeId) -> Option<FileScopeId> {
         let scope = self.scope(scope_id);
@@ -209,6 +218,12 @@ impl<'db> SemanticIndex<'db> {
     pub(crate) fn node_scope(&self, node: NodeWithScopeRef) -> FileScopeId {
         self.scopes_by_node[&node.node_key()]
     }
+
+    /// Checks if there is an import of `__future__.annotations` in the global scope, which affects
+    /// the logic for type inference.
+    pub(super) fn has_future_annotations(&self) -> bool {
+        self.has_future_annotations
+    }
 }
 
 pub struct AncestorsIter<'a> {
@@ -309,14 +324,32 @@ mod tests {
     use ruff_db::parsed::parsed_module;
     use ruff_db::system::DbWithTestSystem;
     use ruff_python_ast as ast;
+    use ruff_text_size::{Ranged, TextRange};
 
     use crate::db::tests::TestDb;
-    use crate::semantic_index::ast_ids::HasScopedUseId;
-    use crate::semantic_index::definition::DefinitionKind;
-    use crate::semantic_index::symbol::{FileScopeId, Scope, ScopeKind, SymbolTable};
+    use crate::semantic_index::ast_ids::{HasScopedUseId, ScopedUseId};
+    use crate::semantic_index::definition::{Definition, DefinitionKind};
+    use crate::semantic_index::symbol::{
+        FileScopeId, Scope, ScopeKind, ScopedSymbolId, SymbolTable,
+    };
+    use crate::semantic_index::use_def::UseDefMap;
     use crate::semantic_index::{global_scope, semantic_index, symbol_table, use_def_map};
     use crate::Db;
 
+    impl UseDefMap<'_> {
+        fn first_public_binding(&self, symbol: ScopedSymbolId) -> Option<Definition<'_>> {
+            self.public_bindings(symbol)
+                .next()
+                .map(|constrained_binding| constrained_binding.binding)
+        }
+
+        fn first_binding_at_use(&self, use_id: ScopedUseId) -> Option<Definition<'_>> {
+            self.bindings_at_use(use_id)
+                .next()
+                .map(|constrained_binding| constrained_binding.binding)
+        }
+    }
+
     struct TestCase {
         db: TestDb,
         file: File,
@@ -375,10 +408,8 @@ mod tests {
         let foo = global_table.symbol_id_by_name("foo").unwrap();
 
         let use_def = use_def_map(&db, scope);
-        let [definition] = use_def.public_definitions(foo) else {
-            panic!("expected one definition");
-        };
-        assert!(matches!(definition.node(&db), DefinitionKind::Import(_)));
+        let binding = use_def.first_public_binding(foo).unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::Import(_)));
     }
 
     #[test]
@@ -407,22 +438,19 @@ mod tests {
         assert!(
             global_table
                 .symbol_by_name("foo")
-                .is_some_and(|symbol| { symbol.is_defined() && !symbol.is_used() }),
+                .is_some_and(|symbol| { symbol.is_bound() && !symbol.is_used() }),
             "symbols that are defined get the defined flag"
         );
 
         let use_def = use_def_map(&db, scope);
-        let [definition] = use_def.public_definitions(
-            global_table
-                .symbol_id_by_name("foo")
-                .expect("symbol to exist"),
-        ) else {
-            panic!("expected one definition");
-        };
-        assert!(matches!(
-            definition.node(&db),
-            DefinitionKind::ImportFrom(_)
-        ));
+        let binding = use_def
+            .first_public_binding(
+                global_table
+                    .symbol_id_by_name("foo")
+                    .expect("symbol to exist"),
+            )
+            .unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::ImportFrom(_)));
     }
 
     #[test]
@@ -435,18 +463,32 @@ mod tests {
         assert!(
             global_table
                 .symbol_by_name("foo")
-                .is_some_and(|symbol| { !symbol.is_defined() && symbol.is_used() }),
-            "a symbol used but not defined in a scope should have only the used flag"
+                .is_some_and(|symbol| { !symbol.is_bound() && symbol.is_used() }),
+            "a symbol used but not bound in a scope should have only the used flag"
         );
         let use_def = use_def_map(&db, scope);
-        let [definition] =
-            use_def.public_definitions(global_table.symbol_id_by_name("x").expect("symbol exists"))
-        else {
-            panic!("expected one definition");
-        };
+        let binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("x").expect("symbol exists"))
+            .unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::Assignment(_)));
+    }
+
+    #[test]
+    fn augmented_assignment() {
+        let TestCase { db, file } = test_case("x += 1");
+        let scope = global_scope(&db, file);
+        let global_table = symbol_table(&db, scope);
+
+        assert_eq!(names(&global_table), vec!["x"]);
+
+        let use_def = use_def_map(&db, scope);
+        let binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("x").unwrap())
+            .unwrap();
+
         assert!(matches!(
-            definition.node(&db),
-            DefinitionKind::Assignment(_)
+            binding.kind(&db),
+            DefinitionKind::AugmentedAssignment(_)
         ));
     }
 
@@ -478,15 +520,10 @@ y = 2
         assert_eq!(names(&class_table), vec!["x"]);
 
         let use_def = index.use_def_map(class_scope_id);
-        let [definition] =
-            use_def.public_definitions(class_table.symbol_id_by_name("x").expect("symbol exists"))
-        else {
-            panic!("expected one definition");
-        };
-        assert!(matches!(
-            definition.node(&db),
-            DefinitionKind::Assignment(_)
-        ));
+        let binding = use_def
+            .first_public_binding(class_table.symbol_id_by_name("x").expect("symbol exists"))
+            .unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::Assignment(_)));
     }
 
     #[test]
@@ -516,17 +553,298 @@ y = 2
         assert_eq!(names(&function_table), vec!["x"]);
 
         let use_def = index.use_def_map(function_scope_id);
-        let [definition] = use_def.public_definitions(
-            function_table
-                .symbol_id_by_name("x")
-                .expect("symbol exists"),
-        ) else {
-            panic!("expected one definition");
+        let binding = use_def
+            .first_public_binding(
+                function_table
+                    .symbol_id_by_name("x")
+                    .expect("symbol exists"),
+            )
+            .unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::Assignment(_)));
+    }
+
+    #[test]
+    fn function_parameter_symbols() {
+        let TestCase { db, file } = test_case(
+            "
+def f(a: str, /, b: str, c: int = 1, *args, d: int = 2, **kwargs):
+    pass
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let global_table = symbol_table(&db, global_scope(&db, file));
+
+        assert_eq!(names(&global_table), vec!["str", "int", "f"]);
+
+        let [(function_scope_id, _function_scope)] = index
+            .child_scopes(FileScopeId::global())
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("Expected a function scope")
         };
-        assert!(matches!(
-            definition.node(&db),
-            DefinitionKind::Assignment(_)
-        ));
+
+        let function_table = index.symbol_table(function_scope_id);
+        assert_eq!(
+            names(&function_table),
+            vec!["a", "b", "c", "args", "d", "kwargs"],
+        );
+
+        let use_def = index.use_def_map(function_scope_id);
+        for name in ["a", "b", "c", "d"] {
+            let binding = use_def
+                .first_public_binding(
+                    function_table
+                        .symbol_id_by_name(name)
+                        .expect("symbol exists"),
+                )
+                .unwrap();
+            assert!(matches!(
+                binding.kind(&db),
+                DefinitionKind::ParameterWithDefault(_)
+            ));
+        }
+        for name in ["args", "kwargs"] {
+            let binding = use_def
+                .first_public_binding(
+                    function_table
+                        .symbol_id_by_name(name)
+                        .expect("symbol exists"),
+                )
+                .unwrap();
+            assert!(matches!(binding.kind(&db), DefinitionKind::Parameter(_)));
+        }
+    }
+
+    #[test]
+    fn lambda_parameter_symbols() {
+        let TestCase { db, file } = test_case("lambda a, b, c=1, *args, d=2, **kwargs: None");
+
+        let index = semantic_index(&db, file);
+        let global_table = symbol_table(&db, global_scope(&db, file));
+
+        assert!(names(&global_table).is_empty());
+
+        let [(lambda_scope_id, _lambda_scope)] = index
+            .child_scopes(FileScopeId::global())
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("Expected a lambda scope")
+        };
+
+        let lambda_table = index.symbol_table(lambda_scope_id);
+        assert_eq!(
+            names(&lambda_table),
+            vec!["a", "b", "c", "args", "d", "kwargs"],
+        );
+
+        let use_def = index.use_def_map(lambda_scope_id);
+        for name in ["a", "b", "c", "d"] {
+            let binding = use_def
+                .first_public_binding(lambda_table.symbol_id_by_name(name).expect("symbol exists"))
+                .unwrap();
+            assert!(matches!(
+                binding.kind(&db),
+                DefinitionKind::ParameterWithDefault(_)
+            ));
+        }
+        for name in ["args", "kwargs"] {
+            let binding = use_def
+                .first_public_binding(lambda_table.symbol_id_by_name(name).expect("symbol exists"))
+                .unwrap();
+            assert!(matches!(binding.kind(&db), DefinitionKind::Parameter(_)));
+        }
+    }
+
+    /// Test case to validate that the comprehension scope is correctly identified and that the target
+    /// variable is defined only in the comprehension scope and not in the global scope.
+    #[test]
+    fn comprehension_scope() {
+        let TestCase { db, file } = test_case(
+            "
+[x for x, y in iter1]
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let global_table = index.symbol_table(FileScopeId::global());
+
+        assert_eq!(names(&global_table), vec!["iter1"]);
+
+        let [(comprehension_scope_id, comprehension_scope)] = index
+            .child_scopes(FileScopeId::global())
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("expected one child scope")
+        };
+
+        assert_eq!(comprehension_scope.kind(), ScopeKind::Comprehension);
+        assert_eq!(
+            comprehension_scope_id.to_scope_id(&db, file).name(&db),
+            "<listcomp>"
+        );
+
+        let comprehension_symbol_table = index.symbol_table(comprehension_scope_id);
+
+        assert_eq!(names(&comprehension_symbol_table), vec!["x", "y"]);
+
+        let use_def = index.use_def_map(comprehension_scope_id);
+        for name in ["x", "y"] {
+            let binding = use_def
+                .first_public_binding(
+                    comprehension_symbol_table
+                        .symbol_id_by_name(name)
+                        .expect("symbol exists"),
+                )
+                .unwrap();
+            assert!(matches!(
+                binding.kind(&db),
+                DefinitionKind::Comprehension(_)
+            ));
+        }
+    }
+
+    /// Test case to validate that the `x` variable used in the comprehension is referencing the
+    /// `x` variable defined by the inner generator (`for x in iter2`) and not the outer one.
+    #[test]
+    fn multiple_generators() {
+        let TestCase { db, file } = test_case(
+            "
+[x for x in iter1 for x in iter2]
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let [(comprehension_scope_id, _)] = index
+            .child_scopes(FileScopeId::global())
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("expected one child scope")
+        };
+
+        let use_def = index.use_def_map(comprehension_scope_id);
+
+        let module = parsed_module(&db, file).syntax();
+        let element = module.body[0]
+            .as_expr_stmt()
+            .unwrap()
+            .value
+            .as_list_comp_expr()
+            .unwrap()
+            .elt
+            .as_name_expr()
+            .unwrap();
+        let element_use_id =
+            element.scoped_use_id(&db, comprehension_scope_id.to_scope_id(&db, file));
+
+        let binding = use_def.first_binding_at_use(element_use_id).unwrap();
+        let DefinitionKind::Comprehension(comprehension) = binding.kind(&db) else {
+            panic!("expected generator definition")
+        };
+        let target = comprehension.target();
+        let name = target.id().as_str();
+
+        assert_eq!(name, "x");
+        assert_eq!(target.range(), TextRange::new(23.into(), 24.into()));
+    }
+
+    /// Test case to validate that the nested comprehension creates a new scope which is a child of
+    /// the outer comprehension scope and the variables are correctly defined in the respective
+    /// scopes.
+    #[test]
+    fn nested_generators() {
+        let TestCase { db, file } = test_case(
+            "
+[{x for x in iter2} for y in iter1]
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let global_table = index.symbol_table(FileScopeId::global());
+
+        assert_eq!(names(&global_table), vec!["iter1"]);
+
+        let [(comprehension_scope_id, comprehension_scope)] = index
+            .child_scopes(FileScopeId::global())
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("expected one child scope")
+        };
+
+        assert_eq!(comprehension_scope.kind(), ScopeKind::Comprehension);
+        assert_eq!(
+            comprehension_scope_id.to_scope_id(&db, file).name(&db),
+            "<listcomp>"
+        );
+
+        let comprehension_symbol_table = index.symbol_table(comprehension_scope_id);
+
+        assert_eq!(names(&comprehension_symbol_table), vec!["y", "iter2"]);
+
+        let [(inner_comprehension_scope_id, inner_comprehension_scope)] = index
+            .child_scopes(comprehension_scope_id)
+            .collect::<Vec<_>>()[..]
+        else {
+            panic!("expected one inner generator scope")
+        };
+
+        assert_eq!(inner_comprehension_scope.kind(), ScopeKind::Comprehension);
+        assert_eq!(
+            inner_comprehension_scope_id
+                .to_scope_id(&db, file)
+                .name(&db),
+            "<setcomp>"
+        );
+
+        let inner_comprehension_symbol_table = index.symbol_table(inner_comprehension_scope_id);
+
+        assert_eq!(names(&inner_comprehension_symbol_table), vec!["x"]);
+    }
+
+    #[test]
+    fn with_item_definition() {
+        let TestCase { db, file } = test_case(
+            "
+with item1 as x, item2 as y:
+    pass
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let global_table = index.symbol_table(FileScopeId::global());
+
+        assert_eq!(names(&global_table), vec!["item1", "x", "item2", "y"]);
+
+        let use_def = index.use_def_map(FileScopeId::global());
+        for name in ["x", "y"] {
+            let binding = use_def
+                .first_public_binding(global_table.symbol_id_by_name(name).expect("symbol exists"))
+                .expect("Expected with item definition for {name}");
+            assert!(matches!(binding.kind(&db), DefinitionKind::WithItem(_)));
+        }
+    }
+
+    #[test]
+    fn with_item_unpacked_definition() {
+        let TestCase { db, file } = test_case(
+            "
+with context() as (x, y):
+    pass
+",
+        );
+
+        let index = semantic_index(&db, file);
+        let global_table = index.symbol_table(FileScopeId::global());
+
+        assert_eq!(names(&global_table), vec!["context", "x", "y"]);
+
+        let use_def = index.use_def_map(FileScopeId::global());
+        for name in ["x", "y"] {
+            let binding = use_def
+                .first_public_binding(global_table.symbol_id_by_name(name).expect("symbol exists"))
+                .expect("Expected with item definition for {name}");
+            assert!(matches!(binding.kind(&db), DefinitionKind::WithItem(_)));
+        }
     }
 
     #[test]
@@ -562,14 +880,14 @@ def func():
         assert_eq!(names(&func2_table), vec!["y"]);
 
         let use_def = index.use_def_map(FileScopeId::global());
-        let [definition] = use_def.public_definitions(
-            global_table
-                .symbol_id_by_name("func")
-                .expect("symbol exists"),
-        ) else {
-            panic!("expected one definition");
-        };
-        assert!(matches!(definition.node(&db), DefinitionKind::Function(_)));
+        let binding = use_def
+            .first_public_binding(
+                global_table
+                    .symbol_id_by_name("func")
+                    .expect("symbol exists"),
+            )
+            .unwrap();
+        assert!(matches!(binding.kind(&db), DefinitionKind::Function(_)));
     }
 
     #[test]
@@ -637,7 +955,7 @@ class C[T]:
         assert!(
             ann_table
                 .symbol_by_name("T")
-                .is_some_and(|s| s.is_defined() && !s.is_used()),
+                .is_some_and(|s| s.is_bound() && !s.is_used()),
             "type parameters are defined by the scope that introduces them"
         );
 
@@ -669,10 +987,8 @@ class C[T]:
         };
         let x_use_id = x_use_expr_name.scoped_use_id(&db, scope);
         let use_def = use_def_map(&db, scope);
-        let [definition] = use_def.use_definitions(x_use_id) else {
-            panic!("expected one definition");
-        };
-        let DefinitionKind::Assignment(assignment) = definition.node(&db) else {
+        let binding = use_def.first_binding_at_use(x_use_id).unwrap();
+        let DefinitionKind::Assignment(assignment) = binding.kind(&db) else {
             panic!("should be an assignment definition")
         };
         let ast::Expr::NumberLiteral(ast::ExprNumberLiteral {
@@ -720,7 +1036,7 @@ class C[T]:
         }
 
         let TestCase { db, file } = test_case(
-            r#"
+            r"
 class Test:
     def foo():
         def bar():
@@ -729,7 +1045,7 @@ class Test:
         pass
 
 def x():
-    pass"#,
+    pass",
         );
 
         let index = semantic_index(&db, file);
@@ -762,4 +1078,136 @@ def x():
             vec!["bar", "foo", "Test", "<module>"]
         );
     }
+
+    #[test]
+    fn match_stmt() {
+        let TestCase { db, file } = test_case(
+            "
+match subject:
+    case a: ...
+    case [b, c, *d]: ...
+    case e as f: ...
+    case {'x': g, **h}: ...
+    case Foo(i, z=j): ...
+    case k | l: ...
+    case _: ...
+",
+        );
+
+        let global_scope_id = global_scope(&db, file);
+        let global_table = symbol_table(&db, global_scope_id);
+
+        assert!(global_table.symbol_by_name("Foo").unwrap().is_used());
+        assert_eq!(
+            names(&global_table),
+            vec!["subject", "a", "b", "c", "d", "e", "f", "g", "h", "Foo", "i", "j", "k", "l"]
+        );
+
+        let use_def = use_def_map(&db, global_scope_id);
+        for (name, expected_index) in [
+            ("a", 0),
+            ("b", 0),
+            ("c", 1),
+            ("d", 2),
+            ("e", 0),
+            ("f", 1),
+            ("g", 0),
+            ("h", 1),
+            ("i", 0),
+            ("j", 1),
+            ("k", 0),
+            ("l", 1),
+        ] {
+            let binding = use_def
+                .first_public_binding(global_table.symbol_id_by_name(name).expect("symbol exists"))
+                .expect("Expected with item definition for {name}");
+            if let DefinitionKind::MatchPattern(pattern) = binding.kind(&db) {
+                assert_eq!(pattern.index(), expected_index);
+            } else {
+                panic!("Expected match pattern definition for {name}");
+            }
+        }
+    }
+
+    #[test]
+    fn nested_match_case() {
+        let TestCase { db, file } = test_case(
+            "
+match 1:
+    case first:
+        match 2:
+            case second:
+                pass
+",
+        );
+
+        let global_scope_id = global_scope(&db, file);
+        let global_table = symbol_table(&db, global_scope_id);
+
+        assert_eq!(names(&global_table), vec!["first", "second"]);
+
+        let use_def = use_def_map(&db, global_scope_id);
+        for (name, expected_index) in [("first", 0), ("second", 0)] {
+            let binding = use_def
+                .first_public_binding(global_table.symbol_id_by_name(name).expect("symbol exists"))
+                .expect("Expected with item definition for {name}");
+            if let DefinitionKind::MatchPattern(pattern) = binding.kind(&db) {
+                assert_eq!(pattern.index(), expected_index);
+            } else {
+                panic!("Expected match pattern definition for {name}");
+            }
+        }
+    }
+
+    #[test]
+    fn for_loops_single_assignment() {
+        let TestCase { db, file } = test_case("for x in a: pass");
+        let scope = global_scope(&db, file);
+        let global_table = symbol_table(&db, scope);
+
+        assert_eq!(&names(&global_table), &["a", "x"]);
+
+        let use_def = use_def_map(&db, scope);
+        let binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("x").unwrap())
+            .unwrap();
+
+        assert!(matches!(binding.kind(&db), DefinitionKind::For(_)));
+    }
+
+    #[test]
+    fn for_loops_simple_unpacking() {
+        let TestCase { db, file } = test_case("for (x, y) in a: pass");
+        let scope = global_scope(&db, file);
+        let global_table = symbol_table(&db, scope);
+
+        assert_eq!(&names(&global_table), &["a", "x", "y"]);
+
+        let use_def = use_def_map(&db, scope);
+        let x_binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("x").unwrap())
+            .unwrap();
+        let y_binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("y").unwrap())
+            .unwrap();
+
+        assert!(matches!(x_binding.kind(&db), DefinitionKind::For(_)));
+        assert!(matches!(y_binding.kind(&db), DefinitionKind::For(_)));
+    }
+
+    #[test]
+    fn for_loops_complex_unpacking() {
+        let TestCase { db, file } = test_case("for [((a,) b), (c, d)] in e: pass");
+        let scope = global_scope(&db, file);
+        let global_table = symbol_table(&db, scope);
+
+        assert_eq!(&names(&global_table), &["e", "a", "b", "c", "d"]);
+
+        let use_def = use_def_map(&db, scope);
+        let binding = use_def
+            .first_public_binding(global_table.symbol_id_by_name("a").unwrap())
+            .unwrap();
+
+        assert!(matches!(binding.kind(&db), DefinitionKind::For(_)));
+    }
 }

crates/red_knot_python_semantic/src/semantic_index/ast_ids.rs

@@ -26,9 +26,9 @@ use crate::Db;
 /// ```
 #[derive(Debug)]
 pub(crate) struct AstIds {
-    /// Maps expressions to their expression id. Uses `NodeKey` because it avoids cloning [`Parsed`].
+    /// Maps expressions to their expression id.
     expressions_map: FxHashMap<ExpressionNodeKey, ScopedExpressionId>,
-    /// Maps expressions which "use" a symbol (that is, [`ExprName`]) to a use id.
+    /// Maps expressions which "use" a symbol (that is, [`ast::ExprName`]) to a use id.
     uses_map: FxHashMap<ExpressionNodeKey, ScopedUseId>,
 }
 

crates/red_knot_python_semantic/src/semantic_index/builder.rs

@@ -7,24 +7,31 @@ use ruff_db::parsed::ParsedModule;
 use ruff_index::IndexVec;
 use ruff_python_ast as ast;
 use ruff_python_ast::name::Name;
-use ruff_python_ast::visitor::{walk_expr, walk_stmt, Visitor};
+use ruff_python_ast::visitor::{walk_expr, walk_pattern, walk_stmt, Visitor};
+use ruff_python_ast::AnyParameterRef;
 
 use crate::ast_node_ref::AstNodeRef;
 use crate::semantic_index::ast_ids::node_key::ExpressionNodeKey;
 use crate::semantic_index::ast_ids::AstIdsBuilder;
 use crate::semantic_index::definition::{
-    AssignmentDefinitionNodeRef, Definition, DefinitionNodeKey, DefinitionNodeRef,
-    ImportFromDefinitionNodeRef,
+    AssignmentDefinitionNodeRef, ComprehensionDefinitionNodeRef, Definition, DefinitionNodeKey,
+    DefinitionNodeRef, ForStmtDefinitionNodeRef, ImportFromDefinitionNodeRef,
 };
 use crate::semantic_index::expression::Expression;
 use crate::semantic_index::symbol::{
-    FileScopeId, NodeWithScopeKey, NodeWithScopeRef, Scope, ScopeId, ScopedSymbolId, SymbolFlags,
+    FileScopeId, NodeWithScopeKey, NodeWithScopeRef, Scope, ScopeId, ScopedSymbolId,
     SymbolTableBuilder,
 };
 use crate::semantic_index::use_def::{FlowSnapshot, UseDefMapBuilder};
 use crate::semantic_index::SemanticIndex;
 use crate::Db;
 
+use super::constraint::{Constraint, PatternConstraint};
+use super::definition::{
+    DefinitionCategory, ExceptHandlerDefinitionNodeRef, MatchPatternDefinitionNodeRef,
+    WithItemDefinitionNodeRef,
+};
+
 pub(super) struct SemanticIndexBuilder<'db> {
     // Builder state
     db: &'db dyn Db,
@@ -33,9 +40,14 @@ pub(super) struct SemanticIndexBuilder<'db> {
     scope_stack: Vec<FileScopeId>,
     /// The assignment we're currently visiting.
     current_assignment: Option<CurrentAssignment<'db>>,
+    /// The match case we're currently visiting.
+    current_match_case: Option<CurrentMatchCase<'db>>,
     /// Flow states at each `break` in the current loop.
     loop_break_states: Vec<FlowSnapshot>,
 
+    /// Flags about the file's global scope
+    has_future_annotations: bool,
+
     // Semantic Index fields
     scopes: IndexVec<FileScopeId, Scope>,
     scope_ids_by_scope: IndexVec<FileScopeId, ScopeId<'db>>,
@@ -56,8 +68,11 @@ impl<'db> SemanticIndexBuilder<'db> {
             module: parsed,
             scope_stack: Vec::new(),
             current_assignment: None,
+            current_match_case: None,
             loop_break_states: vec![],
 
+            has_future_annotations: false,
+
             scopes: IndexVec::new(),
             symbol_tables: IndexVec::new(),
             ast_ids: IndexVec::new(),
@@ -155,49 +170,102 @@ impl<'db> SemanticIndexBuilder<'db> {
         self.current_use_def_map_mut().restore(state);
     }
 
-    fn flow_merge(&mut self, state: &FlowSnapshot) {
+    fn flow_merge(&mut self, state: FlowSnapshot) {
         self.current_use_def_map_mut().merge(state);
     }
 
-    fn add_or_update_symbol(&mut self, name: Name, flags: SymbolFlags) -> ScopedSymbolId {
-        let symbol_table = self.current_symbol_table();
-        let (symbol_id, added) = symbol_table.add_or_update_symbol(name, flags);
+    fn add_symbol(&mut self, name: Name) -> ScopedSymbolId {
+        let (symbol_id, added) = self.current_symbol_table().add_symbol(name);
         if added {
-            let use_def_map = self.current_use_def_map_mut();
-            use_def_map.add_symbol(symbol_id);
+            self.current_use_def_map_mut().add_symbol(symbol_id);
         }
         symbol_id
     }
 
+    fn mark_symbol_bound(&mut self, id: ScopedSymbolId) {
+        self.current_symbol_table().mark_symbol_bound(id);
+    }
+
+    fn mark_symbol_used(&mut self, id: ScopedSymbolId) {
+        self.current_symbol_table().mark_symbol_used(id);
+    }
+
     fn add_definition<'a>(
         &mut self,
         symbol: ScopedSymbolId,
         definition_node: impl Into<DefinitionNodeRef<'a>>,
     ) -> Definition<'db> {
-        let definition_node = definition_node.into();
+        let definition_node: DefinitionNodeRef<'_> = definition_node.into();
+        #[allow(unsafe_code)]
+        // SAFETY: `definition_node` is guaranteed to be a child of `self.module`
+        let kind = unsafe { definition_node.into_owned(self.module.clone()) };
+        let category = kind.category();
         let definition = Definition::new(
             self.db,
             self.file,
             self.current_scope(),
             symbol,
-            #[allow(unsafe_code)]
-            unsafe {
-                definition_node.into_owned(self.module.clone())
-            },
+            kind,
             countme::Count::default(),
         );
 
-        self.definitions_by_node
+        let existing_definition = self
+            .definitions_by_node
             .insert(definition_node.key(), definition);
-        self.current_use_def_map_mut()
-            .record_definition(symbol, definition);
+        debug_assert_eq!(existing_definition, None);
+
+        if category.is_binding() {
+            self.mark_symbol_bound(symbol);
+        }
+
+        let use_def = self.current_use_def_map_mut();
+        match category {
+            DefinitionCategory::DeclarationAndBinding => {
+                use_def.record_declaration_and_binding(symbol, definition);
+            }
+            DefinitionCategory::Declaration => use_def.record_declaration(symbol, definition),
+            DefinitionCategory::Binding => use_def.record_binding(symbol, definition),
+        }
 
         definition
     }
 
+    fn add_expression_constraint(&mut self, constraint_node: &ast::Expr) -> Expression<'db> {
+        let expression = self.add_standalone_expression(constraint_node);
+        self.current_use_def_map_mut()
+            .record_constraint(Constraint::Expression(expression));
+
+        expression
+    }
+
+    fn add_pattern_constraint(
+        &mut self,
+        subject: &ast::Expr,
+        pattern: &ast::Pattern,
+    ) -> PatternConstraint<'db> {
+        #[allow(unsafe_code)]
+        let (subject, pattern) = unsafe {
+            (
+                AstNodeRef::new(self.module.clone(), subject),
+                AstNodeRef::new(self.module.clone(), pattern),
+            )
+        };
+        let pattern_constraint = PatternConstraint::new(
+            self.db,
+            self.file,
+            self.current_scope(),
+            subject,
+            pattern,
+            countme::Count::default(),
+        );
+        self.current_use_def_map_mut()
+            .record_constraint(Constraint::Pattern(pattern_constraint));
+        pattern_constraint
+    }
+
     /// Record an expression that needs to be a Salsa ingredient, because we need to infer its type
     /// standalone (type narrowing tests, RHS of an assignment.)
-    fn add_standalone_expression(&mut self, expression_node: &ast::Expr) {
+    fn add_standalone_expression(&mut self, expression_node: &ast::Expr) -> Expression<'db> {
         let expression = Expression::new(
             self.db,
             self.file,
@@ -210,6 +278,7 @@ impl<'db> SemanticIndexBuilder<'db> {
         );
         self.expressions_by_node
             .insert(expression_node.into(), expression);
+        expression
     }
 
     fn with_type_params(
@@ -238,10 +307,13 @@ impl<'db> SemanticIndexBuilder<'db> {
                         ..
                     }) => (name, &None, default),
                 };
-                // TODO create Definition for typevars
-                self.add_or_update_symbol(name.id.clone(), SymbolFlags::IS_DEFINED);
-                if let Some(bound) = bound {
-                    self.visit_expr(bound);
+                let symbol = self.add_symbol(name.id.clone());
+                // TODO create Definition for PEP 695 typevars
+                // note that the "bound" on the typevar is a totally different thing than whether
+                // or not a name is "bound" by a typevar declaration; the latter is always true.
+                self.mark_symbol_bound(symbol);
+                if let Some(bounds) = bound {
+                    self.visit_expr(bounds);
                 }
                 if let Some(default) = default {
                     self.visit_expr(default);
@@ -258,6 +330,83 @@ impl<'db> SemanticIndexBuilder<'db> {
         nested_scope
     }
 
+    /// This method does several things:
+    /// - It pushes a new scope onto the stack for visiting
+    ///   a list/dict/set comprehension or generator expression
+    /// - Inside that scope, it visits a list of [`Comprehension`] nodes,
+    ///   assumed to be the "generators" that compose a comprehension
+    ///   (that is, the `for x in y` and `for y in z` parts of `x for x in y for y in z`).
+    /// - Inside that scope, it also calls a closure for visiting the outer `elt`
+    ///   of a list/dict/set comprehension or generator expression
+    /// - It then pops the new scope off the stack
+    ///
+    /// [`Comprehension`]: ast::Comprehension
+    fn with_generators_scope(
+        &mut self,
+        scope: NodeWithScopeRef,
+        generators: &'db [ast::Comprehension],
+        visit_outer_elt: impl FnOnce(&mut Self),
+    ) {
+        let mut generators_iter = generators.iter();
+
+        let Some(generator) = generators_iter.next() else {
+            unreachable!("Expression must contain at least one generator");
+        };
+
+        // The `iter` of the first generator is evaluated in the outer scope, while all subsequent
+        // nodes are evaluated in the inner scope.
+        self.add_standalone_expression(&generator.iter);
+        self.visit_expr(&generator.iter);
+        self.push_scope(scope);
+
+        self.current_assignment = Some(CurrentAssignment::Comprehension {
+            node: generator,
+            first: true,
+        });
+        self.visit_expr(&generator.target);
+        self.current_assignment = None;
+
+        for expr in &generator.ifs {
+            self.visit_expr(expr);
+        }
+
+        for generator in generators_iter {
+            self.add_standalone_expression(&generator.iter);
+            self.visit_expr(&generator.iter);
+
+            self.current_assignment = Some(CurrentAssignment::Comprehension {
+                node: generator,
+                first: false,
+            });
+            self.visit_expr(&generator.target);
+            self.current_assignment = None;
+
+            for expr in &generator.ifs {
+                self.visit_expr(expr);
+            }
+        }
+
+        visit_outer_elt(self);
+        self.pop_scope();
+    }
+
+    fn declare_parameter(&mut self, parameter: AnyParameterRef) {
+        let symbol = self.add_symbol(parameter.name().id().clone());
+
+        let definition = self.add_definition(symbol, parameter);
+
+        if let AnyParameterRef::NonVariadic(with_default) = parameter {
+            // Insert a mapping from the parameter to the same definition.
+            // This ensures that calling `HasTy::ty` on the inner parameter returns
+            // a valid type (and doesn't panic)
+            let existing_definition = self.definitions_by_node.insert(
+                DefinitionNodeRef::from(AnyParameterRef::Variadic(&with_default.parameter)).key(),
+                definition,
+            );
+            debug_assert_eq!(existing_definition, None);
+        }
+    }
+
     pub(super) fn build(mut self) -> SemanticIndex<'db> {
         let module = self.module;
         self.visit_body(module.suite());
@@ -306,6 +455,7 @@ impl<'db> SemanticIndexBuilder<'db> {
             scopes_by_expression: self.scopes_by_expression,
             scopes_by_node: self.scopes_by_node,
             use_def_maps,
+            has_future_annotations: self.has_future_annotations,
         }
     }
 }
@@ -321,10 +471,6 @@ where
                     self.visit_decorator(decorator);
                 }
 
-                let symbol = self
-                    .add_or_update_symbol(function_def.name.id.clone(), SymbolFlags::IS_DEFINED);
-                self.add_definition(symbol, function_def);
-
                 self.with_type_params(
                     NodeWithScopeRef::FunctionTypeParameters(function_def),
                     function_def.type_params.as_deref(),
@@ -335,18 +481,37 @@ where
                         }
 
                         builder.push_scope(NodeWithScopeRef::Function(function_def));
+
+                        // Add symbols and definitions for the parameters to the function scope.
+                        for parameter in &*function_def.parameters {
+                            builder.declare_parameter(parameter);
+                        }
+
                         builder.visit_body(&function_def.body);
                         builder.pop_scope()
                     },
                 );
+                // The default value of the parameters needs to be evaluated in the
+                // enclosing scope.
+                for default in function_def
+                    .parameters
+                    .iter_non_variadic_params()
+                    .filter_map(|param| param.default.as_deref())
+                {
+                    self.visit_expr(default);
+                }
+                // The symbol for the function name itself has to be evaluated
+                // at the end to match the runtime evaluation of parameter defaults
+                // and return-type annotations.
+                let symbol = self.add_symbol(function_def.name.id.clone());
+                self.add_definition(symbol, function_def);
             }
             ast::Stmt::ClassDef(class) => {
                 for decorator in &class.decorator_list {
                     self.visit_decorator(decorator);
                 }
 
-                let symbol =
-                    self.add_or_update_symbol(class.name.id.clone(), SymbolFlags::IS_DEFINED);
+                let symbol = self.add_symbol(class.name.id.clone());
                 self.add_definition(symbol, class);
 
                 self.with_type_params(
@@ -372,7 +537,7 @@ where
                         Name::new(alias.name.id.split('.').next().unwrap())
                     };
 
-                    let symbol = self.add_or_update_symbol(symbol_name, SymbolFlags::IS_DEFINED);
+                    let symbol = self.add_symbol(symbol_name);
                     self.add_definition(symbol, alias);
                 }
             }
@@ -384,8 +549,16 @@ where
                         &alias.name.id
                     };
 
-                    let symbol =
-                        self.add_or_update_symbol(symbol_name.clone(), SymbolFlags::IS_DEFINED);
+                    // Look for imports `from __future__ import annotations`, ignore `as ...`
+                    // We intentionally don't enforce the rules about location of `__future__`
+                    // imports here, we assume the user's intent was to apply the `__future__`
+                    // import, so we still check using it (and will also emit a diagnostic about a
+                    // miss-placed `__future__` import.)
+                    self.has_future_annotations |= alias.name.id == "annotations"
+                        && node.module.as_deref() == Some("__future__");
+
+                    let symbol = self.add_symbol(symbol_name.clone());
+
                     self.add_definition(symbol, ImportFromDefinitionNodeRef { node, alias_index });
                 }
             }
@@ -401,7 +574,6 @@ where
             }
             ast::Stmt::AnnAssign(node) => {
                 debug_assert!(self.current_assignment.is_none());
-                // TODO deferred annotation visiting
                 self.visit_expr(&node.annotation);
                 if let Some(value) = &node.value {
                     self.visit_expr(value);
@@ -410,9 +582,24 @@ where
                 self.visit_expr(&node.target);
                 self.current_assignment = None;
             }
+            ast::Stmt::AugAssign(
+                aug_assign @ ast::StmtAugAssign {
+                    range: _,
+                    target,
+                    op: _,
+                    value,
+                },
+            ) => {
+                debug_assert!(self.current_assignment.is_none());
+                self.visit_expr(value);
+                self.current_assignment = Some(aug_assign.into());
+                self.visit_expr(target);
+                self.current_assignment = None;
+            }
             ast::Stmt::If(node) => {
                 self.visit_expr(&node.test);
                 let pre_if = self.flow_snapshot();
+                self.add_expression_constraint(&node.test);
                 self.visit_body(&node.body);
                 let mut post_clauses: Vec<FlowSnapshot> = vec![];
                 for clause in &node.elif_else_clauses {
@@ -425,7 +612,7 @@ where
                     self.visit_elif_else_clause(clause);
                 }
                 for post_clause_state in post_clauses {
-                    self.flow_merge(&post_clause_state);
+                    self.flow_merge(post_clause_state);
                 }
                 let has_else = node
                     .elif_else_clauses
@@ -434,17 +621,26 @@ where
                 if !has_else {
                     // if there's no else clause, then it's possible we took none of the branches,
                     // and the pre_if state can reach here
-                    self.flow_merge(&pre_if);
+                    self.flow_merge(pre_if);
                 }
             }
-            ast::Stmt::While(node) => {
-                self.visit_expr(&node.test);
+            ast::Stmt::While(ast::StmtWhile {
+                test,
+                body,
+                orelse,
+                range: _,
+            }) => {
+                self.visit_expr(test);
 
                 let pre_loop = self.flow_snapshot();
 
                 // Save aside any break states from an outer loop
                 let saved_break_states = std::mem::take(&mut self.loop_break_states);
-                self.visit_body(&node.body);
+
+                // TODO: definitions created inside the body should be fully visible
+                // to other statements/expressions inside the body --Alex/Carl
+                self.visit_body(body);
+
                 // Get the break states from the body of this loop, and restore the saved outer
                 // ones.
                 let break_states =
@@ -452,18 +648,147 @@ where
 
                 // We may execute the `else` clause without ever executing the body, so merge in
                 // the pre-loop state before visiting `else`.
-                self.flow_merge(&pre_loop);
-                self.visit_body(&node.orelse);
+                self.flow_merge(pre_loop);
+                self.visit_body(orelse);
 
                 // Breaking out of a while loop bypasses the `else` clause, so merge in the break
                 // states after visiting `else`.
                 for break_state in break_states {
-                    self.flow_merge(&break_state);
+                    self.flow_merge(break_state);
                 }
             }
+            ast::Stmt::With(ast::StmtWith { items, body, .. }) => {
+                for item in items {
+                    self.visit_expr(&item.context_expr);
+                    if let Some(optional_vars) = item.optional_vars.as_deref() {
+                        self.add_standalone_expression(&item.context_expr);
+                        self.current_assignment = Some(item.into());
+                        self.visit_expr(optional_vars);
+                        self.current_assignment = None;
+                    }
+                }
+                self.visit_body(body);
+            }
             ast::Stmt::Break(_) => {
                 self.loop_break_states.push(self.flow_snapshot());
             }
+
+            ast::Stmt::For(
+                for_stmt @ ast::StmtFor {
+                    range: _,
+                    is_async: _,
+                    target,
+                    iter,
+                    body,
+                    orelse,
+                },
+            ) => {
+                self.add_standalone_expression(iter);
+                self.visit_expr(iter);
+
+                let pre_loop = self.flow_snapshot();
+                let saved_break_states = std::mem::take(&mut self.loop_break_states);
+
+                debug_assert!(self.current_assignment.is_none());
+                self.current_assignment = Some(for_stmt.into());
+                self.visit_expr(target);
+                self.current_assignment = None;
+
+                // TODO: Definitions created by loop variables
+                // (and definitions created inside the body)
+                // are fully visible to other statements/expressions inside the body --Alex/Carl
+                self.visit_body(body);
+
+                let break_states =
+                    std::mem::replace(&mut self.loop_break_states, saved_break_states);
+
+                // We may execute the `else` clause without ever executing the body, so merge in
+                // the pre-loop state before visiting `else`.
+                self.flow_merge(pre_loop);
+                self.visit_body(orelse);
+
+                // Breaking out of a `for` loop bypasses the `else` clause, so merge in the break
+                // states after visiting `else`.
+                for break_state in break_states {
+                    self.flow_merge(break_state);
+                }
+            }
+            ast::Stmt::Match(ast::StmtMatch {
+                subject,
+                cases,
+                range: _,
+            }) => {
+                self.add_standalone_expression(subject);
+                self.visit_expr(subject);
+
+                let after_subject = self.flow_snapshot();
+                let Some((first, remaining)) = cases.split_first() else {
+                    return;
+                };
+                self.add_pattern_constraint(subject, &first.pattern);
+                self.visit_match_case(first);
+
+                let mut post_case_snapshots = vec![];
+                for case in remaining {
+                    post_case_snapshots.push(self.flow_snapshot());
+                    self.flow_restore(after_subject.clone());
+                    self.add_pattern_constraint(subject, &case.pattern);
+                    self.visit_match_case(case);
+                }
+                for post_clause_state in post_case_snapshots {
+                    self.flow_merge(post_clause_state);
+                }
+                if !cases
+                    .last()
+                    .is_some_and(|case| case.guard.is_none() && case.pattern.is_wildcard())
+                {
+                    self.flow_merge(after_subject);
+                }
+            }
+            ast::Stmt::Try(ast::StmtTry {
+                body,
+                handlers,
+                orelse,
+                finalbody,
+                is_star,
+                range: _,
+            }) => {
+                self.visit_body(body);
+
+                for except_handler in handlers {
+                    let ast::ExceptHandler::ExceptHandler(except_handler) = except_handler;
+                    let ast::ExceptHandlerExceptHandler {
+                        name: symbol_name,
+                        type_: handled_exceptions,
+                        body: handler_body,
+                        range: _,
+                    } = except_handler;
+
+                    if let Some(handled_exceptions) = handled_exceptions {
+                        self.visit_expr(handled_exceptions);
+                    }
+
+                    // If `handled_exceptions` above was `None`, it's something like `except as e:`,
+                    // which is invalid syntax. However, it's still pretty obvious here that the user
+                    // *wanted* `e` to be bound, so we should still create a definition here nonetheless.
+                    if let Some(symbol_name) = symbol_name {
+                        let symbol = self.add_symbol(symbol_name.id.clone());
+
+                        self.add_definition(
+                            symbol,
+                            DefinitionNodeRef::ExceptHandler(ExceptHandlerDefinitionNodeRef {
+                                handler: except_handler,
+                                is_star: *is_star,
+                            }),
+                        );
+                    }
+
+                    self.visit_body(handler_body);
+                }
+
+                self.visit_body(orelse);
+                self.visit_body(finalbody);
+            }
             _ => {
                 walk_stmt(self, stmt);
             }
@@ -476,16 +801,19 @@ where
         self.current_ast_ids().record_expression(expr);
 
         match expr {
-            ast::Expr::Name(name_node) => {
-                let ast::ExprName { id, ctx, .. } = name_node;
-                let flags = match ctx {
-                    ast::ExprContext::Load => SymbolFlags::IS_USED,
-                    ast::ExprContext::Store => SymbolFlags::IS_DEFINED,
-                    ast::ExprContext::Del => SymbolFlags::IS_DEFINED,
-                    ast::ExprContext::Invalid => SymbolFlags::empty(),
+            ast::Expr::Name(name_node @ ast::ExprName { id, ctx, .. }) => {
+                let (is_use, is_definition) = match (ctx, self.current_assignment) {
+                    (ast::ExprContext::Store, Some(CurrentAssignment::AugAssign(_))) => {
+                        // For augmented assignment, the target expression is also used.
+                        (true, true)
+                    }
+                    (ast::ExprContext::Load, _) => (true, false),
+                    (ast::ExprContext::Store, _) => (false, true),
+                    (ast::ExprContext::Del, _) => (false, true),
+                    (ast::ExprContext::Invalid, _) => (false, false),
                 };
-                let symbol = self.add_or_update_symbol(id.clone(), flags);
-                if flags.contains(SymbolFlags::IS_DEFINED) {
+                let symbol = self.add_symbol(id.clone());
+                if is_definition {
                     match self.current_assignment {
                         Some(CurrentAssignment::Assign(assignment)) => {
                             self.add_definition(
@@ -499,14 +827,51 @@ where
                         Some(CurrentAssignment::AnnAssign(ann_assign)) => {
                             self.add_definition(symbol, ann_assign);
                         }
+                        Some(CurrentAssignment::AugAssign(aug_assign)) => {
+                            self.add_definition(symbol, aug_assign);
+                        }
+                        Some(CurrentAssignment::For(node)) => {
+                            self.add_definition(
+                                symbol,
+                                ForStmtDefinitionNodeRef {
+                                    iterable: &node.iter,
+                                    target: name_node,
+                                    is_async: node.is_async,
+                                },
+                            );
+                        }
                         Some(CurrentAssignment::Named(named)) => {
+                            // TODO(dhruvmanila): If the current scope is a comprehension, then the
+                            // named expression is implicitly nonlocal. This is yet to be
+                            // implemented.
                             self.add_definition(symbol, named);
                         }
+                        Some(CurrentAssignment::Comprehension { node, first }) => {
+                            self.add_definition(
+                                symbol,
+                                ComprehensionDefinitionNodeRef {
+                                    iterable: &node.iter,
+                                    target: name_node,
+                                    first,
+                                    is_async: node.is_async,
+                                },
+                            );
+                        }
+                        Some(CurrentAssignment::WithItem(with_item)) => {
+                            self.add_definition(
+                                symbol,
+                                WithItemDefinitionNodeRef {
+                                    node: with_item,
+                                    target: name_node,
+                                },
+                            );
+                        }
                         None => {}
                     }
                 }
 
-                if flags.contains(SymbolFlags::IS_USED) {
+                if is_use {
+                    self.mark_symbol_used(symbol);
                     let use_id = self.current_ast_ids().record_use(expr);
                     self.current_use_def_map_mut().record_use(symbol, use_id);
                 }
@@ -515,17 +880,33 @@ where
             }
             ast::Expr::Named(node) => {
                 debug_assert!(self.current_assignment.is_none());
-                self.current_assignment = Some(node.into());
                 // TODO walrus in comprehensions is implicitly nonlocal
+                self.visit_expr(&node.value);
+                self.current_assignment = Some(node.into());
                 self.visit_expr(&node.target);
                 self.current_assignment = None;
-                self.visit_expr(&node.value);
             }
             ast::Expr::Lambda(lambda) => {
                 if let Some(parameters) = &lambda.parameters {
+                    // The default value of the parameters needs to be evaluated in the
+                    // enclosing scope.
+                    for default in parameters
+                        .iter_non_variadic_params()
+                        .filter_map(|param| param.default.as_deref())
+                    {
+                        self.visit_expr(default);
+                    }
                     self.visit_parameters(parameters);
                 }
                 self.push_scope(NodeWithScopeRef::Lambda(lambda));
+
+                // Add symbols and definitions for the parameters to the lambda scope.
+                if let Some(parameters) = &lambda.parameters {
+                    for parameter in &**parameters {
+                        self.declare_parameter(parameter);
+                    }
+                }
+
                 self.visit_expr(lambda.body.as_ref());
                 self.pop_scope();
             }
@@ -541,20 +922,139 @@ where
                 let post_body = self.flow_snapshot();
                 self.flow_restore(pre_if);
                 self.visit_expr(orelse);
-                self.flow_merge(&post_body);
+                self.flow_merge(post_body);
+            }
+            ast::Expr::ListComp(
+                list_comprehension @ ast::ExprListComp {
+                    elt, generators, ..
+                },
+            ) => {
+                self.with_generators_scope(
+                    NodeWithScopeRef::ListComprehension(list_comprehension),
+                    generators,
+                    |builder| builder.visit_expr(elt),
+                );
+            }
+            ast::Expr::SetComp(
+                set_comprehension @ ast::ExprSetComp {
+                    elt, generators, ..
+                },
+            ) => {
+                self.with_generators_scope(
+                    NodeWithScopeRef::SetComprehension(set_comprehension),
+                    generators,
+                    |builder| builder.visit_expr(elt),
+                );
+            }
+            ast::Expr::Generator(
+                generator @ ast::ExprGenerator {
+                    elt, generators, ..
+                },
+            ) => {
+                self.with_generators_scope(
+                    NodeWithScopeRef::GeneratorExpression(generator),
+                    generators,
+                    |builder| builder.visit_expr(elt),
+                );
+            }
+            ast::Expr::DictComp(
+                dict_comprehension @ ast::ExprDictComp {
+                    key,
+                    value,
+                    generators,
+                    ..
+                },
+            ) => {
+                self.with_generators_scope(
+                    NodeWithScopeRef::DictComprehension(dict_comprehension),
+                    generators,
+                    |builder| {
+                        builder.visit_expr(key);
+                        builder.visit_expr(value);
+                    },
+                );
             }
             _ => {
                 walk_expr(self, expr);
             }
         }
     }
+
+    fn visit_parameters(&mut self, parameters: &'ast ast::Parameters) {
+        // Intentionally avoid walking default expressions, as we handle them in the enclosing
+        // scope.
+        for parameter in parameters.iter().map(ast::AnyParameterRef::as_parameter) {
+            self.visit_parameter(parameter);
+        }
+    }
+
+    fn visit_match_case(&mut self, match_case: &'ast ast::MatchCase) {
+        debug_assert!(self.current_match_case.is_none());
+        self.current_match_case = Some(CurrentMatchCase::new(&match_case.pattern));
+        self.visit_pattern(&match_case.pattern);
+        self.current_match_case = None;
+
+        if let Some(expr) = &match_case.guard {
+            self.visit_expr(expr);
+        }
+        self.visit_body(&match_case.body);
+    }
+
+    fn visit_pattern(&mut self, pattern: &'ast ast::Pattern) {
+        if let ast::Pattern::MatchStar(ast::PatternMatchStar {
+            name: Some(name),
+            range: _,
+        }) = pattern
+        {
+            let symbol = self.add_symbol(name.id().clone());
+            let state = self.current_match_case.as_ref().unwrap();
+            self.add_definition(
+                symbol,
+                MatchPatternDefinitionNodeRef {
+                    pattern: state.pattern,
+                    identifier: name,
+                    index: state.index,
+                },
+            );
+        }
+
+        walk_pattern(self, pattern);
+
+        if let ast::Pattern::MatchAs(ast::PatternMatchAs {
+            name: Some(name), ..
+        })
+        | ast::Pattern::MatchMapping(ast::PatternMatchMapping {
+            rest: Some(name), ..
+        }) = pattern
+        {
+            let symbol = self.add_symbol(name.id().clone());
+            let state = self.current_match_case.as_ref().unwrap();
+            self.add_definition(
+                symbol,
+                MatchPatternDefinitionNodeRef {
+                    pattern: state.pattern,
+                    identifier: name,
+                    index: state.index,
+                },
+            );
+        }
+
+        self.current_match_case.as_mut().unwrap().index += 1;
+    }
 }
 
 #[derive(Copy, Clone, Debug)]
 enum CurrentAssignment<'a> {
     Assign(&'a ast::StmtAssign),
     AnnAssign(&'a ast::StmtAnnAssign),
+    AugAssign(&'a ast::StmtAugAssign),
+    For(&'a ast::StmtFor),
     Named(&'a ast::ExprNamed),
+    Comprehension {
+        node: &'a ast::Comprehension,
+        first: bool,
+    },
+    WithItem(&'a ast::WithItem),
 }
 
 impl<'a> From<&'a ast::StmtAssign> for CurrentAssignment<'a> {
@@ -569,8 +1069,50 @@ impl<'a> From<&'a ast::StmtAnnAssign> for CurrentAssignment<'a> {
     }
 }
 
+impl<'a> From<&'a ast::StmtAugAssign> for CurrentAssignment<'a> {
+    fn from(value: &'a ast::StmtAugAssign) -> Self {
+        Self::AugAssign(value)
+    }
+}
+
+impl<'a> From<&'a ast::StmtFor> for CurrentAssignment<'a> {
+    fn from(value: &'a ast::StmtFor) -> Self {
+        Self::For(value)
+    }
+}
+
 impl<'a> From<&'a ast::ExprNamed> for CurrentAssignment<'a> {
     fn from(value: &'a ast::ExprNamed) -> Self {
         Self::Named(value)
     }
 }
+
+impl<'a> From<&'a ast::WithItem> for CurrentAssignment<'a> {
+    fn from(value: &'a ast::WithItem) -> Self {
+        Self::WithItem(value)
+    }
+}
+
+struct CurrentMatchCase<'a> {
+    /// The pattern that's part of the current match case.
+    pattern: &'a ast::Pattern,
+
+    /// The index of the sub-pattern that's being currently visited within the pattern.
+    ///
+    /// For example:
+    /// ```py
+    /// match subject:
+    ///     case a as b: ...
+    ///     case [a, b]: ...
+    ///     case a | b: ...
+    /// ```
+    ///
+    /// In all of the above cases, the index would be 0 for `a` and 1 for `b`.
+    index: u32,
+}
+
+impl<'a> CurrentMatchCase<'a> {
+    fn new(pattern: &'a ast::Pattern) -> Self {
+        Self { pattern, index: 0 }
+    }
+}

crates/red_knot_python_semantic/src/semantic_index/constraint.rs

@@ -0,0 +1,39 @@
+use ruff_db::files::File;
+use ruff_python_ast as ast;
+
+use crate::ast_node_ref::AstNodeRef;
+use crate::db::Db;
+use crate::semantic_index::expression::Expression;
+use crate::semantic_index::symbol::{FileScopeId, ScopeId};
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(crate) enum Constraint<'db> {
+    Expression(Expression<'db>),
+    Pattern(PatternConstraint<'db>),
+}
+
+#[salsa::tracked]
+pub(crate) struct PatternConstraint<'db> {
+    #[id]
+    pub(crate) file: File,
+
+    #[id]
+    pub(crate) file_scope: FileScopeId,
+
+    #[no_eq]
+    #[return_ref]
+    pub(crate) subject: AstNodeRef<ast::Expr>,
+
+    #[no_eq]
+    #[return_ref]
+    pub(crate) pattern: AstNodeRef<ast::Pattern>,
+
+    #[no_eq]
+    count: countme::Count<PatternConstraint<'static>>,
+}
+
+impl<'db> PatternConstraint<'db> {
+    pub(crate) fn scope(self, db: &'db dyn Db) -> ScopeId<'db> {
+        self.file_scope(db).to_scope_id(db, self.file(db))
+    }
+}

crates/red_knot_python_semantic/src/semantic_index/definition.rs

@@ -23,7 +23,7 @@ pub struct Definition<'db> {
 
     #[no_eq]
     #[return_ref]
-    pub(crate) node: DefinitionKind,
+    pub(crate) kind: DefinitionKind,
 
     #[no_eq]
     count: countme::Count<Definition<'static>>,
@@ -33,17 +33,36 @@ impl<'db> Definition<'db> {
     pub(crate) fn scope(self, db: &'db dyn Db) -> ScopeId<'db> {
         self.file_scope(db).to_scope_id(db, self.file(db))
     }
+
+    pub(crate) fn category(self, db: &'db dyn Db) -> DefinitionCategory {
+        self.kind(db).category()
+    }
+
+    pub(crate) fn is_declaration(self, db: &'db dyn Db) -> bool {
+        self.kind(db).category().is_declaration()
+    }
+
+    pub(crate) fn is_binding(self, db: &'db dyn Db) -> bool {
+        self.kind(db).category().is_binding()
+    }
 }
 
 #[derive(Copy, Clone, Debug)]
 pub(crate) enum DefinitionNodeRef<'a> {
     Import(&'a ast::Alias),
     ImportFrom(ImportFromDefinitionNodeRef<'a>),
+    For(ForStmtDefinitionNodeRef<'a>),
     Function(&'a ast::StmtFunctionDef),
     Class(&'a ast::StmtClassDef),
     NamedExpression(&'a ast::ExprNamed),
     Assignment(AssignmentDefinitionNodeRef<'a>),
     AnnotatedAssignment(&'a ast::StmtAnnAssign),
+    AugmentedAssignment(&'a ast::StmtAugAssign),
+    Comprehension(ComprehensionDefinitionNodeRef<'a>),
+    Parameter(ast::AnyParameterRef<'a>),
+    WithItem(WithItemDefinitionNodeRef<'a>),
+    MatchPattern(MatchPatternDefinitionNodeRef<'a>),
+    ExceptHandler(ExceptHandlerDefinitionNodeRef<'a>),
 }
 
 impl<'a> From<&'a ast::StmtFunctionDef> for DefinitionNodeRef<'a> {
@@ -70,6 +89,12 @@ impl<'a> From<&'a ast::StmtAnnAssign> for DefinitionNodeRef<'a> {
     }
 }
 
+impl<'a> From<&'a ast::StmtAugAssign> for DefinitionNodeRef<'a> {
+    fn from(node: &'a ast::StmtAugAssign) -> Self {
+        Self::AugmentedAssignment(node)
+    }
+}
+
 impl<'a> From<&'a ast::Alias> for DefinitionNodeRef<'a> {
     fn from(node_ref: &'a ast::Alias) -> Self {
         Self::Import(node_ref)
@@ -82,12 +107,42 @@ impl<'a> From<ImportFromDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
     }
 }
 
+impl<'a> From<ForStmtDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
+    fn from(value: ForStmtDefinitionNodeRef<'a>) -> Self {
+        Self::For(value)
+    }
+}
+
 impl<'a> From<AssignmentDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
     fn from(node_ref: AssignmentDefinitionNodeRef<'a>) -> Self {
         Self::Assignment(node_ref)
     }
 }
 
+impl<'a> From<WithItemDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
+    fn from(node_ref: WithItemDefinitionNodeRef<'a>) -> Self {
+        Self::WithItem(node_ref)
+    }
+}
+
+impl<'a> From<ComprehensionDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
+    fn from(node: ComprehensionDefinitionNodeRef<'a>) -> Self {
+        Self::Comprehension(node)
+    }
+}
+
+impl<'a> From<ast::AnyParameterRef<'a>> for DefinitionNodeRef<'a> {
+    fn from(node: ast::AnyParameterRef<'a>) -> Self {
+        Self::Parameter(node)
+    }
+}
+
+impl<'a> From<MatchPatternDefinitionNodeRef<'a>> for DefinitionNodeRef<'a> {
+    fn from(node: MatchPatternDefinitionNodeRef<'a>) -> Self {
+        Self::MatchPattern(node)
+    }
+}
+
 #[derive(Copy, Clone, Debug)]
 pub(crate) struct ImportFromDefinitionNodeRef<'a> {
     pub(crate) node: &'a ast::StmtImportFrom,
@@ -100,6 +155,44 @@ pub(crate) struct AssignmentDefinitionNodeRef<'a> {
     pub(crate) target: &'a ast::ExprName,
 }
 
+#[derive(Copy, Clone, Debug)]
+pub(crate) struct WithItemDefinitionNodeRef<'a> {
+    pub(crate) node: &'a ast::WithItem,
+    pub(crate) target: &'a ast::ExprName,
+}
+
+#[derive(Copy, Clone, Debug)]
+pub(crate) struct ForStmtDefinitionNodeRef<'a> {
+    pub(crate) iterable: &'a ast::Expr,
+    pub(crate) target: &'a ast::ExprName,
+    pub(crate) is_async: bool,
+}
+
+#[derive(Copy, Clone, Debug)]
+pub(crate) struct ExceptHandlerDefinitionNodeRef<'a> {
+    pub(crate) handler: &'a ast::ExceptHandlerExceptHandler,
+    pub(crate) is_star: bool,
+}
+
+#[derive(Copy, Clone, Debug)]
+pub(crate) struct ComprehensionDefinitionNodeRef<'a> {
+    pub(crate) iterable: &'a ast::Expr,
+    pub(crate) target: &'a ast::ExprName,
+    pub(crate) first: bool,
+    pub(crate) is_async: bool,
+}
+
+#[derive(Copy, Clone, Debug)]
+pub(crate) struct MatchPatternDefinitionNodeRef<'a> {
+    /// The outermost pattern node in which the identifier being defined occurs.
+    pub(crate) pattern: &'a ast::Pattern,
+    /// The identifier being defined.
+    pub(crate) identifier: &'a ast::Identifier,
+    /// The index of the identifier in the pattern when visiting the `pattern` node in evaluation
+    /// order.
+    pub(crate) index: u32,
+}
+
 impl DefinitionNodeRef<'_> {
     #[allow(unsafe_code)]
     pub(super) unsafe fn into_owned(self, parsed: ParsedModule) -> DefinitionKind {
@@ -131,6 +224,59 @@ impl DefinitionNodeRef<'_> {
             DefinitionNodeRef::AnnotatedAssignment(assign) => {
                 DefinitionKind::AnnotatedAssignment(AstNodeRef::new(parsed, assign))
             }
+            DefinitionNodeRef::AugmentedAssignment(augmented_assignment) => {
+                DefinitionKind::AugmentedAssignment(AstNodeRef::new(parsed, augmented_assignment))
+            }
+            DefinitionNodeRef::For(ForStmtDefinitionNodeRef {
+                iterable,
+                target,
+                is_async,
+            }) => DefinitionKind::For(ForStmtDefinitionKind {
+                iterable: AstNodeRef::new(parsed.clone(), iterable),
+                target: AstNodeRef::new(parsed, target),
+                is_async,
+            }),
+            DefinitionNodeRef::Comprehension(ComprehensionDefinitionNodeRef {
+                iterable,
+                target,
+                first,
+                is_async,
+            }) => DefinitionKind::Comprehension(ComprehensionDefinitionKind {
+                iterable: AstNodeRef::new(parsed.clone(), iterable),
+                target: AstNodeRef::new(parsed, target),
+                first,
+                is_async,
+            }),
+            DefinitionNodeRef::Parameter(parameter) => match parameter {
+                ast::AnyParameterRef::Variadic(parameter) => {
+                    DefinitionKind::Parameter(AstNodeRef::new(parsed, parameter))
+                }
+                ast::AnyParameterRef::NonVariadic(parameter) => {
+                    DefinitionKind::ParameterWithDefault(AstNodeRef::new(parsed, parameter))
+                }
+            },
+            DefinitionNodeRef::WithItem(WithItemDefinitionNodeRef { node, target }) => {
+                DefinitionKind::WithItem(WithItemDefinitionKind {
+                    node: AstNodeRef::new(parsed.clone(), node),
+                    target: AstNodeRef::new(parsed, target),
+                })
+            }
+            DefinitionNodeRef::MatchPattern(MatchPatternDefinitionNodeRef {
+                pattern,
+                identifier,
+                index,
+            }) => DefinitionKind::MatchPattern(MatchPatternDefinitionKind {
+                pattern: AstNodeRef::new(parsed.clone(), pattern),
+                identifier: AstNodeRef::new(parsed, identifier),
+                index,
+            }),
+            DefinitionNodeRef::ExceptHandler(ExceptHandlerDefinitionNodeRef {
+                handler,
+                is_star,
+            }) => DefinitionKind::ExceptHandler(ExceptHandlerDefinitionKind {
+                handler: AstNodeRef::new(parsed.clone(), handler),
+                is_star,
+            }),
         }
     }
 
@@ -148,10 +294,61 @@ impl DefinitionNodeRef<'_> {
                 target,
             }) => target.into(),
             Self::AnnotatedAssignment(node) => node.into(),
+            Self::AugmentedAssignment(node) => node.into(),
+            Self::For(ForStmtDefinitionNodeRef {
+                iterable: _,
+                target,
+                is_async: _,
+            }) => target.into(),
+            Self::Comprehension(ComprehensionDefinitionNodeRef { target, .. }) => target.into(),
+            Self::Parameter(node) => match node {
+                ast::AnyParameterRef::Variadic(parameter) => parameter.into(),
+                ast::AnyParameterRef::NonVariadic(parameter) => parameter.into(),
+            },
+            Self::WithItem(WithItemDefinitionNodeRef { node: _, target }) => target.into(),
+            Self::MatchPattern(MatchPatternDefinitionNodeRef { identifier, .. }) => {
+                identifier.into()
+            }
+            Self::ExceptHandler(ExceptHandlerDefinitionNodeRef { handler, .. }) => handler.into(),
         }
     }
 }
 
+#[derive(Clone, Copy, Debug)]
+pub(crate) enum DefinitionCategory {
+    /// A Definition which binds a value to a name (e.g. `x = 1`).
+    Binding,
+    /// A Definition which declares the upper-bound of acceptable types for this name (`x: int`).
+    Declaration,
+    /// A Definition which both declares a type and binds a value (e.g. `x: int = 1`).
+    DeclarationAndBinding,
+}
+
+impl DefinitionCategory {
+    /// True if this definition establishes a "declared type" for the symbol.
+    ///
+    /// If so, any assignments reached by this definition are in error if they assign a value of a
+    /// type not assignable to the declared type.
+    ///
+    /// Annotations establish a declared type. So do function and class definitions, and imports.
+    pub(crate) fn is_declaration(self) -> bool {
+        matches!(
+            self,
+            DefinitionCategory::Declaration | DefinitionCategory::DeclarationAndBinding
+        )
+    }
+
+    /// True if this definition assigns a value to the symbol.
+    ///
+    /// False only for annotated assignments without a RHS.
+    pub(crate) fn is_binding(self) -> bool {
+        matches!(
+            self,
+            DefinitionCategory::Binding | DefinitionCategory::DeclarationAndBinding
+        )
+    }
+}
+
 #[derive(Clone, Debug)]
 pub enum DefinitionKind {
     Import(AstNodeRef<ast::Alias>),
@@ -161,6 +358,103 @@ pub enum DefinitionKind {
     NamedExpression(AstNodeRef<ast::ExprNamed>),
     Assignment(AssignmentDefinitionKind),
     AnnotatedAssignment(AstNodeRef<ast::StmtAnnAssign>),
+    AugmentedAssignment(AstNodeRef<ast::StmtAugAssign>),
+    For(ForStmtDefinitionKind),
+    Comprehension(ComprehensionDefinitionKind),
+    Parameter(AstNodeRef<ast::Parameter>),
+    ParameterWithDefault(AstNodeRef<ast::ParameterWithDefault>),
+    WithItem(WithItemDefinitionKind),
+    MatchPattern(MatchPatternDefinitionKind),
+    ExceptHandler(ExceptHandlerDefinitionKind),
+}
+
+impl DefinitionKind {
+    pub(crate) fn category(&self) -> DefinitionCategory {
+        match self {
+            // functions, classes, and imports always bind, and we consider them declarations
+            DefinitionKind::Function(_)
+            | DefinitionKind::Class(_)
+            | DefinitionKind::Import(_)
+            | DefinitionKind::ImportFrom(_) => DefinitionCategory::DeclarationAndBinding,
+            // a parameter always binds a value, but is only a declaration if annotated
+            DefinitionKind::Parameter(parameter) => {
+                if parameter.annotation.is_some() {
+                    DefinitionCategory::DeclarationAndBinding
+                } else {
+                    DefinitionCategory::Binding
+                }
+            }
+            // presence of a default is irrelevant, same logic as for a no-default parameter
+            DefinitionKind::ParameterWithDefault(parameter_with_default) => {
+                if parameter_with_default.parameter.annotation.is_some() {
+                    DefinitionCategory::DeclarationAndBinding
+                } else {
+                    DefinitionCategory::Binding
+                }
+            }
+            // annotated assignment is always a declaration, only a binding if there is a RHS
+            DefinitionKind::AnnotatedAssignment(ann_assign) => {
+                if ann_assign.value.is_some() {
+                    DefinitionCategory::DeclarationAndBinding
+                } else {
+                    DefinitionCategory::Declaration
+                }
+            }
+            // all of these bind values without declaring a type
+            DefinitionKind::NamedExpression(_)
+            | DefinitionKind::Assignment(_)
+            | DefinitionKind::AugmentedAssignment(_)
+            | DefinitionKind::For(_)
+            | DefinitionKind::Comprehension(_)
+            | DefinitionKind::WithItem(_)
+            | DefinitionKind::MatchPattern(_)
+            | DefinitionKind::ExceptHandler(_) => DefinitionCategory::Binding,
+        }
+    }
+}
+
+#[derive(Clone, Debug)]
+#[allow(dead_code)]
+pub struct MatchPatternDefinitionKind {
+    pattern: AstNodeRef<ast::Pattern>,
+    identifier: AstNodeRef<ast::Identifier>,
+    index: u32,
+}
+
+impl MatchPatternDefinitionKind {
+    pub(crate) fn pattern(&self) -> &ast::Pattern {
+        self.pattern.node()
+    }
+
+    pub(crate) fn index(&self) -> u32 {
+        self.index
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct ComprehensionDefinitionKind {
+    iterable: AstNodeRef<ast::Expr>,
+    target: AstNodeRef<ast::ExprName>,
+    first: bool,
+    is_async: bool,
+}
+
+impl ComprehensionDefinitionKind {
+    pub(crate) fn iterable(&self) -> &ast::Expr {
+        self.iterable.node()
+    }
+
+    pub(crate) fn target(&self) -> &ast::ExprName {
+        self.target.node()
+    }
+
+    pub(crate) fn is_first(&self) -> bool {
+        self.first
+    }
+
+    pub(crate) fn is_async(&self) -> bool {
+        self.is_async
+    }
 }
 
 #[derive(Clone, Debug)]
@@ -180,7 +474,6 @@ impl ImportFromDefinitionKind {
 }
 
 #[derive(Clone, Debug)]
-#[allow(dead_code)]
 pub struct AssignmentDefinitionKind {
     assignment: AstNodeRef<ast::StmtAssign>,
     target: AstNodeRef<ast::ExprName>,
@@ -190,6 +483,67 @@ impl AssignmentDefinitionKind {
     pub(crate) fn assignment(&self) -> &ast::StmtAssign {
         self.assignment.node()
     }
+
+    pub(crate) fn target(&self) -> &ast::ExprName {
+        self.target.node()
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct WithItemDefinitionKind {
+    node: AstNodeRef<ast::WithItem>,
+    target: AstNodeRef<ast::ExprName>,
+}
+
+impl WithItemDefinitionKind {
+    pub(crate) fn node(&self) -> &ast::WithItem {
+        self.node.node()
+    }
+
+    pub(crate) fn target(&self) -> &ast::ExprName {
+        self.target.node()
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct ForStmtDefinitionKind {
+    iterable: AstNodeRef<ast::Expr>,
+    target: AstNodeRef<ast::ExprName>,
+    is_async: bool,
+}
+
+impl ForStmtDefinitionKind {
+    pub(crate) fn iterable(&self) -> &ast::Expr {
+        self.iterable.node()
+    }
+
+    pub(crate) fn target(&self) -> &ast::ExprName {
+        self.target.node()
+    }
+
+    pub(crate) fn is_async(&self) -> bool {
+        self.is_async
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct ExceptHandlerDefinitionKind {
+    handler: AstNodeRef<ast::ExceptHandlerExceptHandler>,
+    is_star: bool,
+}
+
+impl ExceptHandlerDefinitionKind {
+    pub(crate) fn node(&self) -> &ast::ExceptHandlerExceptHandler {
+        self.handler.node()
+    }
+
+    pub(crate) fn handled_exceptions(&self) -> Option<&ast::Expr> {
+        self.node().type_.as_deref()
+    }
+
+    pub(crate) fn is_star(&self) -> bool {
+        self.is_star
+    }
 }
 
 #[derive(Copy, Clone, Eq, PartialEq, Hash, Debug)]
@@ -230,3 +584,39 @@ impl From<&ast::StmtAnnAssign> for DefinitionNodeKey {
         Self(NodeKey::from_node(node))
     }
 }
+
+impl From<&ast::StmtAugAssign> for DefinitionNodeKey {
+    fn from(node: &ast::StmtAugAssign) -> Self {
+        Self(NodeKey::from_node(node))
+    }
+}
+
+impl From<&ast::StmtFor> for DefinitionNodeKey {
+    fn from(value: &ast::StmtFor) -> Self {
+        Self(NodeKey::from_node(value))
+    }
+}
+
+impl From<&ast::Parameter> for DefinitionNodeKey {
+    fn from(node: &ast::Parameter) -> Self {
+        Self(NodeKey::from_node(node))
+    }
+}
+
+impl From<&ast::ParameterWithDefault> for DefinitionNodeKey {
+    fn from(node: &ast::ParameterWithDefault) -> Self {
+        Self(NodeKey::from_node(node))
+    }
+}
+
+impl From<&ast::Identifier> for DefinitionNodeKey {
+    fn from(identifier: &ast::Identifier) -> Self {
+        Self(NodeKey::from_node(identifier))
+    }
+}
+
+impl From<&ast::ExceptHandlerExceptHandler> for DefinitionNodeKey {
+    fn from(handler: &ast::ExceptHandlerExceptHandler) -> Self {
+        Self(NodeKey::from_node(handler))
+    }
+}

crates/red_knot_python_semantic/src/semantic_index/expression.rs

@@ -21,7 +21,7 @@ pub(crate) struct Expression<'db> {
     /// The expression node.
     #[no_eq]
     #[return_ref]
-    pub(crate) node: AstNodeRef<ast::Expr>,
+    pub(crate) node_ref: AstNodeRef<ast::Expr>,
 
     #[no_eq]
     count: countme::Count<Expression<'static>>,

crates/red_knot_python_semantic/src/semantic_index/symbol.rs

@@ -44,16 +44,16 @@ impl Symbol {
     }
 
     /// Is the symbol defined in its containing scope?
-    pub fn is_defined(&self) -> bool {
-        self.flags.contains(SymbolFlags::IS_DEFINED)
+    pub fn is_bound(&self) -> bool {
+        self.flags.contains(SymbolFlags::IS_BOUND)
     }
 }
 
 bitflags! {
     #[derive(Copy, Clone, Debug, Eq, PartialEq)]
-    pub(super) struct SymbolFlags: u8 {
+    struct SymbolFlags: u8 {
         const IS_USED         = 1 << 0;
-        const IS_DEFINED      = 1 << 1;
+        const IS_BOUND      = 1 << 1;
         /// TODO: This flag is not yet set by anything
         const MARKED_GLOBAL   = 1 << 2;
         /// TODO: This flag is not yet set by anything
@@ -114,6 +114,10 @@ impl<'db> ScopeId<'db> {
             NodeWithScopeKind::ClassTypeParameters(_)
                 | NodeWithScopeKind::FunctionTypeParameters(_)
                 | NodeWithScopeKind::Function(_)
+                | NodeWithScopeKind::ListComprehension(_)
+                | NodeWithScopeKind::SetComprehension(_)
+                | NodeWithScopeKind::DictComprehension(_)
+                | NodeWithScopeKind::GeneratorExpression(_)
         )
     }
 
@@ -127,6 +131,10 @@ impl<'db> ScopeId<'db> {
             NodeWithScopeKind::Function(function)
             | NodeWithScopeKind::FunctionTypeParameters(function) => function.name.as_str(),
             NodeWithScopeKind::Lambda(_) => "<lambda>",
+            NodeWithScopeKind::ListComprehension(_) => "<listcomp>",
+            NodeWithScopeKind::SetComprehension(_) => "<setcomp>",
+            NodeWithScopeKind::DictComprehension(_) => "<dictcomp>",
+            NodeWithScopeKind::GeneratorExpression(_) => "<generator>",
         }
     }
 }
@@ -141,6 +149,10 @@ impl FileScopeId {
         FileScopeId::from_u32(0)
     }
 
+    pub fn is_global(self) -> bool {
+        self == FileScopeId::global()
+    }
+
     pub fn to_scope_id(self, db: &dyn Db, file: File) -> ScopeId<'_> {
         let index = semantic_index(db, file);
         index.scope_ids_by_scope[self]
@@ -170,6 +182,13 @@ pub enum ScopeKind {
     Annotation,
     Class,
     Function,
+    Comprehension,
+}
+
+impl ScopeKind {
+    pub const fn is_comprehension(self) -> bool {
+        matches!(self, ScopeKind::Comprehension)
+    }
 }
 
 /// Symbol table for a specific [`Scope`].
@@ -253,11 +272,7 @@ impl SymbolTableBuilder {
         }
     }
 
-    pub(super) fn add_or_update_symbol(
-        &mut self,
-        name: Name,
-        flags: SymbolFlags,
-    ) -> (ScopedSymbolId, bool) {
+    pub(super) fn add_symbol(&mut self, name: Name) -> (ScopedSymbolId, bool) {
         let hash = SymbolTable::hash_name(&name);
         let entry = self
             .table
@@ -266,15 +281,9 @@ impl SymbolTableBuilder {
             .from_hash(hash, |id| self.table.symbols[*id].name() == &name);
 
         match entry {
-            RawEntryMut::Occupied(entry) => {
-                let symbol = &mut self.table.symbols[*entry.key()];
-                symbol.insert_flags(flags);
-
-                (*entry.key(), false)
-            }
+            RawEntryMut::Occupied(entry) => (*entry.key(), false),
             RawEntryMut::Vacant(entry) => {
-                let mut symbol = Symbol::new(name);
-                symbol.insert_flags(flags);
+                let symbol = Symbol::new(name);
 
                 let id = self.table.symbols.push(symbol);
                 entry.insert_with_hasher(hash, id, (), |id| {
@@ -285,6 +294,14 @@ impl SymbolTableBuilder {
         }
     }
 
+    pub(super) fn mark_symbol_bound(&mut self, id: ScopedSymbolId) {
+        self.table.symbols[id].insert_flags(SymbolFlags::IS_BOUND);
+    }
+
+    pub(super) fn mark_symbol_used(&mut self, id: ScopedSymbolId) {
+        self.table.symbols[id].insert_flags(SymbolFlags::IS_USED);
+    }
+
     pub(super) fn finish(mut self) -> SymbolTable {
         self.table.shrink_to_fit();
         self.table
@@ -300,6 +317,10 @@ pub(crate) enum NodeWithScopeRef<'a> {
     Lambda(&'a ast::ExprLambda),
     FunctionTypeParameters(&'a ast::StmtFunctionDef),
     ClassTypeParameters(&'a ast::StmtClassDef),
+    ListComprehension(&'a ast::ExprListComp),
+    SetComprehension(&'a ast::ExprSetComp),
+    DictComprehension(&'a ast::ExprDictComp),
+    GeneratorExpression(&'a ast::ExprGenerator),
 }
 
 impl NodeWithScopeRef<'_> {
@@ -326,6 +347,18 @@ impl NodeWithScopeRef<'_> {
             NodeWithScopeRef::ClassTypeParameters(class) => {
                 NodeWithScopeKind::ClassTypeParameters(AstNodeRef::new(module, class))
             }
+            NodeWithScopeRef::ListComprehension(comprehension) => {
+                NodeWithScopeKind::ListComprehension(AstNodeRef::new(module, comprehension))
+            }
+            NodeWithScopeRef::SetComprehension(comprehension) => {
+                NodeWithScopeKind::SetComprehension(AstNodeRef::new(module, comprehension))
+            }
+            NodeWithScopeRef::DictComprehension(comprehension) => {
+                NodeWithScopeKind::DictComprehension(AstNodeRef::new(module, comprehension))
+            }
+            NodeWithScopeRef::GeneratorExpression(generator) => {
+                NodeWithScopeKind::GeneratorExpression(AstNodeRef::new(module, generator))
+            }
         }
     }
 
@@ -337,6 +370,10 @@ impl NodeWithScopeRef<'_> {
             NodeWithScopeRef::Lambda(_) => ScopeKind::Function,
             NodeWithScopeRef::FunctionTypeParameters(_)
             | NodeWithScopeRef::ClassTypeParameters(_) => ScopeKind::Annotation,
+            NodeWithScopeRef::ListComprehension(_)
+            | NodeWithScopeRef::SetComprehension(_)
+            | NodeWithScopeRef::DictComprehension(_)
+            | NodeWithScopeRef::GeneratorExpression(_) => ScopeKind::Comprehension,
         }
     }
 
@@ -356,6 +393,18 @@ impl NodeWithScopeRef<'_> {
             NodeWithScopeRef::ClassTypeParameters(class) => {
                 NodeWithScopeKey::ClassTypeParameters(NodeKey::from_node(class))
             }
+            NodeWithScopeRef::ListComprehension(comprehension) => {
+                NodeWithScopeKey::ListComprehension(NodeKey::from_node(comprehension))
+            }
+            NodeWithScopeRef::SetComprehension(comprehension) => {
+                NodeWithScopeKey::SetComprehension(NodeKey::from_node(comprehension))
+            }
+            NodeWithScopeRef::DictComprehension(comprehension) => {
+                NodeWithScopeKey::DictComprehension(NodeKey::from_node(comprehension))
+            }
+            NodeWithScopeRef::GeneratorExpression(generator) => {
+                NodeWithScopeKey::GeneratorExpression(NodeKey::from_node(generator))
+            }
         }
     }
 }
@@ -369,6 +418,10 @@ pub enum NodeWithScopeKind {
     Function(AstNodeRef<ast::StmtFunctionDef>),
     FunctionTypeParameters(AstNodeRef<ast::StmtFunctionDef>),
     Lambda(AstNodeRef<ast::ExprLambda>),
+    ListComprehension(AstNodeRef<ast::ExprListComp>),
+    SetComprehension(AstNodeRef<ast::ExprSetComp>),
+    DictComprehension(AstNodeRef<ast::ExprDictComp>),
+    GeneratorExpression(AstNodeRef<ast::ExprGenerator>),
 }
 
 #[derive(Copy, Clone, Debug, Eq, PartialEq, Hash)]
@@ -379,4 +432,8 @@ pub(crate) enum NodeWithScopeKey {
     Function(NodeKey),
     FunctionTypeParameters(NodeKey),
     Lambda(NodeKey),
+    ListComprehension(NodeKey),
+    SetComprehension(NodeKey),
+    DictComprehension(NodeKey),
+    GeneratorExpression(NodeKey),
 }

crates/red_knot_python_semantic/src/semantic_index/use_def.rs

@@ -1,4 +1,79 @@
-//! Build a map from each use of a symbol to the definitions visible from that use.
+//! First, some terminology:
+//!
+//! * A "binding" gives a new value to a variable. This includes many different Python statements
+//!   (assignment statements of course, but also imports, `def` and `class` statements, `as`
+//!   clauses in `with` and `except` statements, match patterns, and others) and even one
+//!   expression kind (named expressions). It notably does not include annotated assignment
+//!   statements without a right-hand side value; these do not assign any new value to the
+//!   variable. We consider function parameters to be bindings as well, since (from the perspective
+//!   of the function's internal scope), a function parameter begins the scope bound to a value.
+//!
+//! * A "declaration" establishes an upper bound type for the values that a variable may be
+//!   permitted to take on. Annotated assignment statements (with or without an RHS value) are
+//!   declarations; annotated function parameters are also declarations. We consider `def` and
+//!   `class` statements to also be declarations, so as to prohibit accidentally shadowing them.
+//!
+//! Annotated assignments with a right-hand side, and annotated function parameters, are both
+//! bindings and declarations.
+//!
+//! We use [`Definition`] as the universal term (and Salsa tracked struct) encompassing both
+//! bindings and declarations. (This sacrifices a bit of type safety in exchange for improved
+//! performance via fewer Salsa tracked structs and queries, since most declarations -- typed
+//! parameters and annotated assignments with RHS -- are both bindings and declarations.)
+//!
+//! At any given use of a variable, we can ask about both its "declared type" and its "inferred
+//! type". These may be different, but the inferred type must always be assignable to the declared
+//! type; that is, the declared type is always wider, and the inferred type may be more precise. If
+//! we see an invalid assignment, we emit a diagnostic and abandon our inferred type, deferring to
+//! the declared type (this allows an explicit annotation to override bad inference, without a
+//! cast), maintaining the invariant.
+//!
+//! The **inferred type** represents the most precise type we believe encompasses all possible
+//! values for the variable at a given use. It is based on a union of the bindings which can reach
+//! that use through some control flow path, and the narrowing constraints that control flow must
+//! have passed through between the binding and the use. For example, in this code:
+//!
+//! ```python
+//! x = 1 if flag else None
+//! if x is not None:
+//!     use(x)
+//! ```
+//!
+//! For the use of `x` on the third line, the inferred type should be `Literal[1]`. This is based
+//! on the binding on the first line, which assigns the type `Literal[1] | None`, and the narrowing
+//! constraint on the second line, which rules out the type `None`, since control flow must pass
+//! through this constraint to reach the use in question.
+//!
+//! The **declared type** represents the code author's declaration (usually through a type
+//! annotation) that a given variable should not be assigned any type outside the declared type. In
+//! our model, declared types are also control-flow-sensitive; we allow the code author to
+//! explicitly re-declare the same variable with a different type. So for a given binding of a
+//! variable, we will want to ask which declarations of that variable can reach that binding, in
+//! order to determine whether the binding is permitted, or should be a type error. For example:
+//!
+//! ```python
+//! from pathlib import Path
+//! def f(path: str):
+//!     path: Path = Path(path)
+//! ```
+//!
+//! In this function, the initial declared type of `path` is `str`, meaning that the assignment
+//! `path = Path(path)` would be a type error, since it assigns to `path` a value whose type is not
+//! assignable to `str`. This is the purpose of declared types: they prevent accidental assignment
+//! of the wrong type to a variable.
+//!
+//! But in some cases it is useful to "shadow" or "re-declare" a variable with a new type, and we
+//! permit this, as long as it is done with an explicit re-annotation. So `path: Path =
+//! Path(path)`, with the explicit `: Path` annotation, is permitted.
+//!
+//! The general rule is that whatever declaration(s) can reach a given binding determine the
+//! validity of that binding. If there is a path in which the symbol is not declared, that is a
+//! declaration of `Unknown`. If multiple declarations can reach a binding, we union them, but by
+//! default we also issue a type error, since this implicit union of declared types may hide an
+//! error.
+//!
+//! To support type inference, we build a map from each use of a symbol to the bindings live at
+//! that use, and the type narrowing constraints that apply to each binding.
 //!
 //! Let's take this code sample:
 //!
@@ -13,342 +88,488 @@
 //! z = x
 //! ```
 //!
-//! In this snippet, we have four definitions of `x` (the statements assigning `1`, `2`, `3`,
-//! and `4` to it), and two uses of `x` (the `y = x` and `z = x` assignments). The first
-//! [`Definition`] of `x` is never visible to any use, because it's immediately replaced by the
-//! second definition, before any use happens. (A linter could thus flag the statement `x = 1`
-//! as likely superfluous.)
+//! In this snippet, we have four bindings of `x` (the statements assigning `1`, `2`, `3`, and `4`
+//! to it), and two uses of `x` (the `y = x` and `z = x` assignments). The first binding of `x`
+//! does not reach any use, because it's immediately replaced by the second binding, before any use
+//! happens. (A linter could thus flag the statement `x = 1` as likely superfluous.)
 //!
-//! The first use of `x` has one definition visible to it: the assignment `x = 2`.
+//! The first use of `x` has one live binding: the assignment `x = 2`.
 //!
 //! Things get a bit more complex when we have branches. We will definitely take either the `if` or
-//! the `else` branch. Thus, the second use of `x` has two definitions visible to it: `x = 3` and
-//! `x = 4`. The `x = 2` definition is no longer visible, because it must be replaced by either `x
-//! = 3` or `x = 4`, no matter which branch was taken. We don't know which branch was taken, so we
-//! must consider both definitions as visible, which means eventually we would (in type inference)
-//! look at these two definitions and infer a type of `Literal[3, 4]` -- the union of `Literal[3]`
-//! and `Literal[4]` -- for the second use of `x`.
+//! the `else` branch. Thus, the second use of `x` has two live bindings: `x = 3` and `x = 4`. The
+//! `x = 2` assignment is no longer visible, because it must be replaced by either `x = 3` or `x =
+//! 4`, no matter which branch was taken. We don't know which branch was taken, so we must consider
+//! both bindings as live, which means eventually we would (in type inference) look at these two
+//! bindings and infer a type of `Literal[3, 4]` -- the union of `Literal[3]` and `Literal[4]` --
+//! for the second use of `x`.
 //!
 //! So that's one question our use-def map needs to answer: given a specific use of a symbol, which
-//! definition(s) is/are visible from that use. In
-//! [`AstIds`](crate::semantic_index::ast_ids::AstIds) we number all uses (that means a `Name` node
-//! with `Load` context) so we have a `ScopedUseId` to efficiently represent each use.
+//! binding(s) can reach that use. In [`AstIds`](crate::semantic_index::ast_ids::AstIds) we number
+//! all uses (that means a `Name` node with `Load` context) so we have a `ScopedUseId` to
+//! efficiently represent each use.
 //!
-//! The other case we need to handle is when a symbol is referenced from a different scope (the
-//! most obvious example of this is an import). We call this "public" use of a symbol. So the other
-//! question we need to be able to answer is, what are the publicly-visible definitions of each
-//! symbol?
+//! We also need to know, for a given definition of a symbol, what type narrowing constraints apply
+//! to it. For instance, in this code sample:
 //!
-//! Technically, public use of a symbol could also occur from any point in control flow of the
-//! scope where the symbol is defined (via inline imports and import cycles, in the case of an
-//! import, or via a function call partway through the local scope that ends up using a symbol from
-//! the scope via a global or nonlocal reference.) But modeling this fully accurately requires
-//! whole-program analysis that isn't tractable for an efficient incremental compiler, since it
-//! means a given symbol could have a different type every place it's referenced throughout the
-//! program, depending on the shape of arbitrarily-sized call/import graphs. So we follow other
-//! Python type-checkers in making the simplifying assumption that usually the scope will finish
-//! execution before its symbols are made visible to other scopes; for instance, most imports will
-//! import from a complete module, not a partially-executed module. (We may want to get a little
-//! smarter than this in the future, in particular for closures, but for now this is where we
-//! start.)
+//! ```python
+//! x = 1 if flag else None
+//! if x is not None:
+//!     use(x)
+//! ```
 //!
-//! So this means that the publicly-visible definitions of a symbol are the definitions still
-//! visible at the end of the scope.
+//! At the use of `x`, the live binding of `x` is `1 if flag else None`, which would infer as the
+//! type `Literal[1] | None`. But the constraint `x is not None` dominates this use, which means we
+//! can rule out the possibility that `x` is `None` here, which should give us the type
+//! `Literal[1]` for this use.
 //!
-//! The data structure we build to answer these two questions is the `UseDefMap`. It has a
-//! `definitions_by_use` vector indexed by [`ScopedUseId`] and a `public_definitions` vector
-//! indexed by [`ScopedSymbolId`]. The values in each of these vectors are (in principle) a list of
-//! visible definitions at that use, or at the end of the scope for that symbol.
+//! For declared types, we need to be able to answer the question "given a binding to a symbol,
+//! which declarations of that symbol can reach the binding?" This allows us to emit a diagnostic
+//! if the binding is attempting to bind a value of a type that is not assignable to the declared
+//! type for that symbol, at that point in control flow.
 //!
-//! In order to avoid vectors-of-vectors and all the allocations that would entail, we don't
-//! actually store these "list of visible definitions" as a vector of [`Definition`] IDs. Instead,
-//! the values in `definitions_by_use` and `public_definitions` are a [`Definitions`] struct that
-//! keeps a [`Range`] into a third vector of [`Definition`] IDs, `all_definitions`. The trick with
-//! this representation is that it requires that the definitions visible at any given use of a
-//! symbol are stored sequentially in `all_definitions`.
+//! We also need to know, given a declaration of a symbol, what the inferred type of that symbol is
+//! at that point. This allows us to emit a diagnostic in a case like `x = "foo"; x: int`. The
+//! binding `x = "foo"` occurs before the declaration `x: int`, so according to our
+//! control-flow-sensitive interpretation of declarations, the assignment is not an error. But the
+//! declaration is an error, since it would violate the "inferred type must be assignable to
+//! declared type" rule.
 //!
-//! There is another special kind of possible "definition" for a symbol: it might be unbound in the
-//! scope. (This isn't equivalent to "zero visible definitions", since we may go through an `if`
-//! that has a definition for the symbol, leaving us with one visible definition, but still also
-//! the "unbound" possibility, since we might not have taken the `if` branch.)
+//! Another case we need to handle is when a symbol is referenced from a different scope (for
+//! example, an import or a nonlocal reference). We call this "public" use of a symbol. For public
+//! use of a symbol, we prefer the declared type, if there are any declarations of that symbol; if
+//! not, we fall back to the inferred type. So we also need to know which declarations and bindings
+//! can reach the end of the scope.
 //!
-//! The simplest way to model "unbound" would be as an actual [`Definition`] itself: the initial
-//! visible [`Definition`] for each symbol in a scope. But actually modeling it this way would
-//! dramatically increase the number of [`Definition`] that Salsa must track. Since "unbound" is a
-//! special definition in that all symbols share it, and it doesn't have any additional per-symbol
-//! state, we can represent it more efficiently: we use the `may_be_unbound` boolean on the
-//! [`Definitions`] struct. If this flag is `true`, it means the symbol/use really has one
-//! additional visible "definition", which is the unbound state. If this flag is `false`, it means
-//! we've eliminated the possibility of unbound: every path we've followed includes a definition
+//! Technically, public use of a symbol could occur from any point in control flow of the scope
+//! where the symbol is defined (via inline imports and import cycles, in the case of an import, or
+//! via a function call partway through the local scope that ends up using a symbol from the scope
+//! via a global or nonlocal reference.) But modeling this fully accurately requires whole-program
+//! analysis that isn't tractable for an efficient analysis, since it means a given symbol could
+//! have a different type every place it's referenced throughout the program, depending on the
+//! shape of arbitrarily-sized call/import graphs. So we follow other Python type checkers in
+//! making the simplifying assumption that usually the scope will finish execution before its
+//! symbols are made visible to other scopes; for instance, most imports will import from a
+//! complete module, not a partially-executed module. (We may want to get a little smarter than
+//! this in the future for some closures, but for now this is where we start.)
+//!
+//! The data structure we build to answer these questions is the `UseDefMap`. It has a
+//! `bindings_by_use` vector of [`SymbolBindings`] indexed by [`ScopedUseId`], a
+//! `declarations_by_binding` vector of [`SymbolDeclarations`] indexed by [`ScopedDefinitionId`], a
+//! `bindings_by_declaration` vector of [`SymbolBindings`] indexed by [`ScopedDefinitionId`], and
+//! `public_bindings` and `public_definitions` vectors indexed by [`ScopedSymbolId`]. The values in
+//! each of these vectors are (in principle) a list of live bindings at that use/definition, or at
+//! the end of the scope for that symbol, with a list of the dominating constraints for each
+//! binding.
+//!
+//! In order to avoid vectors-of-vectors-of-vectors and all the allocations that would entail, we
+//! don't actually store these "list of visible definitions" as a vector of [`Definition`].
+//! Instead, [`SymbolBindings`] and [`SymbolDeclarations`] are structs which use bit-sets to track
+//! definitions (and constraints, in the case of bindings) in terms of [`ScopedDefinitionId`] and
+//! [`ScopedConstraintId`], which are indices into the `all_definitions` and `all_constraints`
+//! indexvecs in the [`UseDefMap`].
+//!
+//! There is another special kind of possible "definition" for a symbol: there might be a path from
+//! the scope entry to a given use in which the symbol is never bound.
+//!
+//! The simplest way to model "unbound" would be as a "binding" itself: the initial "binding" for
+//! each symbol in a scope. But actually modeling it this way would unnecessarily increase the
+//! number of [`Definition`]s that Salsa must track. Since "unbound" is special in that all symbols
+//! share it, and it doesn't have any additional per-symbol state, and constraints are irrelevant
+//! to it, we can represent it more efficiently: we use the `may_be_unbound` boolean on the
+//! [`SymbolBindings`] struct. If this flag is `true` for a use of a symbol, it means the symbol
+//! has a path to the use in which it is never bound. If this flag is `false`, it means we've
+//! eliminated the possibility of unbound: every control flow path to the use includes a binding
 //! for this symbol.
 //!
-//! To build a [`UseDefMap`], the [`UseDefMapBuilder`] is notified of each new use and definition
-//! as they are encountered by the
+//! To build a [`UseDefMap`], the [`UseDefMapBuilder`] is notified of each new use, definition, and
+//! constraint as they are encountered by the
 //! [`SemanticIndexBuilder`](crate::semantic_index::builder::SemanticIndexBuilder) AST visit. For
-//! each symbol, the builder tracks the currently-visible definitions for that symbol. When we hit
-//! a use of a symbol, it records the currently-visible definitions for that symbol as the visible
-//! definitions for that use. When we reach the end of the scope, it records the currently-visible
-//! definitions for each symbol as the public definitions of that symbol.
+//! each symbol, the builder tracks the `SymbolState` (`SymbolBindings` and `SymbolDeclarations`)
+//! for that symbol. When we hit a use or definition of a symbol, we record the necessary parts of
+//! the current state for that symbol that we need for that use or definition. When we reach the
+//! end of the scope, it records the state for each symbol as the public definitions of that
+//! symbol.
 //!
-//! Let's walk through the above example. Initially we record for `x` that it has no visible
-//! definitions, and may be unbound. When we see `x = 1`, we record that as the sole visible
-//! definition of `x`, and flip `may_be_unbound` to `false`. Then we see `x = 2`, and it replaces
-//! `x = 1` as the sole visible definition of `x`. When we get to `y = x`, we record that the
-//! visible definitions for that use of `x` are just the `x = 2` definition.
+//! Let's walk through the above example. Initially we record for `x` that it has no bindings, and
+//! may be unbound. When we see `x = 1`, we record that as the sole live binding of `x`, and flip
+//! `may_be_unbound` to `false`. Then we see `x = 2`, and we replace `x = 1` as the sole live
+//! binding of `x`. When we get to `y = x`, we record that the live bindings for that use of `x`
+//! are just the `x = 2` definition.
 //!
 //! Then we hit the `if` branch. We visit the `test` node (`flag` in this case), since that will
-//! happen regardless. Then we take a pre-branch snapshot of the currently visible definitions for
-//! all symbols, which we'll need later. Then we go ahead and visit the `if` body. When we see `x =
-//! 3`, it replaces `x = 2` as the sole visible definition of `x`. At the end of the `if` body, we
-//! take another snapshot of the currently-visible definitions; we'll call this the post-if-body
-//! snapshot.
+//! happen regardless. Then we take a pre-branch snapshot of the current state for all symbols,
+//! which we'll need later. Then we record `flag` as a possible constraint on the current binding
+//! (`x = 2`), and go ahead and visit the `if` body. When we see `x = 3`, it replaces `x = 2`
+//! (constrained by `flag`) as the sole live binding of `x`. At the end of the `if` body, we take
+//! another snapshot of the current symbol state; we'll call this the post-if-body snapshot.
 //!
 //! Now we need to visit the `else` clause. The conditions when entering the `else` clause should
 //! be the pre-if conditions; if we are entering the `else` clause, we know that the `if` test
 //! failed and we didn't execute the `if` body. So we first reset the builder to the pre-if state,
-//! using the snapshot we took previously (meaning we now have `x = 2` as the sole visible
-//! definition for `x` again), then visit the `else` clause, where `x = 4` replaces `x = 2` as the
-//! sole visible definition of `x`.
+//! using the snapshot we took previously (meaning we now have `x = 2` as the sole binding for `x`
+//! again), then visit the `else` clause, where `x = 4` replaces `x = 2` as the sole live binding
+//! of `x`.
 //!
 //! Now we reach the end of the if/else, and want to visit the following code. The state here needs
 //! to reflect that we might have gone through the `if` branch, or we might have gone through the
 //! `else` branch, and we don't know which. So we need to "merge" our current builder state
-//! (reflecting the end-of-else state, with `x = 4` as the only visible definition) with our
-//! post-if-body snapshot (which has `x = 3` as the only visible definition). The result of this
-//! merge is that we now have two visible definitions of `x`: `x = 3` and `x = 4`.
+//! (reflecting the end-of-else state, with `x = 4` as the only live binding) with our post-if-body
+//! snapshot (which has `x = 3` as the only live binding). The result of this merge is that we now
+//! have two live bindings of `x`: `x = 3` and `x = 4`.
 //!
 //! The [`UseDefMapBuilder`] itself just exposes methods for taking a snapshot, resetting to a
 //! snapshot, and merging a snapshot into the current state. The logic using these methods lives in
 //! [`SemanticIndexBuilder`](crate::semantic_index::builder::SemanticIndexBuilder), e.g. where it
 //! visits a `StmtIf` node.
-//!
-//! (In the future we may have some other questions we want to answer as well, such as "is this
-//! definition used?", which will require tracking a bit more info in our map, e.g. a "used" bit
-//! for each [`Definition`] which is flipped to true when we record that definition for a use.)
+use self::symbol_state::{
+    BindingIdWithConstraintsIterator, ConstraintIdIterator, DeclarationIdIterator,
+    ScopedConstraintId, ScopedDefinitionId, SymbolBindings, SymbolDeclarations, SymbolState,
+};
 use crate::semantic_index::ast_ids::ScopedUseId;
 use crate::semantic_index::definition::Definition;
 use crate::semantic_index::symbol::ScopedSymbolId;
 use ruff_index::IndexVec;
-use std::ops::Range;
+use rustc_hash::FxHashMap;
 
-/// All definitions that can reach a given use of a name.
+use super::constraint::Constraint;
+
+mod bitset;
+mod symbol_state;
+
+/// Applicable definitions and constraints for every use of a name.
 #[derive(Debug, PartialEq, Eq)]
 pub(crate) struct UseDefMap<'db> {
-    // TODO store constraints with definitions for type narrowing
-    /// Definition IDs array for `definitions_by_use` and `public_definitions` to slice into.
-    all_definitions: Vec<Definition<'db>>,
+    /// Array of [`Definition`] in this scope.
+    all_definitions: IndexVec<ScopedDefinitionId, Definition<'db>>,
 
-    /// Definitions that can reach a [`ScopedUseId`].
-    definitions_by_use: IndexVec<ScopedUseId, Definitions>,
+    /// Array of [`Constraint`] in this scope.
+    all_constraints: IndexVec<ScopedConstraintId, Constraint<'db>>,
 
-    /// Definitions of each symbol visible at end of scope.
-    public_definitions: IndexVec<ScopedSymbolId, Definitions>,
+    /// [`SymbolBindings`] reaching a [`ScopedUseId`].
+    bindings_by_use: IndexVec<ScopedUseId, SymbolBindings>,
+
+    /// [`SymbolBindings`] or [`SymbolDeclarations`] reaching a given [`Definition`].
+    ///
+    /// If the definition is a binding (only) -- `x = 1` for example -- then we need
+    /// [`SymbolDeclarations`] to know whether this binding is permitted by the live declarations.
+    ///
+    /// If the definition is a declaration (only) -- `x: int` for example -- then we need
+    /// [`SymbolBindings`] to know whether this declaration is consistent with the previously
+    /// inferred type.
+    ///
+    /// If the definition is both a declaration and a binding -- `x: int = 1` for example -- then
+    /// we don't actually need anything here, all we'll need to validate is that our own RHS is a
+    /// valid assignment to our own annotation.
+    definitions_by_definition: FxHashMap<Definition<'db>, SymbolDefinitions>,
+
+    /// [`SymbolState`] visible at end of scope for each symbol.
+    public_symbols: IndexVec<ScopedSymbolId, SymbolState>,
 }
 
 impl<'db> UseDefMap<'db> {
-    pub(crate) fn use_definitions(&self, use_id: ScopedUseId) -> &[Definition<'db>] {
-        &self.all_definitions[self.definitions_by_use[use_id].definitions_range.clone()]
+    pub(crate) fn bindings_at_use(
+        &self,
+        use_id: ScopedUseId,
+    ) -> BindingWithConstraintsIterator<'_, 'db> {
+        self.bindings_iterator(&self.bindings_by_use[use_id])
     }
 
     pub(crate) fn use_may_be_unbound(&self, use_id: ScopedUseId) -> bool {
-        self.definitions_by_use[use_id].may_be_unbound
+        self.bindings_by_use[use_id].may_be_unbound()
     }
 
-    pub(crate) fn public_definitions(&self, symbol: ScopedSymbolId) -> &[Definition<'db>] {
-        &self.all_definitions[self.public_definitions[symbol].definitions_range.clone()]
+    pub(crate) fn public_bindings(
+        &self,
+        symbol: ScopedSymbolId,
+    ) -> BindingWithConstraintsIterator<'_, 'db> {
+        self.bindings_iterator(self.public_symbols[symbol].bindings())
     }
 
     pub(crate) fn public_may_be_unbound(&self, symbol: ScopedSymbolId) -> bool {
-        self.public_definitions[symbol].may_be_unbound
+        self.public_symbols[symbol].may_be_unbound()
     }
-}
 
-/// Definitions visible for a symbol at a particular use (or end-of-scope).
-#[derive(Clone, Debug, PartialEq, Eq)]
-struct Definitions {
-    /// [`Range`] in `all_definitions` of the visible definition IDs.
-    definitions_range: Range<usize>,
-    /// Is the symbol possibly unbound at this point?
-    may_be_unbound: bool,
-}
+    pub(crate) fn bindings_at_declaration(
+        &self,
+        declaration: Definition<'db>,
+    ) -> BindingWithConstraintsIterator<'_, 'db> {
+        if let SymbolDefinitions::Bindings(bindings) = &self.definitions_by_definition[&declaration]
+        {
+            self.bindings_iterator(bindings)
+        } else {
+            unreachable!("Declaration has non-Bindings in definitions_by_definition");
+        }
+    }
 
-impl Definitions {
-    /// The default state of a symbol is "no definitions, may be unbound", aka definitely-unbound.
-    fn unbound() -> Self {
-        Self {
-            definitions_range: Range::default(),
-            may_be_unbound: true,
+    pub(crate) fn declarations_at_binding(
+        &self,
+        binding: Definition<'db>,
+    ) -> DeclarationsIterator<'_, 'db> {
+        if let SymbolDefinitions::Declarations(declarations) =
+            &self.definitions_by_definition[&binding]
+        {
+            self.declarations_iterator(declarations)
+        } else {
+            unreachable!("Binding has non-Declarations in definitions_by_definition");
+        }
+    }
+
+    pub(crate) fn public_declarations(
+        &self,
+        symbol: ScopedSymbolId,
+    ) -> DeclarationsIterator<'_, 'db> {
+        let declarations = self.public_symbols[symbol].declarations();
+        self.declarations_iterator(declarations)
+    }
+
+    pub(crate) fn has_public_declarations(&self, symbol: ScopedSymbolId) -> bool {
+        !self.public_symbols[symbol].declarations().is_empty()
+    }
+
+    fn bindings_iterator<'a>(
+        &'a self,
+        bindings: &'a SymbolBindings,
+    ) -> BindingWithConstraintsIterator<'a, 'db> {
+        BindingWithConstraintsIterator {
+            all_definitions: &self.all_definitions,
+            all_constraints: &self.all_constraints,
+            inner: bindings.iter(),
+        }
+    }
+
+    fn declarations_iterator<'a>(
+        &'a self,
+        declarations: &'a SymbolDeclarations,
+    ) -> DeclarationsIterator<'a, 'db> {
+        DeclarationsIterator {
+            all_definitions: &self.all_definitions,
+            inner: declarations.iter(),
+            may_be_undeclared: declarations.may_be_undeclared(),
         }
     }
 }
 
-impl Default for Definitions {
-    fn default() -> Self {
-        Definitions::unbound()
-    }
-}
-
-/// A snapshot of the visible definitions for each symbol at a particular point in control flow.
-#[derive(Clone, Debug)]
-pub(super) struct FlowSnapshot {
-    definitions_by_symbol: IndexVec<ScopedSymbolId, Definitions>,
+/// Either live bindings or live declarations for a symbol.
+#[derive(Debug, PartialEq, Eq)]
+enum SymbolDefinitions {
+    Bindings(SymbolBindings),
+    Declarations(SymbolDeclarations),
 }
 
 #[derive(Debug)]
+pub(crate) struct BindingWithConstraintsIterator<'map, 'db> {
+    all_definitions: &'map IndexVec<ScopedDefinitionId, Definition<'db>>,
+    all_constraints: &'map IndexVec<ScopedConstraintId, Constraint<'db>>,
+    inner: BindingIdWithConstraintsIterator<'map>,
+}
+
+impl<'map, 'db> Iterator for BindingWithConstraintsIterator<'map, 'db> {
+    type Item = BindingWithConstraints<'map, 'db>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.inner
+            .next()
+            .map(|def_id_with_constraints| BindingWithConstraints {
+                binding: self.all_definitions[def_id_with_constraints.definition],
+                constraints: ConstraintsIterator {
+                    all_constraints: self.all_constraints,
+                    constraint_ids: def_id_with_constraints.constraint_ids,
+                },
+            })
+    }
+}
+
+impl std::iter::FusedIterator for BindingWithConstraintsIterator<'_, '_> {}
+
+pub(crate) struct BindingWithConstraints<'map, 'db> {
+    pub(crate) binding: Definition<'db>,
+    pub(crate) constraints: ConstraintsIterator<'map, 'db>,
+}
+
+pub(crate) struct ConstraintsIterator<'map, 'db> {
+    all_constraints: &'map IndexVec<ScopedConstraintId, Constraint<'db>>,
+    constraint_ids: ConstraintIdIterator<'map>,
+}
+
+impl<'map, 'db> Iterator for ConstraintsIterator<'map, 'db> {
+    type Item = Constraint<'db>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.constraint_ids
+            .next()
+            .map(|constraint_id| self.all_constraints[constraint_id])
+    }
+}
+
+impl std::iter::FusedIterator for ConstraintsIterator<'_, '_> {}
+
+pub(crate) struct DeclarationsIterator<'map, 'db> {
+    all_definitions: &'map IndexVec<ScopedDefinitionId, Definition<'db>>,
+    inner: DeclarationIdIterator<'map>,
+    may_be_undeclared: bool,
+}
+
+impl DeclarationsIterator<'_, '_> {
+    pub(crate) fn may_be_undeclared(&self) -> bool {
+        self.may_be_undeclared
+    }
+}
+
+impl<'map, 'db> Iterator for DeclarationsIterator<'map, 'db> {
+    type Item = Definition<'db>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.inner.next().map(|def_id| self.all_definitions[def_id])
+    }
+}
+
+impl std::iter::FusedIterator for DeclarationsIterator<'_, '_> {}
+
+/// A snapshot of the definitions and constraints state at a particular point in control flow.
+#[derive(Clone, Debug)]
+pub(super) struct FlowSnapshot {
+    symbol_states: IndexVec<ScopedSymbolId, SymbolState>,
+}
+
+#[derive(Debug, Default)]
 pub(super) struct UseDefMapBuilder<'db> {
-    /// Definition IDs array for `definitions_by_use` and `definitions_by_symbol` to slice into.
-    all_definitions: Vec<Definition<'db>>,
+    /// Append-only array of [`Definition`].
+    all_definitions: IndexVec<ScopedDefinitionId, Definition<'db>>,
 
-    /// Visible definitions at each so-far-recorded use.
-    definitions_by_use: IndexVec<ScopedUseId, Definitions>,
+    /// Append-only array of [`Constraint`].
+    all_constraints: IndexVec<ScopedConstraintId, Constraint<'db>>,
 
-    /// Currently visible definitions for each symbol.
-    definitions_by_symbol: IndexVec<ScopedSymbolId, Definitions>,
+    /// Live bindings at each so-far-recorded use.
+    bindings_by_use: IndexVec<ScopedUseId, SymbolBindings>,
+
+    /// Live bindings or declarations for each so-far-recorded definition.
+    definitions_by_definition: FxHashMap<Definition<'db>, SymbolDefinitions>,
+
+    /// Currently live bindings and declarations for each symbol.
+    symbol_states: IndexVec<ScopedSymbolId, SymbolState>,
 }
 
 impl<'db> UseDefMapBuilder<'db> {
     pub(super) fn new() -> Self {
-        Self {
-            all_definitions: Vec::new(),
-            definitions_by_use: IndexVec::new(),
-            definitions_by_symbol: IndexVec::new(),
-        }
+        Self::default()
     }
 
     pub(super) fn add_symbol(&mut self, symbol: ScopedSymbolId) {
-        let new_symbol = self.definitions_by_symbol.push(Definitions::unbound());
+        let new_symbol = self.symbol_states.push(SymbolState::undefined());
         debug_assert_eq!(symbol, new_symbol);
     }
 
-    pub(super) fn record_definition(
+    pub(super) fn record_binding(&mut self, symbol: ScopedSymbolId, binding: Definition<'db>) {
+        let def_id = self.all_definitions.push(binding);
+        let symbol_state = &mut self.symbol_states[symbol];
+        self.definitions_by_definition.insert(
+            binding,
+            SymbolDefinitions::Declarations(symbol_state.declarations().clone()),
+        );
+        symbol_state.record_binding(def_id);
+    }
+
+    pub(super) fn record_constraint(&mut self, constraint: Constraint<'db>) {
+        let constraint_id = self.all_constraints.push(constraint);
+        for state in &mut self.symbol_states {
+            state.record_constraint(constraint_id);
+        }
+    }
+
+    pub(super) fn record_declaration(
+        &mut self,
+        symbol: ScopedSymbolId,
+        declaration: Definition<'db>,
+    ) {
+        let def_id = self.all_definitions.push(declaration);
+        let symbol_state = &mut self.symbol_states[symbol];
+        self.definitions_by_definition.insert(
+            declaration,
+            SymbolDefinitions::Bindings(symbol_state.bindings().clone()),
+        );
+        symbol_state.record_declaration(def_id);
+    }
+
+    pub(super) fn record_declaration_and_binding(
         &mut self,
         symbol: ScopedSymbolId,
         definition: Definition<'db>,
     ) {
-        // We have a new definition of a symbol; this replaces any previous definitions in this
-        // path.
-        let def_idx = self.all_definitions.len();
-        self.all_definitions.push(definition);
-        self.definitions_by_symbol[symbol] = Definitions {
-            #[allow(clippy::range_plus_one)]
-            definitions_range: def_idx..(def_idx + 1),
-            may_be_unbound: false,
-        };
+        // We don't need to store anything in self.definitions_by_definition.
+        let def_id = self.all_definitions.push(definition);
+        let symbol_state = &mut self.symbol_states[symbol];
+        symbol_state.record_declaration(def_id);
+        symbol_state.record_binding(def_id);
     }
 
     pub(super) fn record_use(&mut self, symbol: ScopedSymbolId, use_id: ScopedUseId) {
-        // We have a use of a symbol; clone the currently visible definitions for that symbol, and
-        // record them as the visible definitions for this use.
+        // We have a use of a symbol; clone the current bindings for that symbol, and record them
+        // as the live bindings for this use.
         let new_use = self
-            .definitions_by_use
-            .push(self.definitions_by_symbol[symbol].clone());
+            .bindings_by_use
+            .push(self.symbol_states[symbol].bindings().clone());
         debug_assert_eq!(use_id, new_use);
     }
 
     /// Take a snapshot of the current visible-symbols state.
     pub(super) fn snapshot(&self) -> FlowSnapshot {
         FlowSnapshot {
-            definitions_by_symbol: self.definitions_by_symbol.clone(),
+            symbol_states: self.symbol_states.clone(),
         }
     }
 
-    /// Restore the current builder visible-definitions state to the given snapshot.
+    /// Restore the current builder symbols state to the given snapshot.
     pub(super) fn restore(&mut self, snapshot: FlowSnapshot) {
-        // We never remove symbols from `definitions_by_symbol` (it's an IndexVec, and the symbol
+        // We never remove symbols from `symbol_states` (it's an IndexVec, and the symbol
         // IDs must line up), so the current number of known symbols must always be equal to or
         // greater than the number of known symbols in a previously-taken snapshot.
-        let num_symbols = self.definitions_by_symbol.len();
-        debug_assert!(num_symbols >= snapshot.definitions_by_symbol.len());
+        let num_symbols = self.symbol_states.len();
+        debug_assert!(num_symbols >= snapshot.symbol_states.len());
 
         // Restore the current visible-definitions state to the given snapshot.
-        self.definitions_by_symbol = snapshot.definitions_by_symbol;
+        self.symbol_states = snapshot.symbol_states;
 
         // If the snapshot we are restoring is missing some symbols we've recorded since, we need
         // to fill them in so the symbol IDs continue to line up. Since they don't exist in the
-        // snapshot, the correct state to fill them in with is "unbound", the default.
-        self.definitions_by_symbol
-            .resize(num_symbols, Definitions::unbound());
+        // snapshot, the correct state to fill them in with is "undefined".
+        self.symbol_states
+            .resize(num_symbols, SymbolState::undefined());
     }
 
     /// Merge the given snapshot into the current state, reflecting that we might have taken either
-    /// path to get here. The new visible-definitions state for each symbol should include
-    /// definitions from both the prior state and the snapshot.
-    pub(super) fn merge(&mut self, snapshot: &FlowSnapshot) {
-        // The tricky thing about merging two Ranges pointing into `all_definitions` is that if the
-        // two Ranges aren't already adjacent in `all_definitions`, we will have to copy at least
-        // one or the other of the ranges to the end of `all_definitions` so as to make them
-        // adjacent. We can't ever move things around in `all_definitions` because previously
-        // recorded uses may still have ranges pointing to any part of it; all we can do is append.
-        // It's possible we may end up with some old entries in `all_definitions` that nobody is
-        // pointing to, but that's OK.
-
-        // We never remove symbols from `definitions_by_symbol` (it's an IndexVec, and the symbol
+    /// path to get here. The new state for each symbol should include definitions from both the
+    /// prior state and the snapshot.
+    pub(super) fn merge(&mut self, snapshot: FlowSnapshot) {
+        // We never remove symbols from `symbol_states` (it's an IndexVec, and the symbol
         // IDs must line up), so the current number of known symbols must always be equal to or
         // greater than the number of known symbols in a previously-taken snapshot.
-        debug_assert!(self.definitions_by_symbol.len() >= snapshot.definitions_by_symbol.len());
+        debug_assert!(self.symbol_states.len() >= snapshot.symbol_states.len());
 
-        for (symbol_id, current) in self.definitions_by_symbol.iter_mut_enumerated() {
-            let Some(snapshot) = snapshot.definitions_by_symbol.get(symbol_id) else {
-                // Symbol not present in snapshot, so it's unbound from that path.
-                current.may_be_unbound = true;
-                continue;
-            };
-
-            // If the symbol can be unbound in either predecessor, it can be unbound post-merge.
-            current.may_be_unbound |= snapshot.may_be_unbound;
-
-            // Merge the definition ranges.
-            let current = &mut current.definitions_range;
-            let snapshot = &snapshot.definitions_range;
-
-            // We never create reversed ranges.
-            debug_assert!(current.end >= current.start);
-            debug_assert!(snapshot.end >= snapshot.start);
-
-            if current == snapshot {
-                // Ranges already identical, nothing to do.
-            } else if snapshot.is_empty() {
-                // Merging from an empty range; nothing to do.
-            } else if (*current).is_empty() {
-                // Merging to an empty range; just use the incoming range.
-                *current = snapshot.clone();
-            } else if snapshot.end >= current.start && snapshot.start <= current.end {
-                // Ranges are adjacent or overlapping, merge them in-place.
-                *current = current.start.min(snapshot.start)..current.end.max(snapshot.end);
-            } else if current.end == self.all_definitions.len() {
-                // Ranges are not adjacent or overlapping, `current` is at the end of
-                // `all_definitions`, we need to copy `snapshot` to the end so they are adjacent
-                // and can be merged into one range.
-                self.all_definitions.extend_from_within(snapshot.clone());
-                current.end = self.all_definitions.len();
-            } else if snapshot.end == self.all_definitions.len() {
-                // Ranges are not adjacent or overlapping, `snapshot` is at the end of
-                // `all_definitions`, we need to copy `current` to the end so they are adjacent and
-                // can be merged into one range.
-                self.all_definitions.extend_from_within(current.clone());
-                current.start = snapshot.start;
-                current.end = self.all_definitions.len();
+        let mut snapshot_definitions_iter = snapshot.symbol_states.into_iter();
+        for current in &mut self.symbol_states {
+            if let Some(snapshot) = snapshot_definitions_iter.next() {
+                current.merge(snapshot);
             } else {
-                // Ranges are not adjacent and neither one is at the end of `all_definitions`, we
-                // have to copy both to the end so they are adjacent and we can merge them.
-                let start = self.all_definitions.len();
-                self.all_definitions.extend_from_within(current.clone());
-                self.all_definitions.extend_from_within(snapshot.clone());
-                current.start = start;
-                current.end = self.all_definitions.len();
+                // Symbol not present in snapshot, so it's unbound/undeclared from that path.
+                current.set_may_be_unbound();
+                current.set_may_be_undeclared();
             }
         }
     }
 
     pub(super) fn finish(mut self) -> UseDefMap<'db> {
         self.all_definitions.shrink_to_fit();
-        self.definitions_by_symbol.shrink_to_fit();
-        self.definitions_by_use.shrink_to_fit();
+        self.all_constraints.shrink_to_fit();
+        self.symbol_states.shrink_to_fit();
+        self.bindings_by_use.shrink_to_fit();
+        self.definitions_by_definition.shrink_to_fit();
 
         UseDefMap {
             all_definitions: self.all_definitions,
-            definitions_by_use: self.definitions_by_use,
-            public_definitions: self.definitions_by_symbol,
+            all_constraints: self.all_constraints,
+            bindings_by_use: self.bindings_by_use,
+            public_symbols: self.symbol_states,
+            definitions_by_definition: self.definitions_by_definition,
         }
     }
 }

crates/red_knot_python_semantic/src/semantic_index/use_def/bitset.rs

@@ -0,0 +1,309 @@
+/// Ordered set of `u32`.
+///
+/// Uses an inline bit-set for small values (up to 64 * B), falls back to heap allocated vector of
+/// blocks for larger values.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub(super) enum BitSet<const B: usize> {
+    /// Bit-set (in 64-bit blocks) for the first 64 * B entries.
+    Inline([u64; B]),
+
+    /// Overflow beyond 64 * B.
+    Heap(Vec<u64>),
+}
+
+impl<const B: usize> Default for BitSet<B> {
+    fn default() -> Self {
+        // B * 64 must fit in a u32, or else we have unusable bits; this assertion makes the
+        // truncating casts to u32 below safe. This would be better as a const assertion, but
+        // that's not possible on stable with const generic params. (B should never really be
+        // anywhere close to this large.)
+        assert!(B * 64 < (u32::MAX as usize));
+        // This implementation requires usize >= 32 bits.
+        static_assertions::const_assert!(usize::BITS >= 32);
+        Self::Inline([0; B])
+    }
+}
+
+impl<const B: usize> BitSet<B> {
+    /// Create and return a new [`BitSet`] with a single `value` inserted.
+    pub(super) fn with(value: u32) -> Self {
+        let mut bitset = Self::default();
+        bitset.insert(value);
+        bitset
+    }
+
+    pub(super) fn is_empty(&self) -> bool {
+        self.blocks().iter().all(|&b| b == 0)
+    }
+
+    /// Convert from Inline to Heap, if needed, and resize the Heap vector, if needed.
+    fn resize(&mut self, value: u32) {
+        let num_blocks_needed = (value / 64) + 1;
+        self.resize_blocks(num_blocks_needed as usize);
+    }
+
+    fn resize_blocks(&mut self, num_blocks_needed: usize) {
+        match self {
+            Self::Inline(blocks) => {
+                let mut vec = blocks.to_vec();
+                vec.resize(num_blocks_needed, 0);
+                *self = Self::Heap(vec);
+            }
+            Self::Heap(vec) => {
+                vec.resize(num_blocks_needed, 0);
+            }
+        }
+    }
+
+    fn blocks_mut(&mut self) -> &mut [u64] {
+        match self {
+            Self::Inline(blocks) => blocks.as_mut_slice(),
+            Self::Heap(blocks) => blocks.as_mut_slice(),
+        }
+    }
+
+    fn blocks(&self) -> &[u64] {
+        match self {
+            Self::Inline(blocks) => blocks.as_slice(),
+            Self::Heap(blocks) => blocks.as_slice(),
+        }
+    }
+
+    /// Insert a value into the [`BitSet`].
+    ///
+    /// Return true if the value was newly inserted, false if already present.
+    pub(super) fn insert(&mut self, value: u32) -> bool {
+        let value_usize = value as usize;
+        let (block, index) = (value_usize / 64, value_usize % 64);
+        if block >= self.blocks().len() {
+            self.resize(value);
+        }
+        let blocks = self.blocks_mut();
+        let missing = blocks[block] & (1 << index) == 0;
+        blocks[block] |= 1 << index;
+        missing
+    }
+
+    /// Intersect in-place with another [`BitSet`].
+    pub(super) fn intersect(&mut self, other: &BitSet<B>) {
+        let my_blocks = self.blocks_mut();
+        let other_blocks = other.blocks();
+        let min_len = my_blocks.len().min(other_blocks.len());
+        for i in 0..min_len {
+            my_blocks[i] &= other_blocks[i];
+        }
+        for block in my_blocks.iter_mut().skip(min_len) {
+            *block = 0;
+        }
+    }
+
+    /// Union in-place with another [`BitSet`].
+    pub(super) fn union(&mut self, other: &BitSet<B>) {
+        let mut max_len = self.blocks().len();
+        let other_len = other.blocks().len();
+        if other_len > max_len {
+            max_len = other_len;
+            self.resize_blocks(max_len);
+        }
+        for (my_block, other_block) in self.blocks_mut().iter_mut().zip(other.blocks()) {
+            *my_block |= other_block;
+        }
+    }
+
+    /// Return an iterator over the values (in ascending order) in this [`BitSet`].
+    pub(super) fn iter(&self) -> BitSetIterator<'_, B> {
+        let blocks = self.blocks();
+        BitSetIterator {
+            blocks,
+            current_block_index: 0,
+            current_block: blocks[0],
+        }
+    }
+}
+
+/// Iterator over values in a [`BitSet`].
+#[derive(Debug)]
+pub(super) struct BitSetIterator<'a, const B: usize> {
+    /// The blocks we are iterating over.
+    blocks: &'a [u64],
+
+    /// The index of the block we are currently iterating through.
+    current_block_index: usize,
+
+    /// The block we are currently iterating through (and zeroing as we go.)
+    current_block: u64,
+}
+
+impl<const B: usize> Iterator for BitSetIterator<'_, B> {
+    type Item = u32;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        while self.current_block == 0 {
+            if self.current_block_index + 1 >= self.blocks.len() {
+                return None;
+            }
+            self.current_block_index += 1;
+            self.current_block = self.blocks[self.current_block_index];
+        }
+        let lowest_bit_set = self.current_block.trailing_zeros();
+        // reset the lowest set bit, without a data dependency on `lowest_bit_set`
+        self.current_block &= self.current_block.wrapping_sub(1);
+        // SAFETY: `lowest_bit_set` cannot be more than 64, `current_block_index` cannot be more
+        // than `B - 1`, and we check above that `B * 64 < u32::MAX`. So both `64 *
+        // current_block_index` and the final value here must fit in u32.
+        #[allow(clippy::cast_possible_truncation)]
+        Some(lowest_bit_set + (64 * self.current_block_index) as u32)
+    }
+}
+
+impl<const B: usize> std::iter::FusedIterator for BitSetIterator<'_, B> {}
+
+#[cfg(test)]
+mod tests {
+    use super::BitSet;
+
+    fn assert_bitset<const B: usize>(bitset: &BitSet<B>, contents: &[u32]) {
+        assert_eq!(bitset.iter().collect::<Vec<_>>(), contents);
+    }
+
+    #[test]
+    fn iter() {
+        let mut b = BitSet::<1>::with(3);
+        b.insert(27);
+        b.insert(6);
+        assert!(matches!(b, BitSet::Inline(_)));
+        assert_bitset(&b, &[3, 6, 27]);
+    }
+
+    #[test]
+    fn iter_overflow() {
+        let mut b = BitSet::<1>::with(140);
+        b.insert(100);
+        b.insert(129);
+        assert!(matches!(b, BitSet::Heap(_)));
+        assert_bitset(&b, &[100, 129, 140]);
+    }
+
+    #[test]
+    fn intersect() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(23);
+        b2.insert(5);
+
+        b1.intersect(&b2);
+        assert_bitset(&b1, &[4]);
+    }
+
+    #[test]
+    fn intersect_mixed_1() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(89);
+        b2.insert(5);
+
+        b1.intersect(&b2);
+        assert_bitset(&b1, &[4]);
+    }
+
+    #[test]
+    fn intersect_mixed_2() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(23);
+        b2.insert(89);
+
+        b1.intersect(&b2);
+        assert_bitset(&b1, &[4]);
+    }
+
+    #[test]
+    fn intersect_heap() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(89);
+        b2.insert(90);
+
+        b1.intersect(&b2);
+        assert_bitset(&b1, &[4]);
+    }
+
+    #[test]
+    fn intersect_heap_2() {
+        let mut b1 = BitSet::<1>::with(89);
+        let mut b2 = BitSet::<1>::with(89);
+        b1.insert(91);
+        b2.insert(90);
+
+        b1.intersect(&b2);
+        assert_bitset(&b1, &[89]);
+    }
+
+    #[test]
+    fn union() {
+        let mut b1 = BitSet::<1>::with(2);
+        let b2 = BitSet::<1>::with(4);
+
+        b1.union(&b2);
+        assert_bitset(&b1, &[2, 4]);
+    }
+
+    #[test]
+    fn union_mixed_1() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(89);
+        b2.insert(5);
+
+        b1.union(&b2);
+        assert_bitset(&b1, &[4, 5, 89]);
+    }
+
+    #[test]
+    fn union_mixed_2() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(23);
+        b2.insert(89);
+
+        b1.union(&b2);
+        assert_bitset(&b1, &[4, 23, 89]);
+    }
+
+    #[test]
+    fn union_heap() {
+        let mut b1 = BitSet::<1>::with(4);
+        let mut b2 = BitSet::<1>::with(4);
+        b1.insert(89);
+        b2.insert(90);
+
+        b1.union(&b2);
+        assert_bitset(&b1, &[4, 89, 90]);
+    }
+
+    #[test]
+    fn union_heap_2() {
+        let mut b1 = BitSet::<1>::with(89);
+        let mut b2 = BitSet::<1>::with(89);
+        b1.insert(91);
+        b2.insert(90);
+
+        b1.union(&b2);
+        assert_bitset(&b1, &[89, 90, 91]);
+    }
+
+    #[test]
+    fn multiple_blocks() {
+        let mut b = BitSet::<2>::with(120);
+        b.insert(45);
+        assert!(matches!(b, BitSet::Inline(_)));
+        assert_bitset(&b, &[45, 120]);
+    }
+
+    #[test]
+    fn empty() {
+        let b = BitSet::<1>::default();
+
+        assert!(b.is_empty());
+    }
+}

crates/red_knot_python_semantic/src/semantic_index/use_def/symbol_state.rs

@@ -0,0 +1,588 @@
+//! Track live bindings per symbol, applicable constraints per binding, and live declarations.
+//!
+//! These data structures operate entirely on scope-local newtype-indices for definitions and
+//! constraints, referring to their location in the `all_definitions` and `all_constraints`
+//! indexvecs in [`super::UseDefMapBuilder`].
+//!
+//! We need to track arbitrary associations between bindings and constraints, not just a single set
+//! of currently dominating constraints (where "dominating" means "control flow must have passed
+//! through it to reach this point"), because we can have dominating constraints that apply to some
+//! bindings but not others, as in this code:
+//!
+//! ```python
+//! x = 1 if flag else None
+//! if x is not None:
+//!     if flag2:
+//!         x = 2 if flag else None
+//!     x
+//! ```
+//!
+//! The `x is not None` constraint dominates the final use of `x`, but it applies only to the first
+//! binding of `x`, not the second, so `None` is a possible value for `x`.
+//!
+//! And we can't just track, for each binding, an index into a list of dominating constraints,
+//! either, because we can have bindings which are still visible, but subject to constraints that
+//! are no longer dominating, as in this code:
+//!
+//! ```python
+//! x = 0
+//! if flag1:
+//!     x = 1 if flag2 else None
+//!     assert x is not None
+//! x
+//! ```
+//!
+//! From the point of view of the final use of `x`, the `x is not None` constraint no longer
+//! dominates, but it does dominate the `x = 1 if flag2 else None` binding, so we have to keep
+//! track of that.
+//!
+//! The data structures used here ([`BitSet`] and [`smallvec::SmallVec`]) optimize for keeping all
+//! data inline (avoiding lots of scattered allocations) in small-to-medium cases, and falling back
+//! to heap allocation to be able to scale to arbitrary numbers of live bindings and constraints
+//! when needed.
+//!
+//! Tracking live declarations is simpler, since constraints are not involved, but otherwise very
+//! similar to tracking live bindings.
+use super::bitset::{BitSet, BitSetIterator};
+use ruff_index::newtype_index;
+use smallvec::SmallVec;
+
+/// A newtype-index for a definition in a particular scope.
+#[newtype_index]
+pub(super) struct ScopedDefinitionId;
+
+/// A newtype-index for a constraint expression in a particular scope.
+#[newtype_index]
+pub(super) struct ScopedConstraintId;
+
+/// Can reference this * 64 total definitions inline; more will fall back to the heap.
+const INLINE_BINDING_BLOCKS: usize = 3;
+
+/// A [`BitSet`] of [`ScopedDefinitionId`], representing live bindings of a symbol in a scope.
+type Bindings = BitSet<INLINE_BINDING_BLOCKS>;
+type BindingsIterator<'a> = BitSetIterator<'a, INLINE_BINDING_BLOCKS>;
+
+/// Can reference this * 64 total declarations inline; more will fall back to the heap.
+const INLINE_DECLARATION_BLOCKS: usize = 3;
+
+/// A [`BitSet`] of [`ScopedDefinitionId`], representing live declarations of a symbol in a scope.
+type Declarations = BitSet<INLINE_DECLARATION_BLOCKS>;
+type DeclarationsIterator<'a> = BitSetIterator<'a, INLINE_DECLARATION_BLOCKS>;
+
+/// Can reference this * 64 total constraints inline; more will fall back to the heap.
+const INLINE_CONSTRAINT_BLOCKS: usize = 2;
+
+/// Can keep inline this many live bindings per symbol at a given time; more will go to heap.
+const INLINE_BINDINGS_PER_SYMBOL: usize = 4;
+
+/// One [`BitSet`] of applicable [`ScopedConstraintId`] per live binding.
+type InlineConstraintArray = [BitSet<INLINE_CONSTRAINT_BLOCKS>; INLINE_BINDINGS_PER_SYMBOL];
+type Constraints = SmallVec<InlineConstraintArray>;
+type ConstraintsIterator<'a> = std::slice::Iter<'a, BitSet<INLINE_CONSTRAINT_BLOCKS>>;
+type ConstraintsIntoIterator = smallvec::IntoIter<InlineConstraintArray>;
+
+/// Live declarations for a single symbol at some point in control flow.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) struct SymbolDeclarations {
+    /// [`BitSet`]: which declarations (as [`ScopedDefinitionId`]) can reach the current location?
+    live_declarations: Declarations,
+
+    /// Could the symbol be un-declared at this point?
+    may_be_undeclared: bool,
+}
+
+impl SymbolDeclarations {
+    fn undeclared() -> Self {
+        Self {
+            live_declarations: Declarations::default(),
+            may_be_undeclared: true,
+        }
+    }
+
+    /// Record a newly-encountered declaration for this symbol.
+    fn record_declaration(&mut self, declaration_id: ScopedDefinitionId) {
+        self.live_declarations = Declarations::with(declaration_id.into());
+        self.may_be_undeclared = false;
+    }
+
+    /// Add undeclared as a possibility for this symbol.
+    fn set_may_be_undeclared(&mut self) {
+        self.may_be_undeclared = true;
+    }
+
+    /// Return an iterator over live declarations for this symbol.
+    pub(super) fn iter(&self) -> DeclarationIdIterator {
+        DeclarationIdIterator {
+            inner: self.live_declarations.iter(),
+        }
+    }
+
+    pub(super) fn is_empty(&self) -> bool {
+        self.live_declarations.is_empty()
+    }
+
+    pub(super) fn may_be_undeclared(&self) -> bool {
+        self.may_be_undeclared
+    }
+}
+
+/// Live bindings and narrowing constraints for a single symbol at some point in control flow.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) struct SymbolBindings {
+    /// [`BitSet`]: which bindings (as [`ScopedDefinitionId`]) can reach the current location?
+    live_bindings: Bindings,
+
+    /// For each live binding, which [`ScopedConstraintId`] apply?
+    ///
+    /// This is a [`smallvec::SmallVec`] which should always have one [`BitSet`] of constraints per
+    /// binding in `live_bindings`.
+    constraints: Constraints,
+
+    /// Could the symbol be unbound at this point?
+    may_be_unbound: bool,
+}
+
+impl SymbolBindings {
+    fn unbound() -> Self {
+        Self {
+            live_bindings: Bindings::default(),
+            constraints: Constraints::default(),
+            may_be_unbound: true,
+        }
+    }
+
+    /// Add Unbound as a possibility for this symbol.
+    fn set_may_be_unbound(&mut self) {
+        self.may_be_unbound = true;
+    }
+
+    /// Record a newly-encountered binding for this symbol.
+    pub(super) fn record_binding(&mut self, binding_id: ScopedDefinitionId) {
+        // The new binding replaces all previous live bindings in this path, and has no
+        // constraints.
+        self.live_bindings = Bindings::with(binding_id.into());
+        self.constraints = Constraints::with_capacity(1);
+        self.constraints.push(BitSet::default());
+        self.may_be_unbound = false;
+    }
+
+    /// Add given constraint to all live bindings.
+    pub(super) fn record_constraint(&mut self, constraint_id: ScopedConstraintId) {
+        for bitset in &mut self.constraints {
+            bitset.insert(constraint_id.into());
+        }
+    }
+
+    /// Iterate over currently live bindings for this symbol.
+    pub(super) fn iter(&self) -> BindingIdWithConstraintsIterator {
+        BindingIdWithConstraintsIterator {
+            definitions: self.live_bindings.iter(),
+            constraints: self.constraints.iter(),
+        }
+    }
+
+    pub(super) fn may_be_unbound(&self) -> bool {
+        self.may_be_unbound
+    }
+}
+
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) struct SymbolState {
+    declarations: SymbolDeclarations,
+    bindings: SymbolBindings,
+}
+
+impl SymbolState {
+    /// Return a new [`SymbolState`] representing an unbound, undeclared symbol.
+    pub(super) fn undefined() -> Self {
+        Self {
+            declarations: SymbolDeclarations::undeclared(),
+            bindings: SymbolBindings::unbound(),
+        }
+    }
+
+    /// Add Unbound as a possibility for this symbol.
+    pub(super) fn set_may_be_unbound(&mut self) {
+        self.bindings.set_may_be_unbound();
+    }
+
+    /// Record a newly-encountered binding for this symbol.
+    pub(super) fn record_binding(&mut self, binding_id: ScopedDefinitionId) {
+        self.bindings.record_binding(binding_id);
+    }
+
+    /// Add given constraint to all live bindings.
+    pub(super) fn record_constraint(&mut self, constraint_id: ScopedConstraintId) {
+        self.bindings.record_constraint(constraint_id);
+    }
+
+    /// Add undeclared as a possibility for this symbol.
+    pub(super) fn set_may_be_undeclared(&mut self) {
+        self.declarations.set_may_be_undeclared();
+    }
+
+    /// Record a newly-encountered declaration of this symbol.
+    pub(super) fn record_declaration(&mut self, declaration_id: ScopedDefinitionId) {
+        self.declarations.record_declaration(declaration_id);
+    }
+
+    /// Merge another [`SymbolState`] into this one.
+    pub(super) fn merge(&mut self, b: SymbolState) {
+        let mut a = Self {
+            bindings: SymbolBindings {
+                live_bindings: Bindings::default(),
+                constraints: Constraints::default(),
+                may_be_unbound: self.bindings.may_be_unbound || b.bindings.may_be_unbound,
+            },
+            declarations: SymbolDeclarations {
+                live_declarations: self.declarations.live_declarations.clone(),
+                may_be_undeclared: self.declarations.may_be_undeclared
+                    || b.declarations.may_be_undeclared,
+            },
+        };
+
+        std::mem::swap(&mut a, self);
+        self.declarations
+            .live_declarations
+            .union(&b.declarations.live_declarations);
+
+        let mut a_defs_iter = a.bindings.live_bindings.iter();
+        let mut b_defs_iter = b.bindings.live_bindings.iter();
+        let mut a_constraints_iter = a.bindings.constraints.into_iter();
+        let mut b_constraints_iter = b.bindings.constraints.into_iter();
+
+        let mut opt_a_def: Option<u32> = a_defs_iter.next();
+        let mut opt_b_def: Option<u32> = b_defs_iter.next();
+
+        // Iterate through the definitions from `a` and `b`, always processing the lower definition
+        // ID first, and pushing each definition onto the merged `SymbolState` with its
+        // constraints. If a definition is found in both `a` and `b`, push it with the intersection
+        // of the constraints from the two paths; a constraint that applies from only one possible
+        // path is irrelevant.
+
+        // Helper to push `def`, with constraints in `constraints_iter`, onto `self`.
+        let push = |def, constraints_iter: &mut ConstraintsIntoIterator, merged: &mut Self| {
+            merged.bindings.live_bindings.insert(def);
+            // SAFETY: we only ever create SymbolState with either no definitions and no constraint
+            // bitsets (`::unbound`) or one definition and one constraint bitset (`::with`), and
+            // `::merge` always pushes one definition and one constraint bitset together (just
+            // below), so the number of definitions and the number of constraint bitsets can never
+            // get out of sync.
+            let constraints = constraints_iter
+                .next()
+                .expect("definitions and constraints length mismatch");
+            merged.bindings.constraints.push(constraints);
+        };
+
+        loop {
+            match (opt_a_def, opt_b_def) {
+                (Some(a_def), Some(b_def)) => match a_def.cmp(&b_def) {
+                    std::cmp::Ordering::Less => {
+                        // Next definition ID is only in `a`, push it to `self` and advance `a`.
+                        push(a_def, &mut a_constraints_iter, self);
+                        opt_a_def = a_defs_iter.next();
+                    }
+                    std::cmp::Ordering::Greater => {
+                        // Next definition ID is only in `b`, push it to `self` and advance `b`.
+                        push(b_def, &mut b_constraints_iter, self);
+                        opt_b_def = b_defs_iter.next();
+                    }
+                    std::cmp::Ordering::Equal => {
+                        // Next definition is in both; push to `self` and intersect constraints.
+                        push(a_def, &mut b_constraints_iter, self);
+                        // SAFETY: we only ever create SymbolState with either no definitions and
+                        // no constraint bitsets (`::unbound`) or one definition and one constraint
+                        // bitset (`::with`), and `::merge` always pushes one definition and one
+                        // constraint bitset together (just below), so the number of definitions
+                        // and the number of constraint bitsets can never get out of sync.
+                        let a_constraints = a_constraints_iter
+                            .next()
+                            .expect("definitions and constraints length mismatch");
+                        // If the same definition is visible through both paths, any constraint
+                        // that applies on only one path is irrelevant to the resulting type from
+                        // unioning the two paths, so we intersect the constraints.
+                        self.bindings
+                            .constraints
+                            .last_mut()
+                            .unwrap()
+                            .intersect(&a_constraints);
+                        opt_a_def = a_defs_iter.next();
+                        opt_b_def = b_defs_iter.next();
+                    }
+                },
+                (Some(a_def), None) => {
+                    // We've exhausted `b`, just push the def from `a` and move on to the next.
+                    push(a_def, &mut a_constraints_iter, self);
+                    opt_a_def = a_defs_iter.next();
+                }
+                (None, Some(b_def)) => {
+                    // We've exhausted `a`, just push the def from `b` and move on to the next.
+                    push(b_def, &mut b_constraints_iter, self);
+                    opt_b_def = b_defs_iter.next();
+                }
+                (None, None) => break,
+            }
+        }
+    }
+
+    pub(super) fn bindings(&self) -> &SymbolBindings {
+        &self.bindings
+    }
+
+    pub(super) fn declarations(&self) -> &SymbolDeclarations {
+        &self.declarations
+    }
+
+    /// Could the symbol be unbound?
+    pub(super) fn may_be_unbound(&self) -> bool {
+        self.bindings.may_be_unbound()
+    }
+}
+
+/// The default state of a symbol, if we've seen no definitions of it, is undefined (that is,
+/// both unbound and undeclared).
+impl Default for SymbolState {
+    fn default() -> Self {
+        SymbolState::undefined()
+    }
+}
+
+/// A single binding (as [`ScopedDefinitionId`]) with an iterator of its applicable
+/// [`ScopedConstraintId`].
+#[derive(Debug)]
+pub(super) struct BindingIdWithConstraints<'a> {
+    pub(super) definition: ScopedDefinitionId,
+    pub(super) constraint_ids: ConstraintIdIterator<'a>,
+}
+
+#[derive(Debug)]
+pub(super) struct BindingIdWithConstraintsIterator<'a> {
+    definitions: BindingsIterator<'a>,
+    constraints: ConstraintsIterator<'a>,
+}
+
+impl<'a> Iterator for BindingIdWithConstraintsIterator<'a> {
+    type Item = BindingIdWithConstraints<'a>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match (self.definitions.next(), self.constraints.next()) {
+            (None, None) => None,
+            (Some(def), Some(constraints)) => Some(BindingIdWithConstraints {
+                definition: ScopedDefinitionId::from_u32(def),
+                constraint_ids: ConstraintIdIterator {
+                    wrapped: constraints.iter(),
+                },
+            }),
+            // SAFETY: see above.
+            _ => unreachable!("definitions and constraints length mismatch"),
+        }
+    }
+}
+
+impl std::iter::FusedIterator for BindingIdWithConstraintsIterator<'_> {}
+
+#[derive(Debug)]
+pub(super) struct ConstraintIdIterator<'a> {
+    wrapped: BitSetIterator<'a, INLINE_CONSTRAINT_BLOCKS>,
+}
+
+impl Iterator for ConstraintIdIterator<'_> {
+    type Item = ScopedConstraintId;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.wrapped.next().map(ScopedConstraintId::from_u32)
+    }
+}
+
+impl std::iter::FusedIterator for ConstraintIdIterator<'_> {}
+
+#[derive(Debug)]
+pub(super) struct DeclarationIdIterator<'a> {
+    inner: DeclarationsIterator<'a>,
+}
+
+impl<'a> Iterator for DeclarationIdIterator<'a> {
+    type Item = ScopedDefinitionId;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.inner.next().map(ScopedDefinitionId::from_u32)
+    }
+}
+
+impl std::iter::FusedIterator for DeclarationIdIterator<'_> {}
+
+#[cfg(test)]
+mod tests {
+    use super::{ScopedConstraintId, ScopedDefinitionId, SymbolState};
+
+    fn assert_bindings(symbol: &SymbolState, may_be_unbound: bool, expected: &[&str]) {
+        assert_eq!(symbol.may_be_unbound(), may_be_unbound);
+        let actual = symbol
+            .bindings()
+            .iter()
+            .map(|def_id_with_constraints| {
+                format!(
+                    "{}<{}>",
+                    def_id_with_constraints.definition.as_u32(),
+                    def_id_with_constraints
+                        .constraint_ids
+                        .map(ScopedConstraintId::as_u32)
+                        .map(|idx| idx.to_string())
+                        .collect::<Vec<_>>()
+                        .join(", ")
+                )
+            })
+            .collect::<Vec<_>>();
+        assert_eq!(actual, expected);
+    }
+
+    pub(crate) fn assert_declarations(
+        symbol: &SymbolState,
+        may_be_undeclared: bool,
+        expected: &[u32],
+    ) {
+        assert_eq!(symbol.declarations.may_be_undeclared(), may_be_undeclared);
+        let actual = symbol
+            .declarations()
+            .iter()
+            .map(ScopedDefinitionId::as_u32)
+            .collect::<Vec<_>>();
+        assert_eq!(actual, expected);
+    }
+
+    #[test]
+    fn unbound() {
+        let sym = SymbolState::undefined();
+
+        assert_bindings(&sym, true, &[]);
+    }
+
+    #[test]
+    fn with() {
+        let mut sym = SymbolState::undefined();
+        sym.record_binding(ScopedDefinitionId::from_u32(0));
+
+        assert_bindings(&sym, false, &["0<>"]);
+    }
+
+    #[test]
+    fn set_may_be_unbound() {
+        let mut sym = SymbolState::undefined();
+        sym.record_binding(ScopedDefinitionId::from_u32(0));
+        sym.set_may_be_unbound();
+
+        assert_bindings(&sym, true, &["0<>"]);
+    }
+
+    #[test]
+    fn record_constraint() {
+        let mut sym = SymbolState::undefined();
+        sym.record_binding(ScopedDefinitionId::from_u32(0));
+        sym.record_constraint(ScopedConstraintId::from_u32(0));
+
+        assert_bindings(&sym, false, &["0<0>"]);
+    }
+
+    #[test]
+    fn merge() {
+        // merging the same definition with the same constraint keeps the constraint
+        let mut sym0a = SymbolState::undefined();
+        sym0a.record_binding(ScopedDefinitionId::from_u32(0));
+        sym0a.record_constraint(ScopedConstraintId::from_u32(0));
+
+        let mut sym0b = SymbolState::undefined();
+        sym0b.record_binding(ScopedDefinitionId::from_u32(0));
+        sym0b.record_constraint(ScopedConstraintId::from_u32(0));
+
+        sym0a.merge(sym0b);
+        let mut sym0 = sym0a;
+        assert_bindings(&sym0, false, &["0<0>"]);
+
+        // merging the same definition with differing constraints drops all constraints
+        let mut sym1a = SymbolState::undefined();
+        sym1a.record_binding(ScopedDefinitionId::from_u32(1));
+        sym1a.record_constraint(ScopedConstraintId::from_u32(1));
+
+        let mut sym1b = SymbolState::undefined();
+        sym1b.record_binding(ScopedDefinitionId::from_u32(1));
+        sym1b.record_constraint(ScopedConstraintId::from_u32(2));
+
+        sym1a.merge(sym1b);
+        let sym1 = sym1a;
+        assert_bindings(&sym1, false, &["1<>"]);
+
+        // merging a constrained definition with unbound keeps both
+        let mut sym2a = SymbolState::undefined();
+        sym2a.record_binding(ScopedDefinitionId::from_u32(2));
+        sym2a.record_constraint(ScopedConstraintId::from_u32(3));
+
+        let sym2b = SymbolState::undefined();
+
+        sym2a.merge(sym2b);
+        let sym2 = sym2a;
+        assert_bindings(&sym2, true, &["2<3>"]);
+
+        // merging different definitions keeps them each with their existing constraints
+        sym0.merge(sym2);
+        let sym = sym0;
+        assert_bindings(&sym, true, &["0<0>", "2<3>"]);
+    }
+
+    #[test]
+    fn no_declaration() {
+        let sym = SymbolState::undefined();
+
+        assert_declarations(&sym, true, &[]);
+    }
+
+    #[test]
+    fn record_declaration() {
+        let mut sym = SymbolState::undefined();
+        sym.record_declaration(ScopedDefinitionId::from_u32(1));
+
+        assert_declarations(&sym, false, &[1]);
+    }
+
+    #[test]
+    fn record_declaration_override() {
+        let mut sym = SymbolState::undefined();
+        sym.record_declaration(ScopedDefinitionId::from_u32(1));
+        sym.record_declaration(ScopedDefinitionId::from_u32(2));
+
+        assert_declarations(&sym, false, &[2]);
+    }
+
+    #[test]
+    fn record_declaration_merge() {
+        let mut sym = SymbolState::undefined();
+        sym.record_declaration(ScopedDefinitionId::from_u32(1));
+
+        let mut sym2 = SymbolState::undefined();
+        sym2.record_declaration(ScopedDefinitionId::from_u32(2));
+
+        sym.merge(sym2);
+
+        assert_declarations(&sym, false, &[1, 2]);
+    }
+
+    #[test]
+    fn record_declaration_merge_partial_undeclared() {
+        let mut sym = SymbolState::undefined();
+        sym.record_declaration(ScopedDefinitionId::from_u32(1));
+
+        let sym2 = SymbolState::undefined();
+
+        sym.merge(sym2);
+
+        assert_declarations(&sym, true, &[1]);
+    }
+
+    #[test]
+    fn set_may_be_undeclared() {
+        let mut sym = SymbolState::undefined();
+        sym.record_declaration(ScopedDefinitionId::from_u32(0));
+        sym.set_may_be_undeclared();
+
+        assert_declarations(&sym, true, &[0]);
+    }
+}

crates/red_knot_python_semantic/src/semantic_model.rs

@@ -1,12 +1,14 @@
-use ruff_db::files::File;
+use ruff_db::files::{File, FilePath};
+use ruff_db::source::line_index;
 use ruff_python_ast as ast;
-use ruff_python_ast::{Expr, ExpressionRef, StmtClassDef};
+use ruff_python_ast::{Expr, ExpressionRef};
+use ruff_source_file::LineIndex;
 
 use crate::module_name::ModuleName;
 use crate::module_resolver::{resolve_module, Module};
 use crate::semantic_index::ast_ids::HasScopedAstId;
 use crate::semantic_index::semantic_index;
-use crate::types::{definition_ty, global_symbol_ty_by_name, infer_scope_types, Type};
+use crate::types::{binding_ty, global_symbol_ty, infer_scope_types, Type};
 use crate::Db;
 
 pub struct SemanticModel<'db> {
@@ -25,12 +27,20 @@ impl<'db> SemanticModel<'db> {
         self.db
     }
 
+    pub fn file_path(&self) -> &FilePath {
+        self.file.path(self.db)
+    }
+
+    pub fn line_index(&self) -> LineIndex {
+        line_index(self.db.upcast(), self.file)
+    }
+
     pub fn resolve_module(&self, module_name: ModuleName) -> Option<Module> {
         resolve_module(self.db, module_name)
     }
 
     pub fn global_symbol_ty(&self, module: &Module, symbol_name: &str) -> Type<'db> {
-        global_symbol_ty_by_name(self.db, module.file(), symbol_name)
+        global_symbol_ty(self.db, module.file(), symbol_name)
     }
 }
 
@@ -137,29 +147,24 @@ impl HasTy for ast::Expr {
     }
 }
 
-impl HasTy for ast::StmtFunctionDef {
-    fn ty<'db>(&self, model: &SemanticModel<'db>) -> Type<'db> {
-        let index = semantic_index(model.db, model.file);
-        let definition = index.definition(self);
-        definition_ty(model.db, definition)
-    }
+macro_rules! impl_binding_has_ty {
+    ($ty: ty) => {
+        impl HasTy for $ty {
+            #[inline]
+            fn ty<'db>(&self, model: &SemanticModel<'db>) -> Type<'db> {
+                let index = semantic_index(model.db, model.file);
+                let binding = index.definition(self);
+                binding_ty(model.db, binding)
+            }
+        }
+    };
 }
 
-impl HasTy for StmtClassDef {
-    fn ty<'db>(&self, model: &SemanticModel<'db>) -> Type<'db> {
-        let index = semantic_index(model.db, model.file);
-        let definition = index.definition(self);
-        definition_ty(model.db, definition)
-    }
-}
-
-impl HasTy for ast::Alias {
-    fn ty<'db>(&self, model: &SemanticModel<'db>) -> Type<'db> {
-        let index = semantic_index(model.db, model.file);
-        let definition = index.definition(self);
-        definition_ty(model.db, definition)
-    }
-}
+impl_binding_has_ty!(ast::StmtFunctionDef);
+impl_binding_has_ty!(ast::StmtClassDef);
+impl_binding_has_ty!(ast::Alias);
+impl_binding_has_ty!(ast::Parameter);
+impl_binding_has_ty!(ast::ParameterWithDefault);
 
 #[cfg(test)]
 mod tests {
@@ -171,29 +176,27 @@ mod tests {
     use crate::program::{Program, SearchPathSettings};
     use crate::python_version::PythonVersion;
     use crate::types::Type;
-    use crate::{HasTy, SemanticModel};
+    use crate::{HasTy, ProgramSettings, SemanticModel};
 
-    fn setup_db() -> TestDb {
-        let db = TestDb::new();
-        Program::new(
+    fn setup_db<'a>(files: impl IntoIterator<Item = (&'a str, &'a str)>) -> anyhow::Result<TestDb> {
+        let mut db = TestDb::new();
+        db.write_files(files)?;
+
+        Program::from_settings(
             &db,
-            PythonVersion::default(),
-            SearchPathSettings {
-                extra_paths: vec![],
-                src_root: SystemPathBuf::from("/src"),
-                site_packages: vec![],
-                custom_typeshed: None,
+            &ProgramSettings {
+                target_version: PythonVersion::default(),
+                search_paths: SearchPathSettings::new(SystemPathBuf::from("/src")),
             },
-        );
+        )?;
 
-        db
+        Ok(db)
     }
 
     #[test]
     fn function_ty() -> anyhow::Result<()> {
-        let mut db = setup_db();
+        let db = setup_db([("/src/foo.py", "def test(): pass")])?;
 
-        db.write_file("/src/foo.py", "def test(): pass")?;
         let foo = system_path_to_file(&db, "/src/foo.py").unwrap();
 
         let ast = parsed_module(&db, foo);
@@ -209,9 +212,8 @@ mod tests {
 
     #[test]
     fn class_ty() -> anyhow::Result<()> {
-        let mut db = setup_db();
+        let db = setup_db([("/src/foo.py", "class Test: pass")])?;
 
-        db.write_file("/src/foo.py", "class Test: pass")?;
         let foo = system_path_to_file(&db, "/src/foo.py").unwrap();
 
         let ast = parsed_module(&db, foo);
@@ -227,12 +229,11 @@ mod tests {
 
     #[test]
     fn alias_ty() -> anyhow::Result<()> {
-        let mut db = setup_db();
-
-        db.write_files([
+        let db = setup_db([
             ("/src/foo.py", "class Test: pass"),
             ("/src/bar.py", "from foo import Test"),
         ])?;
+
         let bar = system_path_to_file(&db, "/src/bar.py").unwrap();
 
         let ast = parsed_module(&db, bar);

crates/red_knot_python_semantic/src/site_packages.rs

@@ -13,9 +13,10 @@ use std::io;
 use std::num::NonZeroUsize;
 use std::ops::Deref;
 
-use red_knot_python_semantic::PythonVersion;
 use ruff_db::system::{System, SystemPath, SystemPathBuf};
 
+use crate::PythonVersion;
+
 type SitePackagesDiscoveryResult<T> = Result<T, SitePackagesDiscoveryError>;
 
 /// Abstraction for a Python virtual environment.
@@ -24,7 +25,7 @@ type SitePackagesDiscoveryResult<T> = Result<T, SitePackagesDiscoveryError>;
 /// The format of this file is not defined anywhere, and exactly which keys are present
 /// depends on the tool that was used to create the virtual environment.
 #[derive(Debug)]
-pub struct VirtualEnvironment {
+pub(crate) struct VirtualEnvironment {
     venv_path: SysPrefixPath,
     base_executable_home_path: PythonHomePath,
     include_system_site_packages: bool,
@@ -41,7 +42,7 @@ pub struct VirtualEnvironment {
 }
 
 impl VirtualEnvironment {
-    pub fn new(
+    pub(crate) fn new(
         path: impl AsRef<SystemPath>,
         system: &dyn System,
     ) -> SitePackagesDiscoveryResult<Self> {
@@ -55,7 +56,7 @@ impl VirtualEnvironment {
 
         let venv_path = SysPrefixPath::new(path, system)?;
         let pyvenv_cfg_path = venv_path.join("pyvenv.cfg");
-        tracing::debug!("Attempting to parse virtual environment metadata at {pyvenv_cfg_path}");
+        tracing::debug!("Attempting to parse virtual environment metadata at '{pyvenv_cfg_path}'");
 
         let pyvenv_cfg = system
             .read_to_string(&pyvenv_cfg_path)
@@ -157,7 +158,7 @@ impl VirtualEnvironment {
     /// Return a list of `site-packages` directories that are available from this virtual environment
     ///
     /// See the documentation for `site_packages_dir_from_sys_prefix` for more details.
-    pub fn site_packages_directories(
+    pub(crate) fn site_packages_directories(
         &self,
         system: &dyn System,
     ) -> SitePackagesDiscoveryResult<Vec<SystemPathBuf>> {
@@ -191,7 +192,7 @@ impl VirtualEnvironment {
             } else {
                 tracing::warn!(
                     "Failed to resolve `sys.prefix` of the system Python installation \
-from the `home` value in the `pyvenv.cfg` file at {}. \
+from the `home` value in the `pyvenv.cfg` file at '{}'. \
 System site-packages will not be used for module resolution.",
                     venv_path.join("pyvenv.cfg")
                 );
@@ -204,7 +205,7 @@ System site-packages will not be used for module resolution.",
 }
 
 #[derive(Debug, thiserror::Error)]
-pub enum SitePackagesDiscoveryError {
+pub(crate) enum SitePackagesDiscoveryError {
     #[error("Invalid --venv-path argument: {0} could not be canonicalized")]
     VenvDirCanonicalizationError(SystemPathBuf, #[source] io::Error),
     #[error("Invalid --venv-path argument: {0} does not point to a directory on disk")]
@@ -221,7 +222,7 @@ pub enum SitePackagesDiscoveryError {
 
 /// The various ways in which parsing a `pyvenv.cfg` file could fail
 #[derive(Debug)]
-pub enum PyvenvCfgParseErrorKind {
+pub(crate) enum PyvenvCfgParseErrorKind {
     TooManyEquals { line_number: NonZeroUsize },
     MalformedKeyValuePair { line_number: NonZeroUsize },
     NoHomeKey,
@@ -370,7 +371,7 @@ fn site_packages_directory_from_sys_prefix(
 ///
 /// [`sys.prefix`]: https://docs.python.org/3/library/sys.html#sys.prefix
 #[derive(Debug, PartialEq, Eq, Clone)]
-pub struct SysPrefixPath(SystemPathBuf);
+pub(crate) struct SysPrefixPath(SystemPathBuf);
 
 impl SysPrefixPath {
     fn new(
@@ -425,7 +426,7 @@ impl Deref for SysPrefixPath {
 
 impl fmt::Display for SysPrefixPath {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(f, "`sys.prefix` path {}", self.0)
+        write!(f, "`sys.prefix` path '{}'", self.0)
     }
 }
 
@@ -482,7 +483,7 @@ impl Deref for PythonHomePath {
 
 impl fmt::Display for PythonHomePath {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(f, "`home` location {}", self.0)
+        write!(f, "`home` location '{}'", self.0)
     }
 }
 

crates/red_knot_python_semantic/src/stdlib.rs

@@ -0,0 +1,87 @@
+use crate::module_name::ModuleName;
+use crate::module_resolver::resolve_module;
+use crate::semantic_index::global_scope;
+use crate::semantic_index::symbol::ScopeId;
+use crate::types::{global_symbol_ty, Type};
+use crate::Db;
+
+/// Enumeration of various core stdlib modules, for which we have dedicated Salsa queries.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum CoreStdlibModule {
+    Builtins,
+    Types,
+    Typeshed,
+    TypingExtensions,
+}
+
+impl CoreStdlibModule {
+    fn name(self) -> ModuleName {
+        let module_name = match self {
+            Self::Builtins => "builtins",
+            Self::Types => "types",
+            Self::Typeshed => "_typeshed",
+            Self::TypingExtensions => "typing_extensions",
+        };
+        ModuleName::new_static(module_name)
+            .unwrap_or_else(|| panic!("{module_name} should be a valid module name!"))
+    }
+}
+
+/// Lookup the type of `symbol` in a given core module
+///
+/// Returns `Unbound` if the given core module cannot be resolved for some reason
+fn core_module_symbol_ty<'db>(
+    db: &'db dyn Db,
+    core_module: CoreStdlibModule,
+    symbol: &str,
+) -> Type<'db> {
+    resolve_module(db, core_module.name())
+        .map(|module| global_symbol_ty(db, module.file(), symbol))
+        .unwrap_or(Type::Unbound)
+}
+
+/// Lookup the type of `symbol` in the builtins namespace.
+///
+/// Returns `Unbound` if the `builtins` module isn't available for some reason.
+#[inline]
+pub(crate) fn builtins_symbol_ty<'db>(db: &'db dyn Db, symbol: &str) -> Type<'db> {
+    core_module_symbol_ty(db, CoreStdlibModule::Builtins, symbol)
+}
+
+/// Lookup the type of `symbol` in the `types` module namespace.
+///
+/// Returns `Unbound` if the `types` module isn't available for some reason.
+#[inline]
+pub(crate) fn types_symbol_ty<'db>(db: &'db dyn Db, symbol: &str) -> Type<'db> {
+    core_module_symbol_ty(db, CoreStdlibModule::Types, symbol)
+}
+
+/// Lookup the type of `symbol` in the `_typeshed` module namespace.
+///
+/// Returns `Unbound` if the `_typeshed` module isn't available for some reason.
+#[inline]
+pub(crate) fn typeshed_symbol_ty<'db>(db: &'db dyn Db, symbol: &str) -> Type<'db> {
+    core_module_symbol_ty(db, CoreStdlibModule::Typeshed, symbol)
+}
+
+/// Lookup the type of `symbol` in the `typing_extensions` module namespace.
+///
+/// Returns `Unbound` if the `typing_extensions` module isn't available for some reason.
+#[inline]
+pub(crate) fn typing_extensions_symbol_ty<'db>(db: &'db dyn Db, symbol: &str) -> Type<'db> {
+    core_module_symbol_ty(db, CoreStdlibModule::TypingExtensions, symbol)
+}
+
+/// Get the scope of a core stdlib module.
+///
+/// Can return `None` if a custom typeshed is used that is missing the core module in question.
+fn core_module_scope(db: &dyn Db, core_module: CoreStdlibModule) -> Option<ScopeId<'_>> {
+    resolve_module(db, core_module.name()).map(|module| global_scope(db, module.file()))
+}
+
+/// Get the `builtins` module scope.
+///
+/// Can return `None` if a custom typeshed is used that is missing `builtins.pyi`.
+pub(crate) fn builtins_module_scope(db: &dyn Db) -> Option<ScopeId<'_>> {
+    core_module_scope(db, CoreStdlibModule::Builtins)
+}

crates/red_knot_python_semantic/src/types/builder.rs

@@ -0,0 +1,575 @@
+//! Smart builders for union and intersection types.
+//!
+//! Invariants we maintain here:
+//!   * No single-element union types (should just be the contained type instead.)
+//!   * No single-positive-element intersection types. Single-negative-element are OK, we don't
+//!     have a standalone negation type so there's no other representation for this.
+//!   * The same type should never appear more than once in a union or intersection. (This should
+//!     be expanded to cover subtyping -- see below -- but for now we only implement it for type
+//!     identity.)
+//!   * Disjunctive normal form (DNF): the tree of unions and intersections can never be deeper
+//!     than a union-of-intersections. Unions cannot contain other unions (the inner union just
+//!     flattens into the outer one), intersections cannot contain other intersections (also
+//!     flattens), and intersections cannot contain unions (the intersection distributes over the
+//!     union, inverting it into a union-of-intersections).
+//!
+//! The implication of these invariants is that a [`UnionBuilder`] does not necessarily build a
+//! [`Type::Union`]. For example, if only one type is added to the [`UnionBuilder`], `build()` will
+//! just return that type directly. The same is true for [`IntersectionBuilder`]; for example, if a
+//! union type is added to the intersection, it will distribute and [`IntersectionBuilder::build`]
+//! may end up returning a [`Type::Union`] of intersections.
+//!
+//! In the future we should have these additional invariants, but they aren't implemented yet:
+//!   * No type in a union can be a subtype of any other type in the union (just eliminate the
+//!     subtype from the union).
+//!   * No type in an intersection can be a supertype of any other type in the intersection (just
+//!     eliminate the supertype from the intersection).
+//!   * An intersection containing two non-overlapping types should simplify to [`Type::Never`].
+use crate::types::{builtins_symbol_ty, IntersectionType, Type, UnionType};
+use crate::{Db, FxOrderSet};
+use smallvec::SmallVec;
+
+pub(crate) struct UnionBuilder<'db> {
+    elements: Vec<Type<'db>>,
+    db: &'db dyn Db,
+}
+
+impl<'db> UnionBuilder<'db> {
+    pub(crate) fn new(db: &'db dyn Db) -> Self {
+        Self {
+            db,
+            elements: vec![],
+        }
+    }
+
+    /// Adds a type to this union.
+    pub(crate) fn add(mut self, ty: Type<'db>) -> Self {
+        match ty {
+            Type::Union(union) => {
+                let new_elements = union.elements(self.db);
+                self.elements.reserve(new_elements.len());
+                for element in new_elements {
+                    self = self.add(*element);
+                }
+            }
+            Type::Never => {}
+            _ => {
+                let bool_pair = if let Type::BooleanLiteral(b) = ty {
+                    Some(Type::BooleanLiteral(!b))
+                } else {
+                    None
+                };
+
+                let mut to_add = ty;
+                let mut to_remove = SmallVec::<[usize; 2]>::new();
+                for (index, element) in self.elements.iter().enumerate() {
+                    if Some(*element) == bool_pair {
+                        to_add = builtins_symbol_ty(self.db, "bool");
+                        to_remove.push(index);
+                        // The type we are adding is a BooleanLiteral, which doesn't have any
+                        // subtypes. And we just found that the union already contained our
+                        // mirror-image BooleanLiteral, so it can't also contain bool or any
+                        // supertype of bool. Therefore, we are done.
+                        break;
+                    }
+                    if ty.is_subtype_of(self.db, *element) {
+                        return self;
+                    } else if element.is_subtype_of(self.db, ty) {
+                        to_remove.push(index);
+                    }
+                }
+
+                match to_remove[..] {
+                    [] => self.elements.push(to_add),
+                    [index] => self.elements[index] = to_add,
+                    _ => {
+                        let mut current_index = 0;
+                        let mut to_remove = to_remove.into_iter();
+                        let mut next_to_remove_index = to_remove.next();
+                        self.elements.retain(|_| {
+                            let retain = if Some(current_index) == next_to_remove_index {
+                                next_to_remove_index = to_remove.next();
+                                false
+                            } else {
+                                true
+                            };
+                            current_index += 1;
+                            retain
+                        });
+                        self.elements.push(to_add);
+                    }
+                }
+            }
+        }
+
+        self
+    }
+
+    pub(crate) fn build(self) -> Type<'db> {
+        match self.elements.len() {
+            0 => Type::Never,
+            1 => self.elements[0],
+            _ => Type::Union(UnionType::new(self.db, self.elements.into())),
+        }
+    }
+}
+
+#[derive(Clone)]
+pub(crate) struct IntersectionBuilder<'db> {
+    // Really this builds a union-of-intersections, because we always keep our set-theoretic types
+    // in disjunctive normal form (DNF), a union of intersections. In the simplest case there's
+    // just a single intersection in this vector, and we are building a single intersection type,
+    // but if a union is added to the intersection, we'll distribute ourselves over that union and
+    // create a union of intersections.
+    intersections: Vec<InnerIntersectionBuilder<'db>>,
+    db: &'db dyn Db,
+}
+
+impl<'db> IntersectionBuilder<'db> {
+    pub(crate) fn new(db: &'db dyn Db) -> Self {
+        Self {
+            db,
+            intersections: vec![InnerIntersectionBuilder::new()],
+        }
+    }
+
+    fn empty(db: &'db dyn Db) -> Self {
+        Self {
+            db,
+            intersections: vec![],
+        }
+    }
+
+    pub(crate) fn add_positive(mut self, ty: Type<'db>) -> Self {
+        if let Type::Union(union) = ty {
+            // Distribute ourself over this union: for each union element, clone ourself and
+            // intersect with that union element, then create a new union-of-intersections with all
+            // of those sub-intersections in it. E.g. if `self` is a simple intersection `T1 & T2`
+            // and we add `T3 | T4` to the intersection, we don't get `T1 & T2 & (T3 | T4)` (that's
+            // not in DNF), we distribute the union and get `(T1 & T3) | (T2 & T3) | (T1 & T4) |
+            // (T2 & T4)`. If `self` is already a union-of-intersections `(T1 & T2) | (T3 & T4)`
+            // and we add `T5 | T6` to it, that flattens all the way out to `(T1 & T2 & T5) | (T1 &
+            // T2 & T6) | (T3 & T4 & T5) ...` -- you get the idea.
+            union
+                .elements(self.db)
+                .iter()
+                .map(|elem| self.clone().add_positive(*elem))
+                .fold(IntersectionBuilder::empty(self.db), |mut builder, sub| {
+                    builder.intersections.extend(sub.intersections);
+                    builder
+                })
+        } else {
+            // If we are already a union-of-intersections, distribute the new intersected element
+            // across all of those intersections.
+            for inner in &mut self.intersections {
+                inner.add_positive(self.db, ty);
+            }
+            self
+        }
+    }
+
+    pub(crate) fn add_negative(mut self, ty: Type<'db>) -> Self {
+        // See comments above in `add_positive`; this is just the negated version.
+        if let Type::Union(union) = ty {
+            union
+                .elements(self.db)
+                .iter()
+                .map(|elem| self.clone().add_negative(*elem))
+                .fold(IntersectionBuilder::empty(self.db), |mut builder, sub| {
+                    builder.intersections.extend(sub.intersections);
+                    builder
+                })
+        } else {
+            for inner in &mut self.intersections {
+                inner.add_negative(self.db, ty);
+            }
+            self
+        }
+    }
+
+    pub(crate) fn build(mut self) -> Type<'db> {
+        // Avoid allocating the UnionBuilder unnecessarily if we have just one intersection:
+        if self.intersections.len() == 1 {
+            self.intersections.pop().unwrap().build(self.db)
+        } else {
+            UnionType::from_elements(
+                self.db,
+                self.intersections
+                    .into_iter()
+                    .map(|inner| inner.build(self.db)),
+            )
+        }
+    }
+}
+
+#[derive(Debug, Clone, Default)]
+struct InnerIntersectionBuilder<'db> {
+    positive: FxOrderSet<Type<'db>>,
+    negative: FxOrderSet<Type<'db>>,
+}
+
+impl<'db> InnerIntersectionBuilder<'db> {
+    fn new() -> Self {
+        Self::default()
+    }
+
+    /// Adds a positive type to this intersection.
+    fn add_positive(&mut self, db: &'db dyn Db, ty: Type<'db>) {
+        match ty {
+            Type::Intersection(inter) => {
+                let pos = inter.positive(db);
+                let neg = inter.negative(db);
+                self.positive.extend(pos.difference(&self.negative));
+                self.negative.extend(neg.difference(&self.positive));
+                self.positive.retain(|elem| !neg.contains(elem));
+                self.negative.retain(|elem| !pos.contains(elem));
+            }
+            _ => {
+                if !self.negative.remove(&ty) {
+                    self.positive.insert(ty);
+                };
+            }
+        }
+    }
+
+    /// Adds a negative type to this intersection.
+    fn add_negative(&mut self, db: &'db dyn Db, ty: Type<'db>) {
+        // TODO Any/Unknown actually should not self-cancel
+        match ty {
+            Type::Intersection(intersection) => {
+                let pos = intersection.negative(db);
+                let neg = intersection.positive(db);
+                self.positive.extend(pos.difference(&self.negative));
+                self.negative.extend(neg.difference(&self.positive));
+                self.positive.retain(|elem| !neg.contains(elem));
+                self.negative.retain(|elem| !pos.contains(elem));
+            }
+            Type::Never => {}
+            Type::Unbound => {}
+            _ => {
+                if !self.positive.remove(&ty) {
+                    self.negative.insert(ty);
+                };
+            }
+        }
+    }
+
+    fn simplify(&mut self) {
+        // TODO this should be generalized based on subtyping, for now we just handle a few cases
+
+        // Never is a subtype of all types
+        if self.positive.contains(&Type::Never) {
+            self.positive.retain(Type::is_never);
+            self.negative.clear();
+        }
+
+        if self.positive.contains(&Type::Unbound) {
+            self.positive.retain(Type::is_unbound);
+            self.negative.clear();
+        }
+
+        // None intersects only with object
+        for pos in &self.positive {
+            if let Type::Instance(_) = pos {
+                // could be `object` type
+            } else {
+                self.negative.remove(&Type::None);
+                break;
+            }
+        }
+    }
+
+    fn build(mut self, db: &'db dyn Db) -> Type<'db> {
+        self.simplify();
+        match (self.positive.len(), self.negative.len()) {
+            (0, 0) => Type::Never,
+            (1, 0) => self.positive[0],
+            _ => {
+                self.positive.shrink_to_fit();
+                self.negative.shrink_to_fit();
+                Type::Intersection(IntersectionType::new(db, self.positive, self.negative))
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{IntersectionBuilder, IntersectionType, Type, UnionType};
+    use crate::db::tests::TestDb;
+    use crate::program::{Program, SearchPathSettings};
+    use crate::python_version::PythonVersion;
+    use crate::types::{builtins_symbol_ty, UnionBuilder};
+    use crate::ProgramSettings;
+    use ruff_db::system::{DbWithTestSystem, SystemPathBuf};
+
+    fn setup_db() -> TestDb {
+        let db = TestDb::new();
+
+        let src_root = SystemPathBuf::from("/src");
+        db.memory_file_system()
+            .create_directory_all(&src_root)
+            .unwrap();
+
+        Program::from_settings(
+            &db,
+            &ProgramSettings {
+                target_version: PythonVersion::default(),
+                search_paths: SearchPathSettings::new(src_root),
+            },
+        )
+        .expect("Valid search path settings");
+
+        db
+    }
+
+    #[test]
+    fn build_union() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let t1 = Type::IntLiteral(1);
+        let union = UnionType::from_elements(&db, [t0, t1]).expect_union();
+
+        assert_eq!(union.elements(&db), &[t0, t1]);
+    }
+
+    #[test]
+    fn build_union_single() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let ty = UnionType::from_elements(&db, [t0]);
+        assert_eq!(ty, t0);
+    }
+
+    #[test]
+    fn build_union_empty() {
+        let db = setup_db();
+        let ty = UnionBuilder::new(&db).build();
+        assert_eq!(ty, Type::Never);
+    }
+
+    #[test]
+    fn build_union_never() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let ty = UnionType::from_elements(&db, [t0, Type::Never]);
+        assert_eq!(ty, t0);
+    }
+
+    #[test]
+    fn build_union_bool() {
+        let db = setup_db();
+        let bool_ty = builtins_symbol_ty(&db, "bool");
+
+        let t0 = Type::BooleanLiteral(true);
+        let t1 = Type::BooleanLiteral(true);
+        let t2 = Type::BooleanLiteral(false);
+        let t3 = Type::IntLiteral(17);
+
+        let union = UnionType::from_elements(&db, [t0, t1, t3]).expect_union();
+        assert_eq!(union.elements(&db), &[t0, t3]);
+
+        let union = UnionType::from_elements(&db, [t0, t1, t2, t3]).expect_union();
+        assert_eq!(union.elements(&db), &[bool_ty, t3]);
+    }
+
+    #[test]
+    fn build_union_flatten() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let t1 = Type::IntLiteral(1);
+        let t2 = Type::IntLiteral(2);
+        let u1 = UnionType::from_elements(&db, [t0, t1]);
+        let union = UnionType::from_elements(&db, [u1, t2]).expect_union();
+
+        assert_eq!(union.elements(&db), &[t0, t1, t2]);
+    }
+
+    #[test]
+    fn build_union_simplify_subtype() {
+        let db = setup_db();
+        let t0 = builtins_symbol_ty(&db, "str").to_instance(&db);
+        let t1 = Type::LiteralString;
+        let u0 = UnionType::from_elements(&db, [t0, t1]);
+        let u1 = UnionType::from_elements(&db, [t1, t0]);
+
+        assert_eq!(u0, t0);
+        assert_eq!(u1, t0);
+    }
+
+    #[test]
+    fn build_union_no_simplify_unknown() {
+        let db = setup_db();
+        let t0 = builtins_symbol_ty(&db, "str").to_instance(&db);
+        let t1 = Type::Unknown;
+        let u0 = UnionType::from_elements(&db, [t0, t1]);
+        let u1 = UnionType::from_elements(&db, [t1, t0]);
+
+        assert_eq!(u0.expect_union().elements(&db), &[t0, t1]);
+        assert_eq!(u1.expect_union().elements(&db), &[t1, t0]);
+    }
+
+    #[test]
+    fn build_union_subsume_multiple() {
+        let db = setup_db();
+        let str_ty = builtins_symbol_ty(&db, "str").to_instance(&db);
+        let int_ty = builtins_symbol_ty(&db, "int").to_instance(&db);
+        let object_ty = builtins_symbol_ty(&db, "object").to_instance(&db);
+        let unknown_ty = Type::Unknown;
+
+        let u0 = UnionType::from_elements(&db, [str_ty, unknown_ty, int_ty, object_ty]);
+
+        assert_eq!(u0.expect_union().elements(&db), &[unknown_ty, object_ty]);
+    }
+
+    impl<'db> IntersectionType<'db> {
+        fn pos_vec(self, db: &'db TestDb) -> Vec<Type<'db>> {
+            self.positive(db).into_iter().copied().collect()
+        }
+
+        fn neg_vec(self, db: &'db TestDb) -> Vec<Type<'db>> {
+            self.negative(db).into_iter().copied().collect()
+        }
+    }
+
+    #[test]
+    fn build_intersection() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let ta = Type::Any;
+        let intersection = IntersectionBuilder::new(&db)
+            .add_positive(ta)
+            .add_negative(t0)
+            .build()
+            .expect_intersection();
+
+        assert_eq!(intersection.pos_vec(&db), &[ta]);
+        assert_eq!(intersection.neg_vec(&db), &[t0]);
+    }
+
+    #[test]
+    fn build_intersection_flatten_positive() {
+        let db = setup_db();
+        let ta = Type::Any;
+        let t1 = Type::IntLiteral(1);
+        let t2 = Type::IntLiteral(2);
+        let i0 = IntersectionBuilder::new(&db)
+            .add_positive(ta)
+            .add_negative(t1)
+            .build();
+        let intersection = IntersectionBuilder::new(&db)
+            .add_positive(t2)
+            .add_positive(i0)
+            .build()
+            .expect_intersection();
+
+        assert_eq!(intersection.pos_vec(&db), &[t2, ta]);
+        assert_eq!(intersection.neg_vec(&db), &[t1]);
+    }
+
+    #[test]
+    fn build_intersection_flatten_negative() {
+        let db = setup_db();
+        let ta = Type::Any;
+        let t1 = Type::IntLiteral(1);
+        let t2 = Type::IntLiteral(2);
+        let i0 = IntersectionBuilder::new(&db)
+            .add_positive(ta)
+            .add_negative(t1)
+            .build();
+        let intersection = IntersectionBuilder::new(&db)
+            .add_positive(t2)
+            .add_negative(i0)
+            .build()
+            .expect_intersection();
+
+        assert_eq!(intersection.pos_vec(&db), &[t2, t1]);
+        assert_eq!(intersection.neg_vec(&db), &[ta]);
+    }
+
+    #[test]
+    fn intersection_distributes_over_union() {
+        let db = setup_db();
+        let t0 = Type::IntLiteral(0);
+        let t1 = Type::IntLiteral(1);
+        let ta = Type::Any;
+        let u0 = UnionType::from_elements(&db, [t0, t1]);
+
+        let union = IntersectionBuilder::new(&db)
+            .add_positive(ta)
+            .add_positive(u0)
+            .build()
+            .expect_union();
+        let [Type::Intersection(i0), Type::Intersection(i1)] = union.elements(&db)[..] else {
+            panic!("expected a union of two intersections");
+        };
+        assert_eq!(i0.pos_vec(&db), &[ta, t0]);
+        assert_eq!(i1.pos_vec(&db), &[ta, t1]);
+    }
+
+    #[test]
+    fn build_intersection_self_negation() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_positive(Type::None)
+            .add_negative(Type::None)
+            .build();
+
+        assert_eq!(ty, Type::Never);
+    }
+
+    #[test]
+    fn build_intersection_simplify_negative_never() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_positive(Type::None)
+            .add_negative(Type::Never)
+            .build();
+
+        assert_eq!(ty, Type::None);
+    }
+
+    #[test]
+    fn build_intersection_simplify_positive_never() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_positive(Type::None)
+            .add_positive(Type::Never)
+            .build();
+
+        assert_eq!(ty, Type::Never);
+    }
+
+    #[test]
+    fn build_intersection_simplify_positive_unbound() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_positive(Type::Unbound)
+            .add_positive(Type::IntLiteral(1))
+            .build();
+
+        assert_eq!(ty, Type::Unbound);
+    }
+
+    #[test]
+    fn build_intersection_simplify_negative_unbound() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_negative(Type::Unbound)
+            .add_positive(Type::IntLiteral(1))
+            .build();
+
+        assert_eq!(ty, Type::IntLiteral(1));
+    }
+
+    #[test]
+    fn build_intersection_simplify_negative_none() {
+        let db = setup_db();
+        let ty = IntersectionBuilder::new(&db)
+            .add_negative(Type::None)
+            .add_positive(Type::IntLiteral(1))
+            .build();
+
+        assert_eq!(ty, Type::IntLiteral(1));
+    }
+}

crates/red_knot_python_semantic/src/types/diagnostic.rs

@@ -0,0 +1,111 @@
+use ruff_db::files::File;
+use ruff_text_size::{Ranged, TextRange};
+use std::fmt::Formatter;
+use std::ops::Deref;
+use std::sync::Arc;
+
+#[derive(Debug, Eq, PartialEq)]
+pub struct TypeCheckDiagnostic {
+    // TODO: Don't use string keys for rules
+    pub(super) rule: String,
+    pub(super) message: String,
+    pub(super) range: TextRange,
+    pub(super) file: File,
+}
+
+impl TypeCheckDiagnostic {
+    pub fn rule(&self) -> &str {
+        &self.rule
+    }
+
+    pub fn message(&self) -> &str {
+        &self.message
+    }
+
+    pub fn file(&self) -> File {
+        self.file
+    }
+}
+
+impl Ranged for TypeCheckDiagnostic {
+    fn range(&self) -> TextRange {
+        self.range
+    }
+}
+
+/// A collection of type check diagnostics.
+///
+/// The diagnostics are wrapped in an `Arc` because they need to be cloned multiple times
+/// when going from `infer_expression` to `check_file`. We could consider
+/// making [`TypeCheckDiagnostic`] a Salsa struct to have them Arena-allocated (once the Tables refactor is done).
+/// Using Salsa struct does have the downside that it leaks the Salsa dependency into diagnostics and
+/// each Salsa-struct comes with an overhead.
+#[derive(Default, Eq, PartialEq)]
+pub struct TypeCheckDiagnostics {
+    inner: Vec<std::sync::Arc<TypeCheckDiagnostic>>,
+}
+
+impl TypeCheckDiagnostics {
+    pub fn new() -> Self {
+        Self { inner: Vec::new() }
+    }
+
+    pub(super) fn push(&mut self, diagnostic: TypeCheckDiagnostic) {
+        self.inner.push(Arc::new(diagnostic));
+    }
+
+    pub(crate) fn shrink_to_fit(&mut self) {
+        self.inner.shrink_to_fit();
+    }
+}
+
+impl Extend<TypeCheckDiagnostic> for TypeCheckDiagnostics {
+    fn extend<T: IntoIterator<Item = TypeCheckDiagnostic>>(&mut self, iter: T) {
+        self.inner.extend(iter.into_iter().map(std::sync::Arc::new));
+    }
+}
+
+impl Extend<std::sync::Arc<TypeCheckDiagnostic>> for TypeCheckDiagnostics {
+    fn extend<T: IntoIterator<Item = Arc<TypeCheckDiagnostic>>>(&mut self, iter: T) {
+        self.inner.extend(iter);
+    }
+}
+
+impl<'a> Extend<&'a std::sync::Arc<TypeCheckDiagnostic>> for TypeCheckDiagnostics {
+    fn extend<T: IntoIterator<Item = &'a Arc<TypeCheckDiagnostic>>>(&mut self, iter: T) {
+        self.inner
+            .extend(iter.into_iter().map(std::sync::Arc::clone));
+    }
+}
+
+impl std::fmt::Debug for TypeCheckDiagnostics {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        self.inner.fmt(f)
+    }
+}
+
+impl Deref for TypeCheckDiagnostics {
+    type Target = [std::sync::Arc<TypeCheckDiagnostic>];
+
+    fn deref(&self) -> &Self::Target {
+        &self.inner
+    }
+}
+
+impl IntoIterator for TypeCheckDiagnostics {
+    type Item = Arc<TypeCheckDiagnostic>;
+    type IntoIter = std::vec::IntoIter<std::sync::Arc<TypeCheckDiagnostic>>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        self.inner.into_iter()
+    }
+}
+
+impl<'a> IntoIterator for &'a TypeCheckDiagnostics {
+    type Item = &'a Arc<TypeCheckDiagnostic>;
+    type IntoIter = std::slice::Iter<'a, std::sync::Arc<TypeCheckDiagnostic>>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        self.inner.iter()
+    }
+}

crates/red_knot_python_semantic/src/types/display.rs

@@ -1,14 +1,22 @@
 //! Display implementations for types.
 
-use std::fmt::{Display, Formatter};
+use std::fmt::{self, Display, Formatter};
+
+use ruff_db::display::FormatterJoinExtension;
+use ruff_python_ast::str::Quote;
+use ruff_python_literal::escape::AsciiEscape;
 
 use crate::types::{IntersectionType, Type, UnionType};
 use crate::Db;
+use rustc_hash::FxHashMap;
 
 impl<'db> Type<'db> {
-    pub fn display(&'db self, db: &'db dyn Db) -> DisplayType<'db> {
+    pub fn display(&self, db: &'db dyn Db) -> DisplayType {
         DisplayType { ty: self, db }
     }
+    fn representation(self, db: &'db dyn Db) -> DisplayRepresentation<'db> {
+        DisplayRepresentation { db, ty: self }
+    }
 }
 
 #[derive(Copy, Clone)]
@@ -18,7 +26,41 @@ pub struct DisplayType<'db> {
 }
 
 impl Display for DisplayType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        let representation = self.ty.representation(self.db);
+        if matches!(
+            self.ty,
+            Type::IntLiteral(_)
+                | Type::BooleanLiteral(_)
+                | Type::StringLiteral(_)
+                | Type::BytesLiteral(_)
+                | Type::Class(_)
+                | Type::Function(_)
+                | Type::RevealTypeFunction(_)
+        ) {
+            write!(f, "Literal[{representation}]",)
+        } else {
+            representation.fmt(f)
+        }
+    }
+}
+
+impl fmt::Debug for DisplayType<'_> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        Display::fmt(self, f)
+    }
+}
+
+/// Writes the string representation of a type, which is the value displayed either as
+/// `Literal[<repr>]` or `Literal[<repr1>, <repr2>]` for literal types or as `<repr>` for
+/// non literals
+struct DisplayRepresentation<'db> {
+    ty: Type<'db>,
+    db: &'db dyn Db,
+}
+
+impl Display for DisplayRepresentation<'_> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
         match self.ty {
             Type::Any => f.write_str("Any"),
             Type::Never => f.write_str("Never"),
@@ -29,25 +71,39 @@ impl Display for DisplayType<'_> {
                 write!(f, "<module '{:?}'>", file.path(self.db))
             }
             // TODO functions and classes should display using a fully qualified name
-            Type::Class(class) => write!(f, "Literal[{}]", class.name(self.db)),
-            Type::Instance(class) => f.write_str(&class.name(self.db)),
-            Type::Function(function) => write!(f, "Literal[{}]", function.name(self.db)),
+            Type::Class(class) => f.write_str(class.name(self.db)),
+            Type::Instance(class) => f.write_str(class.name(self.db)),
+            Type::Function(function) | Type::RevealTypeFunction(function) => {
+                f.write_str(function.name(self.db))
+            }
             Type::Union(union) => union.display(self.db).fmt(f),
             Type::Intersection(intersection) => intersection.display(self.db).fmt(f),
-            Type::IntLiteral(n) => write!(f, "Literal[{n}]"),
-            Type::BooleanLiteral(boolean) => {
-                write!(f, "Literal[{}]", if *boolean { "True" } else { "False" })
+            Type::IntLiteral(n) => n.fmt(f),
+            Type::BooleanLiteral(boolean) => f.write_str(if boolean { "True" } else { "False" }),
+            Type::StringLiteral(string) => {
+                write!(f, r#""{}""#, string.value(self.db).replace('"', r#"\""#))
+            }
+            Type::LiteralString => f.write_str("LiteralString"),
+            Type::BytesLiteral(bytes) => {
+                let escape =
+                    AsciiEscape::with_preferred_quote(bytes.value(self.db).as_ref(), Quote::Double);
+
+                escape.bytes_repr().write(f)
+            }
+            Type::Tuple(tuple) => {
+                f.write_str("tuple[")?;
+                let elements = tuple.elements(self.db);
+                if elements.is_empty() {
+                    f.write_str("()")?;
+                } else {
+                    elements.display(self.db).fmt(f)?;
+                }
+                f.write_str("]")
             }
         }
     }
 }
 
-impl std::fmt::Debug for DisplayType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        std::fmt::Display::fmt(self, f)
-    }
-}
-
 impl<'db> UnionType<'db> {
     fn display(&'db self, db: &'db dyn Db) -> DisplayUnionType<'db> {
         DisplayUnionType { db, ty: self }
@@ -60,54 +116,90 @@ struct DisplayUnionType<'db> {
 }
 
 impl Display for DisplayUnionType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        let union = self.ty;
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        let elements = self.ty.elements(self.db);
 
-        let (int_literals, other_types): (Vec<Type>, Vec<Type>) = union
-            .elements(self.db)
-            .iter()
-            .copied()
-            .partition(|ty| matches!(ty, Type::IntLiteral(_)));
+        // Group literal types by kind.
+        let mut grouped_literals = FxHashMap::default();
 
-        let mut first = true;
-        if !int_literals.is_empty() {
-            f.write_str("Literal[")?;
-            let mut nums: Vec<_> = int_literals
-                .into_iter()
-                .filter_map(|ty| {
-                    if let Type::IntLiteral(n) = ty {
-                        Some(n)
-                    } else {
-                        None
-                    }
-                })
-                .collect();
-            nums.sort_unstable();
-            for num in nums {
-                if !first {
-                    f.write_str(", ")?;
-                }
-                write!(f, "{num}")?;
-                first = false;
+        for element in elements {
+            if let Ok(literal_kind) = LiteralTypeKind::try_from(*element) {
+                grouped_literals
+                    .entry(literal_kind)
+                    .or_insert_with(Vec::new)
+                    .push(*element);
             }
-            f.write_str("]")?;
         }
 
-        for ty in other_types {
-            if !first {
-                f.write_str(" | ")?;
-            };
-            first = false;
-            write!(f, "{}", ty.display(self.db))?;
+        let mut join = f.join(" | ");
+
+        for element in elements {
+            if let Ok(literal_kind) = LiteralTypeKind::try_from(*element) {
+                let Some(mut literals) = grouped_literals.remove(&literal_kind) else {
+                    continue;
+                };
+                if literal_kind == LiteralTypeKind::IntLiteral {
+                    literals.sort_unstable_by_key(|ty| ty.expect_int_literal());
+                }
+                join.entry(&DisplayLiteralGroup {
+                    literals,
+                    db: self.db,
+                });
+            } else {
+                join.entry(&element.display(self.db));
+            }
         }
 
+        join.finish()?;
+
+        debug_assert!(grouped_literals.is_empty());
+
         Ok(())
     }
 }
 
-impl std::fmt::Debug for DisplayUnionType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        std::fmt::Display::fmt(self, f)
+impl fmt::Debug for DisplayUnionType<'_> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        Display::fmt(self, f)
+    }
+}
+
+struct DisplayLiteralGroup<'db> {
+    literals: Vec<Type<'db>>,
+    db: &'db dyn Db,
+}
+
+impl Display for DisplayLiteralGroup<'_> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        f.write_str("Literal[")?;
+        f.join(", ")
+            .entries(self.literals.iter().map(|ty| ty.representation(self.db)))
+            .finish()?;
+        f.write_str("]")
+    }
+}
+
+#[derive(Copy, Clone, Debug, Eq, PartialEq, Hash)]
+enum LiteralTypeKind {
+    Class,
+    Function,
+    IntLiteral,
+    StringLiteral,
+    BytesLiteral,
+}
+
+impl TryFrom<Type<'_>> for LiteralTypeKind {
+    type Error = ();
+
+    fn try_from(value: Type<'_>) -> Result<Self, Self::Error> {
+        match value {
+            Type::Class(_) => Ok(Self::Class),
+            Type::Function(_) | Type::RevealTypeFunction(_) => Ok(Self::Function),
+            Type::IntLiteral(_) => Ok(Self::IntLiteral),
+            Type::StringLiteral(_) => Ok(Self::StringLiteral),
+            Type::BytesLiteral(_) => Ok(Self::BytesLiteral),
+            _ => Err(()),
+        }
     }
 }
 
@@ -123,30 +215,159 @@ struct DisplayIntersectionType<'db> {
 }
 
 impl Display for DisplayIntersectionType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        let mut first = true;
-        for (neg, ty) in self
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        let tys = self
             .ty
             .positive(self.db)
             .iter()
-            .map(|ty| (false, ty))
-            .chain(self.ty.negative(self.db).iter().map(|ty| (true, ty)))
-        {
-            if !first {
-                f.write_str(" & ")?;
-            };
-            first = false;
-            if neg {
-                f.write_str("~")?;
-            };
-            write!(f, "{}", ty.display(self.db))?;
-        }
-        Ok(())
+            .map(|&ty| DisplayMaybeNegatedType {
+                ty,
+                db: self.db,
+                negated: false,
+            })
+            .chain(
+                self.ty
+                    .negative(self.db)
+                    .iter()
+                    .map(|&ty| DisplayMaybeNegatedType {
+                        ty,
+                        db: self.db,
+                        negated: true,
+                    }),
+            );
+        f.join(" & ").entries(tys).finish()
     }
 }
 
-impl std::fmt::Debug for DisplayIntersectionType<'_> {
-    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        std::fmt::Display::fmt(self, f)
+impl fmt::Debug for DisplayIntersectionType<'_> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        Display::fmt(self, f)
+    }
+}
+
+struct DisplayMaybeNegatedType<'db> {
+    ty: Type<'db>,
+    db: &'db dyn Db,
+    negated: bool,
+}
+
+impl<'db> Display for DisplayMaybeNegatedType<'db> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        if self.negated {
+            f.write_str("~")?;
+        }
+        self.ty.display(self.db).fmt(f)
+    }
+}
+
+pub(crate) trait TypeArrayDisplay<'db> {
+    fn display(&self, db: &'db dyn Db) -> DisplayTypeArray;
+}
+
+impl<'db> TypeArrayDisplay<'db> for Box<[Type<'db>]> {
+    fn display(&self, db: &'db dyn Db) -> DisplayTypeArray {
+        DisplayTypeArray { types: self, db }
+    }
+}
+
+impl<'db> TypeArrayDisplay<'db> for Vec<Type<'db>> {
+    fn display(&self, db: &'db dyn Db) -> DisplayTypeArray {
+        DisplayTypeArray { types: self, db }
+    }
+}
+
+pub(crate) struct DisplayTypeArray<'b, 'db> {
+    types: &'b [Type<'db>],
+    db: &'db dyn Db,
+}
+
+impl<'db> Display for DisplayTypeArray<'_, 'db> {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        f.join(", ")
+            .entries(self.types.iter().map(|ty| ty.display(self.db)))
+            .finish()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use ruff_db::files::system_path_to_file;
+    use ruff_db::system::{DbWithTestSystem, SystemPathBuf};
+
+    use crate::db::tests::TestDb;
+    use crate::types::{global_symbol_ty, BytesLiteralType, StringLiteralType, Type, UnionType};
+    use crate::{Program, ProgramSettings, PythonVersion, SearchPathSettings};
+
+    fn setup_db() -> TestDb {
+        let db = TestDb::new();
+
+        let src_root = SystemPathBuf::from("/src");
+        db.memory_file_system()
+            .create_directory_all(&src_root)
+            .unwrap();
+
+        Program::from_settings(
+            &db,
+            &ProgramSettings {
+                target_version: PythonVersion::default(),
+                search_paths: SearchPathSettings::new(src_root),
+            },
+        )
+        .expect("Valid search path settings");
+
+        db
+    }
+
+    #[test]
+    fn test_condense_literal_display_by_type() -> anyhow::Result<()> {
+        let mut db = setup_db();
+
+        db.write_dedented(
+            "src/main.py",
+            "
+            def foo(x: int) -> int:
+                return x + 1
+
+            def bar(s: str) -> str:
+                return s
+
+            class A: ...
+            class B: ...
+            ",
+        )?;
+        let mod_file = system_path_to_file(&db, "src/main.py").expect("Expected file to exist.");
+
+        let union_elements = &[
+            Type::Unknown,
+            Type::IntLiteral(-1),
+            global_symbol_ty(&db, mod_file, "A"),
+            Type::StringLiteral(StringLiteralType::new(&db, Box::from("A"))),
+            Type::BytesLiteral(BytesLiteralType::new(&db, Box::from([0]))),
+            Type::BytesLiteral(BytesLiteralType::new(&db, Box::from([7]))),
+            Type::IntLiteral(0),
+            Type::IntLiteral(1),
+            Type::StringLiteral(StringLiteralType::new(&db, Box::from("B"))),
+            global_symbol_ty(&db, mod_file, "foo"),
+            global_symbol_ty(&db, mod_file, "bar"),
+            global_symbol_ty(&db, mod_file, "B"),
+            Type::BooleanLiteral(true),
+            Type::None,
+        ];
+        let union = UnionType::from_elements(&db, union_elements).expect_union();
+        let display = format!("{}", union.display(&db));
+        assert_eq!(
+            display,
+            concat!(
+                "Unknown | ",
+                "Literal[-1, 0, 1] | ",
+                "Literal[A, B] | ",
+                "Literal[\"A\", \"B\"] | ",
+                "Literal[b\"\\x00\", b\"\\x07\"] | ",
+                "Literal[foo, bar] | ",
+                "Literal[True] | ",
+                "None"
+            )
+        );
+        Ok(())
     }
 }

crates/red_knot_python_semantic/src/types/narrow.rs

@@ -0,0 +1,186 @@
+use crate::semantic_index::ast_ids::HasScopedAstId;
+use crate::semantic_index::constraint::{Constraint, PatternConstraint};
+use crate::semantic_index::definition::Definition;
+use crate::semantic_index::expression::Expression;
+use crate::semantic_index::symbol::{ScopeId, ScopedSymbolId, SymbolTable};
+use crate::semantic_index::symbol_table;
+use crate::types::{infer_expression_types, IntersectionBuilder, Type};
+use crate::Db;
+use ruff_python_ast as ast;
+use rustc_hash::FxHashMap;
+use std::sync::Arc;
+
+/// Return the type constraint that `test` (if true) would place on `definition`, if any.
+///
+/// For example, if we have this code:
+///
+/// ```python
+/// y = 1 if flag else None
+/// x = 1 if flag else None
+/// if x is not None:
+///     ...
+/// ```
+///
+/// The `test` expression `x is not None` places the constraint "not None" on the definition of
+/// `x`, so in that case we'd return `Some(Type::Intersection(negative=[Type::None]))`.
+///
+/// But if we called this with the same `test` expression, but the `definition` of `y`, no
+/// constraint is applied to that definition, so we'd just return `None`.
+pub(crate) fn narrowing_constraint<'db>(
+    db: &'db dyn Db,
+    constraint: Constraint<'db>,
+    definition: Definition<'db>,
+) -> Option<Type<'db>> {
+    match constraint {
+        Constraint::Expression(expression) => {
+            all_narrowing_constraints_for_expression(db, expression)
+                .get(&definition.symbol(db))
+                .copied()
+        }
+        Constraint::Pattern(pattern) => all_narrowing_constraints_for_pattern(db, pattern)
+            .get(&definition.symbol(db))
+            .copied(),
+    }
+}
+
+#[salsa::tracked(return_ref)]
+fn all_narrowing_constraints_for_pattern<'db>(
+    db: &'db dyn Db,
+    pattern: PatternConstraint<'db>,
+) -> NarrowingConstraints<'db> {
+    NarrowingConstraintsBuilder::new(db, Constraint::Pattern(pattern)).finish()
+}
+
+#[salsa::tracked(return_ref)]
+fn all_narrowing_constraints_for_expression<'db>(
+    db: &'db dyn Db,
+    expression: Expression<'db>,
+) -> NarrowingConstraints<'db> {
+    NarrowingConstraintsBuilder::new(db, Constraint::Expression(expression)).finish()
+}
+
+type NarrowingConstraints<'db> = FxHashMap<ScopedSymbolId, Type<'db>>;
+
+struct NarrowingConstraintsBuilder<'db> {
+    db: &'db dyn Db,
+    constraint: Constraint<'db>,
+    constraints: NarrowingConstraints<'db>,
+}
+
+impl<'db> NarrowingConstraintsBuilder<'db> {
+    fn new(db: &'db dyn Db, constraint: Constraint<'db>) -> Self {
+        Self {
+            db,
+            constraint,
+            constraints: NarrowingConstraints::default(),
+        }
+    }
+
+    fn finish(mut self) -> NarrowingConstraints<'db> {
+        match self.constraint {
+            Constraint::Expression(expression) => self.evaluate_expression_constraint(expression),
+            Constraint::Pattern(pattern) => self.evaluate_pattern_constraint(pattern),
+        }
+
+        self.constraints.shrink_to_fit();
+        self.constraints
+    }
+
+    fn evaluate_expression_constraint(&mut self, expression: Expression<'db>) {
+        if let ast::Expr::Compare(expr_compare) = expression.node_ref(self.db).node() {
+            self.add_expr_compare(expr_compare, expression);
+        }
+        // TODO other test expression kinds
+    }
+
+    fn evaluate_pattern_constraint(&mut self, pattern: PatternConstraint<'db>) {
+        let subject = pattern.subject(self.db);
+
+        match pattern.pattern(self.db).node() {
+            ast::Pattern::MatchValue(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchSingleton(singleton_pattern) => {
+                self.add_match_pattern_singleton(subject, singleton_pattern);
+            }
+            ast::Pattern::MatchSequence(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchMapping(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchClass(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchStar(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchAs(_) => {
+                // TODO
+            }
+            ast::Pattern::MatchOr(_) => {
+                // TODO
+            }
+        }
+    }
+
+    fn symbols(&self) -> Arc<SymbolTable> {
+        symbol_table(self.db, self.scope())
+    }
+
+    fn scope(&self) -> ScopeId<'db> {
+        match self.constraint {
+            Constraint::Expression(expression) => expression.scope(self.db),
+            Constraint::Pattern(pattern) => pattern.scope(self.db),
+        }
+    }
+
+    fn add_expr_compare(&mut self, expr_compare: &ast::ExprCompare, expression: Expression<'db>) {
+        let ast::ExprCompare {
+            range: _,
+            left,
+            ops,
+            comparators,
+        } = expr_compare;
+
+        if let ast::Expr::Name(ast::ExprName {
+            range: _,
+            id,
+            ctx: _,
+        }) = left.as_ref()
+        {
+            // SAFETY: we should always have a symbol for every Name node.
+            let symbol = self.symbols().symbol_id_by_name(id).unwrap();
+            let scope = self.scope();
+            let inference = infer_expression_types(self.db, expression);
+            for (op, comparator) in std::iter::zip(&**ops, &**comparators) {
+                let comp_ty = inference.expression_ty(comparator.scoped_ast_id(self.db, scope));
+                if matches!(op, ast::CmpOp::IsNot) {
+                    let ty = IntersectionBuilder::new(self.db)
+                        .add_negative(comp_ty)
+                        .build();
+                    self.constraints.insert(symbol, ty);
+                };
+                // TODO other comparison types
+            }
+        }
+    }
+
+    fn add_match_pattern_singleton(
+        &mut self,
+        subject: &ast::Expr,
+        pattern: &ast::PatternMatchSingleton,
+    ) {
+        if let Some(ast::ExprName { id, .. }) = subject.as_name_expr() {
+            // SAFETY: we should always have a symbol for every Name node.
+            let symbol = self.symbols().symbol_id_by_name(id).unwrap();
+
+            let ty = match pattern.value {
+                ast::Singleton::None => Type::None,
+                ast::Singleton::True => Type::BooleanLiteral(true),
+                ast::Singleton::False => Type::BooleanLiteral(false),
+            };
+            self.constraints.insert(symbol, ty);
+        }
+    }
+}

crates/red_knot_python_semantic/vendor/typeshed/source_commit.txt

@@ -1 +0,0 @@
-4ef2d66663fc080fefa379e6ae5fc45d4f8b54eb

crates/red_knot_python_semantic/vendor/typeshed/stdlib/_locale.pyi

@@ -1,100 +0,0 @@
-import sys
-from _typeshed import StrPath
-from collections.abc import Mapping
-
-LC_CTYPE: int
-LC_COLLATE: int
-LC_TIME: int
-LC_MONETARY: int
-LC_NUMERIC: int
-LC_ALL: int
-CHAR_MAX: int
-
-def setlocale(category: int, locale: str | None = None, /) -> str: ...
-def localeconv() -> Mapping[str, int | str | list[int]]: ...
-
-if sys.version_info >= (3, 11):
-    def getencoding() -> str: ...
-
-def strcoll(os1: str, os2: str, /) -> int: ...
-def strxfrm(string: str, /) -> str: ...
-
-# native gettext functions
-# https://docs.python.org/3/library/locale.html#access-to-message-catalogs
-# https://github.com/python/cpython/blob/f4c03484da59049eb62a9bf7777b963e2267d187/Modules/_localemodule.c#L626
-if sys.platform != "win32":
-    LC_MESSAGES: int
-
-    ABDAY_1: int
-    ABDAY_2: int
-    ABDAY_3: int
-    ABDAY_4: int
-    ABDAY_5: int
-    ABDAY_6: int
-    ABDAY_7: int
-
-    ABMON_1: int
-    ABMON_2: int
-    ABMON_3: int
-    ABMON_4: int
-    ABMON_5: int
-    ABMON_6: int
-    ABMON_7: int
-    ABMON_8: int
-    ABMON_9: int
-    ABMON_10: int
-    ABMON_11: int
-    ABMON_12: int
-
-    DAY_1: int
-    DAY_2: int
-    DAY_3: int
-    DAY_4: int
-    DAY_5: int
-    DAY_6: int
-    DAY_7: int
-
-    ERA: int
-    ERA_D_T_FMT: int
-    ERA_D_FMT: int
-    ERA_T_FMT: int
-
-    MON_1: int
-    MON_2: int
-    MON_3: int
-    MON_4: int
-    MON_5: int
-    MON_6: int
-    MON_7: int
-    MON_8: int
-    MON_9: int
-    MON_10: int
-    MON_11: int
-    MON_12: int
-
-    CODESET: int
-    D_T_FMT: int
-    D_FMT: int
-    T_FMT: int
-    T_FMT_AMPM: int
-    AM_STR: int
-    PM_STR: int
-
-    RADIXCHAR: int
-    THOUSEP: int
-    YESEXPR: int
-    NOEXPR: int
-    CRNCYSTR: int
-    ALT_DIGITS: int
-
-    def nl_langinfo(key: int, /) -> str: ...
-
-    # This is dependent on `libintl.h` which is a part of `gettext`
-    # system dependency. These functions might be missing.
-    # But, we always say that they are present.
-    def gettext(msg: str, /) -> str: ...
-    def dgettext(domain: str | None, msg: str, /) -> str: ...
-    def dcgettext(domain: str | None, msg: str, category: int, /) -> str: ...
-    def textdomain(domain: str | None, /) -> str: ...
-    def bindtextdomain(domain: str, dir: StrPath | None, /) -> str: ...
-    def bind_textdomain_codeset(domain: str, codeset: str | None, /) -> str | None: ...

crates/red_knot_python_semantic/vendor/typeshed/stdlib/distutils/cmd.pyi

@@ -1,108 +0,0 @@
-from _typeshed import BytesPath, Incomplete, StrOrBytesPath, StrPath, Unused
-from abc import abstractmethod
-from collections.abc import Callable, Iterable
-from distutils.dist import Distribution
-from distutils.file_util import _BytesPathT, _StrPathT
-from typing import Any, ClassVar, Literal, TypeVar, overload
-from typing_extensions import TypeVarTuple, Unpack
-
-_CommandT = TypeVar("_CommandT", bound=Command)
-_Ts = TypeVarTuple("_Ts")
-
-class Command:
-    distribution: Distribution
-    # Any to work around variance issues
-    sub_commands: ClassVar[list[tuple[str, Callable[[Any], bool] | None]]]
-    def __init__(self, dist: Distribution) -> None: ...
-    @abstractmethod
-    def initialize_options(self) -> None: ...
-    @abstractmethod
-    def finalize_options(self) -> None: ...
-    @abstractmethod
-    def run(self) -> None: ...
-    def announce(self, msg: str, level: int = 1) -> None: ...
-    def debug_print(self, msg: str) -> None: ...
-    def ensure_string(self, option: str, default: str | None = None) -> None: ...
-    def ensure_string_list(self, option: str) -> None: ...
-    def ensure_filename(self, option: str) -> None: ...
-    def ensure_dirname(self, option: str) -> None: ...
-    def get_command_name(self) -> str: ...
-    def set_undefined_options(self, src_cmd: str, *option_pairs: tuple[str, str]) -> None: ...
-    def get_finalized_command(self, command: str, create: bool | Literal[0, 1] = 1) -> Command: ...
-    @overload
-    def reinitialize_command(self, command: str, reinit_subcommands: bool | Literal[0, 1] = 0) -> Command: ...
-    @overload
-    def reinitialize_command(self, command: _CommandT, reinit_subcommands: bool | Literal[0, 1] = 0) -> _CommandT: ...
-    def run_command(self, command: str) -> None: ...
-    def get_sub_commands(self) -> list[str]: ...
-    def warn(self, msg: str) -> None: ...
-    def execute(
-        self, func: Callable[[Unpack[_Ts]], Unused], args: tuple[Unpack[_Ts]], msg: str | None = None, level: int = 1
-    ) -> None: ...
-    def mkpath(self, name: str, mode: int = 0o777) -> None: ...
-    @overload
-    def copy_file(
-        self,
-        infile: StrPath,
-        outfile: _StrPathT,
-        preserve_mode: bool | Literal[0, 1] = 1,
-        preserve_times: bool | Literal[0, 1] = 1,
-        link: str | None = None,
-        level: Unused = 1,
-    ) -> tuple[_StrPathT | str, bool]: ...
-    @overload
-    def copy_file(
-        self,
-        infile: BytesPath,
-        outfile: _BytesPathT,
-        preserve_mode: bool | Literal[0, 1] = 1,
-        preserve_times: bool | Literal[0, 1] = 1,
-        link: str | None = None,
-        level: Unused = 1,
-    ) -> tuple[_BytesPathT | bytes, bool]: ...
-    def copy_tree(
-        self,
-        infile: StrPath,
-        outfile: str,
-        preserve_mode: bool | Literal[0, 1] = 1,
-        preserve_times: bool | Literal[0, 1] = 1,
-        preserve_symlinks: bool | Literal[0, 1] = 0,
-        level: Unused = 1,
-    ) -> list[str]: ...
-    @overload
-    def move_file(self, src: StrPath, dst: _StrPathT, level: Unused = 1) -> _StrPathT | str: ...
-    @overload
-    def move_file(self, src: BytesPath, dst: _BytesPathT, level: Unused = 1) -> _BytesPathT | bytes: ...
-    def spawn(self, cmd: Iterable[str], search_path: bool | Literal[0, 1] = 1, level: Unused = 1) -> None: ...
-    @overload
-    def make_archive(
-        self,
-        base_name: str,
-        format: str,
-        root_dir: StrOrBytesPath | None = None,
-        base_dir: str | None = None,
-        owner: str | None = None,
-        group: str | None = None,
-    ) -> str: ...
-    @overload
-    def make_archive(
-        self,
-        base_name: StrPath,
-        format: str,
-        root_dir: StrOrBytesPath,
-        base_dir: str | None = None,
-        owner: str | None = None,
-        group: str | None = None,
-    ) -> str: ...
-    def make_file(
-        self,
-        infiles: str | list[str] | tuple[str, ...],
-        outfile: StrOrBytesPath,
-        func: Callable[[Unpack[_Ts]], Unused],
-        args: tuple[Unpack[_Ts]],
-        exec_msg: str | None = None,
-        skip_msg: str | None = None,
-        level: Unused = 1,
-    ) -> None: ...
-    def ensure_finalized(self) -> None: ...
-    def dump_options(self, header: Incomplete | None = None, indent: str = "") -> None: ...

crates/red_knot_python_semantic/vendor/typeshed/stdlib/distutils/debug.pyi

@@ -1 +0,0 @@
-DEBUG: bool | None

crates/red_knot_python_semantic/vendor/typeshed/stdlib/distutils/dist.pyi

@@ -1,149 +0,0 @@
-from _typeshed import Incomplete, StrOrBytesPath, StrPath, SupportsWrite
-from collections.abc import Iterable, MutableMapping
-from distutils.cmd import Command
-from re import Pattern
-from typing import IO, ClassVar, Literal, TypeVar, overload
-from typing_extensions import TypeAlias
-
-command_re: Pattern[str]
-
-_OptionsList: TypeAlias = list[tuple[str, str | None, str, int] | tuple[str, str | None, str]]
-_CommandT = TypeVar("_CommandT", bound=Command)
-
-class DistributionMetadata:
-    def __init__(self, path: StrOrBytesPath | None = None) -> None: ...
-    name: str | None
-    version: str | None
-    author: str | None
-    author_email: str | None
-    maintainer: str | None
-    maintainer_email: str | None
-    url: str | None
-    license: str | None
-    description: str | None
-    long_description: str | None
-    keywords: str | list[str] | None
-    platforms: str | list[str] | None
-    classifiers: str | list[str] | None
-    download_url: str | None
-    provides: list[str] | None
-    requires: list[str] | None
-    obsoletes: list[str] | None
-    def read_pkg_file(self, file: IO[str]) -> None: ...
-    def write_pkg_info(self, base_dir: StrPath) -> None: ...
-    def write_pkg_file(self, file: SupportsWrite[str]) -> None: ...
-    def get_name(self) -> str: ...
-    def get_version(self) -> str: ...
-    def get_fullname(self) -> str: ...
-    def get_author(self) -> str: ...
-    def get_author_email(self) -> str: ...
-    def get_maintainer(self) -> str: ...
-    def get_maintainer_email(self) -> str: ...
-    def get_contact(self) -> str: ...
-    def get_contact_email(self) -> str: ...
-    def get_url(self) -> str: ...
-    def get_license(self) -> str: ...
-    def get_licence(self) -> str: ...
-    def get_description(self) -> str: ...
-    def get_long_description(self) -> str: ...
-    def get_keywords(self) -> str | list[str]: ...
-    def get_platforms(self) -> str | list[str]: ...
-    def get_classifiers(self) -> str | list[str]: ...
-    def get_download_url(self) -> str: ...
-    def get_requires(self) -> list[str]: ...
-    def set_requires(self, value: Iterable[str]) -> None: ...
-    def get_provides(self) -> list[str]: ...
-    def set_provides(self, value: Iterable[str]) -> None: ...
-    def get_obsoletes(self) -> list[str]: ...
-    def set_obsoletes(self, value: Iterable[str]) -> None: ...
-
-class Distribution:
-    cmdclass: dict[str, type[Command]]
-    metadata: DistributionMetadata
-    def __init__(self, attrs: MutableMapping[str, Incomplete] | None = None) -> None: ...
-    def get_option_dict(self, command: str) -> dict[str, tuple[str, str]]: ...
-    def parse_config_files(self, filenames: Iterable[str] | None = None) -> None: ...
-    @overload
-    def get_command_obj(self, command: str, create: Literal[1, True] = 1) -> Command: ...
-    @overload
-    def get_command_obj(self, command: str, create: Literal[0, False]) -> Command | None: ...
-    global_options: ClassVar[_OptionsList]
-    common_usage: ClassVar[str]
-    display_options: ClassVar[_OptionsList]
-    display_option_names: ClassVar[list[str]]
-    negative_opt: ClassVar[dict[str, str]]
-    verbose: int
-    dry_run: int
-    help: int
-    command_packages: list[str] | None
-    script_name: str | None
-    script_args: list[str] | None
-    command_options: dict[str, dict[str, tuple[str, str]]]
-    dist_files: list[tuple[str, str, str]]
-    packages: Incomplete
-    package_data: dict[str, list[str]]
-    package_dir: Incomplete
-    py_modules: Incomplete
-    libraries: Incomplete
-    headers: Incomplete
-    ext_modules: Incomplete
-    ext_package: Incomplete
-    include_dirs: Incomplete
-    extra_path: Incomplete
-    scripts: Incomplete
-    data_files: Incomplete
-    password: str
-    command_obj: Incomplete
-    have_run: Incomplete
-    want_user_cfg: bool
-    def dump_option_dicts(
-        self, header: Incomplete | None = None, commands: Incomplete | None = None, indent: str = ""
-    ) -> None: ...
-    def find_config_files(self): ...
-    commands: Incomplete
-    def parse_command_line(self): ...
-    def finalize_options(self) -> None: ...
-    def handle_display_options(self, option_order): ...
-    def print_command_list(self, commands, header, max_length) -> None: ...
-    def print_commands(self) -> None: ...
-    def get_command_list(self): ...
-    def get_command_packages(self): ...
-    def get_command_class(self, command: str) -> type[Command]: ...
-    @overload
-    def reinitialize_command(self, command: str, reinit_subcommands: bool = False) -> Command: ...
-    @overload
-    def reinitialize_command(self, command: _CommandT, reinit_subcommands: bool = False) -> _CommandT: ...
-    def announce(self, msg, level: int = 2) -> None: ...
-    def run_commands(self) -> None: ...
-    def run_command(self, command: str) -> None: ...
-    def has_pure_modules(self) -> bool: ...
-    def has_ext_modules(self) -> bool: ...
-    def has_c_libraries(self) -> bool: ...
-    def has_modules(self) -> bool: ...
-    def has_headers(self) -> bool: ...
-    def has_scripts(self) -> bool: ...
-    def has_data_files(self) -> bool: ...
-    def is_pure(self) -> bool: ...
-
-    # Getter methods generated in __init__
-    def get_name(self) -> str: ...
-    def get_version(self) -> str: ...
-    def get_fullname(self) -> str: ...
-    def get_author(self) -> str: ...
-    def get_author_email(self) -> str: ...
-    def get_maintainer(self) -> str: ...
-    def get_maintainer_email(self) -> str: ...
-    def get_contact(self) -> str: ...
-    def get_contact_email(self) -> str: ...
-    def get_url(self) -> str: ...
-    def get_license(self) -> str: ...
-    def get_licence(self) -> str: ...
-    def get_description(self) -> str: ...
-    def get_long_description(self) -> str: ...
-    def get_keywords(self) -> str | list[str]: ...
-    def get_platforms(self) -> str | list[str]: ...
-    def get_classifiers(self) -> str | list[str]: ...
-    def get_download_url(self) -> str: ...
-    def get_requires(self) -> list[str]: ...
-    def get_provides(self) -> list[str]: ...
-    def get_obsoletes(self) -> list[str]: ...

crates/red_knot_python_semantic/vendor/typeshed/stdlib/pyexpat/errors.pyi

@@ -1,49 +0,0 @@
-import sys
-
-codes: dict[str, int]
-messages: dict[int, str]
-
-XML_ERROR_ABORTED: str
-XML_ERROR_ASYNC_ENTITY: str
-XML_ERROR_ATTRIBUTE_EXTERNAL_ENTITY_REF: str
-XML_ERROR_BAD_CHAR_REF: str
-XML_ERROR_BINARY_ENTITY_REF: str
-XML_ERROR_CANT_CHANGE_FEATURE_ONCE_PARSING: str
-XML_ERROR_DUPLICATE_ATTRIBUTE: str
-XML_ERROR_ENTITY_DECLARED_IN_PE: str
-XML_ERROR_EXTERNAL_ENTITY_HANDLING: str
-XML_ERROR_FEATURE_REQUIRES_XML_DTD: str
-XML_ERROR_FINISHED: str
-XML_ERROR_INCOMPLETE_PE: str
-XML_ERROR_INCORRECT_ENCODING: str
-XML_ERROR_INVALID_TOKEN: str
-XML_ERROR_JUNK_AFTER_DOC_ELEMENT: str
-XML_ERROR_MISPLACED_XML_PI: str
-XML_ERROR_NOT_STANDALONE: str
-XML_ERROR_NOT_SUSPENDED: str
-XML_ERROR_NO_ELEMENTS: str
-XML_ERROR_NO_MEMORY: str
-XML_ERROR_PARAM_ENTITY_REF: str
-XML_ERROR_PARTIAL_CHAR: str
-XML_ERROR_PUBLICID: str
-XML_ERROR_RECURSIVE_ENTITY_REF: str
-XML_ERROR_SUSPENDED: str
-XML_ERROR_SUSPEND_PE: str
-XML_ERROR_SYNTAX: str
-XML_ERROR_TAG_MISMATCH: str
-XML_ERROR_TEXT_DECL: str
-XML_ERROR_UNBOUND_PREFIX: str
-XML_ERROR_UNCLOSED_CDATA_SECTION: str
-XML_ERROR_UNCLOSED_TOKEN: str
-XML_ERROR_UNDECLARING_PREFIX: str
-XML_ERROR_UNDEFINED_ENTITY: str
-XML_ERROR_UNEXPECTED_STATE: str
-XML_ERROR_UNKNOWN_ENCODING: str
-XML_ERROR_XML_DECL: str
-if sys.version_info >= (3, 11):
-    XML_ERROR_RESERVED_PREFIX_XML: str
-    XML_ERROR_RESERVED_PREFIX_XMLNS: str
-    XML_ERROR_RESERVED_NAMESPACE_URI: str
-    XML_ERROR_INVALID_ARGUMENT: str
-    XML_ERROR_NO_BUFFER: str
-    XML_ERROR_AMPLIFICATION_LIMIT_BREACH: str

crates/red_knot_python_semantic/vendor/typeshed/stdlib/pyexpat/model.pyi

@@ -1,11 +0,0 @@
-XML_CTYPE_ANY: int
-XML_CTYPE_CHOICE: int
-XML_CTYPE_EMPTY: int
-XML_CTYPE_MIXED: int
-XML_CTYPE_NAME: int
-XML_CTYPE_SEQ: int
-
-XML_CQUANT_NONE: int
-XML_CQUANT_OPT: int
-XML_CQUANT_PLUS: int
-XML_CQUANT_REP: int

crates/red_knot_server/Cargo.toml

@@ -11,10 +11,8 @@ repository = { workspace = true }
 license = { workspace = true }
 
 [dependencies]
-red_knot_python_semantic = { workspace = true }
 red_knot_workspace = { workspace = true }
 ruff_db = { workspace = true }
-ruff_linter = { workspace = true }
 ruff_notebook = { workspace = true }
 ruff_python_ast = { workspace = true }
 ruff_source_file = { workspace = true }

crates/red_knot_server/src/edit/notebook.rs

@@ -1,5 +1,6 @@
 use anyhow::Ok;
 use lsp_types::NotebookCellKind;
+use ruff_notebook::CellMetadata;
 use rustc_hash::{FxBuildHasher, FxHashMap};
 
 use crate::{PositionEncoding, TextDocument};
@@ -65,7 +66,7 @@ impl NotebookDocument {
                 NotebookCellKind::Code => ruff_notebook::Cell::Code(ruff_notebook::CodeCell {
                     execution_count: None,
                     id: None,
-                    metadata: serde_json::Value::Null,
+                    metadata: CellMetadata::default(),
                     outputs: vec![],
                     source: ruff_notebook::SourceValue::String(
                         cell.document.contents().to_string(),
@@ -75,7 +76,7 @@ impl NotebookDocument {
                     ruff_notebook::Cell::Markdown(ruff_notebook::MarkdownCell {
                         attachments: None,
                         id: None,
-                        metadata: serde_json::Value::Null,
+                        metadata: CellMetadata::default(),
                         source: ruff_notebook::SourceValue::String(
                             cell.document.contents().to_string(),
                         ),

crates/red_knot_server/src/server.rs

@@ -3,11 +3,11 @@
 use std::num::NonZeroUsize;
 use std::panic::PanicInfo;
 
-use lsp_server as lsp;
-use lsp_types as types;
+use lsp_server::Message;
 use lsp_types::{
-    ClientCapabilities, DiagnosticOptions, NotebookCellSelector, NotebookDocumentSyncOptions,
-    NotebookSelector, TextDocumentSyncCapability, TextDocumentSyncOptions,
+    ClientCapabilities, DiagnosticOptions, DiagnosticServerCapabilities, MessageType,
+    ServerCapabilities, TextDocumentSyncCapability, TextDocumentSyncKind, TextDocumentSyncOptions,
+    Url,
 };
 
 use self::connection::{Connection, ConnectionInitializer};
@@ -74,7 +74,7 @@ impl Server {
             init_params.client_info.as_ref(),
         );
 
-        let mut workspace_for_url = |url: lsp_types::Url| {
+        let mut workspace_for_url = |url: Url| {
             let Some(workspace_settings) = workspace_settings.as_mut() else {
                 return (url, ClientSettings::default());
             };
@@ -93,13 +93,18 @@ impl Server {
             }).collect())
             .or_else(|| {
                 tracing::warn!("No workspace(s) were provided during initialization. Using the current working directory as a default workspace...");
-                let uri = types::Url::from_file_path(std::env::current_dir().ok()?).ok()?;
+                let uri = Url::from_file_path(std::env::current_dir().ok()?).ok()?;
                 Some(vec![workspace_for_url(uri)])
             })
             .ok_or_else(|| {
                 anyhow::anyhow!("Failed to get the current working directory while creating a default workspace.")
             })?;
 
+        if workspaces.len() > 1 {
+            // TODO(dhruvmanila): Support multi-root workspaces
+            anyhow::bail!("Multi-root workspaces are not supported yet");
+        }
+
         Ok(Self {
             connection,
             worker_threads,
@@ -149,7 +154,7 @@ impl Server {
             try_show_message(
                 "The Ruff language server exited with a panic. See the logs for more details."
                     .to_string(),
-                lsp_types::MessageType::ERROR,
+                MessageType::ERROR,
             )
             .ok();
         }));
@@ -182,9 +187,9 @@ impl Server {
                 break;
             }
             let task = match msg {
-                lsp::Message::Request(req) => api::request(req),
-                lsp::Message::Notification(notification) => api::notification(notification),
-                lsp::Message::Response(response) => scheduler.response(response),
+                Message::Request(req) => api::request(req),
+                Message::Notification(notification) => api::notification(notification),
+                Message::Response(response) => scheduler.response(response),
             };
             scheduler.dispatch(task);
         }
@@ -206,28 +211,17 @@ impl Server {
             .unwrap_or_default()
     }
 
-    fn server_capabilities(position_encoding: PositionEncoding) -> types::ServerCapabilities {
-        types::ServerCapabilities {
+    fn server_capabilities(position_encoding: PositionEncoding) -> ServerCapabilities {
+        ServerCapabilities {
             position_encoding: Some(position_encoding.into()),
-            diagnostic_provider: Some(types::DiagnosticServerCapabilities::Options(
-                DiagnosticOptions {
-                    identifier: Some(crate::DIAGNOSTIC_NAME.into()),
-                    ..Default::default()
-                },
-            )),
-            notebook_document_sync: Some(types::OneOf::Left(NotebookDocumentSyncOptions {
-                save: Some(false),
-                notebook_selector: [NotebookSelector::ByCells {
-                    notebook: None,
-                    cells: vec![NotebookCellSelector {
-                        language: "python".to_string(),
-                    }],
-                }]
-                .to_vec(),
+            diagnostic_provider: Some(DiagnosticServerCapabilities::Options(DiagnosticOptions {
+                identifier: Some(crate::DIAGNOSTIC_NAME.into()),
+                ..Default::default()
             })),
             text_document_sync: Some(TextDocumentSyncCapability::Options(
                 TextDocumentSyncOptions {
                     open_close: Some(true),
+                    change: Some(TextDocumentSyncKind::INCREMENTAL),
                     ..Default::default()
                 },
             )),

crates/red_knot_server/src/server/api.rs

@@ -1,13 +1,15 @@
-use crate::{server::schedule::Task, session::Session, system::url_to_system_path};
 use lsp_server as server;
 
+use crate::server::schedule::Task;
+use crate::session::Session;
+use crate::system::{url_to_any_system_path, AnySystemPath};
+
 mod diagnostics;
 mod notifications;
 mod requests;
 mod traits;
 
 use notifications as notification;
-use red_knot_workspace::db::RootDatabase;
 use requests as request;
 
 use self::traits::{NotificationHandler, RequestHandler};
@@ -43,6 +45,7 @@ pub(super) fn notification<'a>(notif: server::Notification) -> Task<'a> {
     match notif.method.as_str() {
         notification::DidCloseTextDocumentHandler::METHOD => local_notification_task::<notification::DidCloseTextDocumentHandler>(notif),
         notification::DidOpenTextDocumentHandler::METHOD => local_notification_task::<notification::DidOpenTextDocumentHandler>(notif),
+        notification::DidChangeTextDocumentHandler::METHOD => local_notification_task::<notification::DidChangeTextDocumentHandler>(notif),
         notification::DidOpenNotebookHandler::METHOD => {
             local_notification_task::<notification::DidOpenNotebookHandler>(notif)
         }
@@ -82,12 +85,18 @@ fn background_request_task<'a, R: traits::BackgroundDocumentRequestHandler>(
     Ok(Task::background(schedule, move |session: &Session| {
         let url = R::document_url(&params).into_owned();
 
-        let Ok(path) = url_to_system_path(&url) else {
+        let Ok(path) = url_to_any_system_path(&url) else {
             return Box::new(|_, _| {});
         };
-        let db = session
-            .workspace_db_for_path(path.as_std_path())
-            .map(RootDatabase::snapshot);
+        let db = match path {
+            AnySystemPath::System(path) => {
+                match session.workspace_db_for_path(path.as_std_path()) {
+                    Some(db) => db.snapshot(),
+                    None => session.default_workspace_db().snapshot(),
+                }
+            }
+            AnySystemPath::SystemVirtual(_) => session.default_workspace_db().snapshot(),
+        };
 
         let Some(snapshot) = session.take_snapshot(url) else {
             return Box::new(|_, _| {});

crates/red_knot_server/src/server/api/notifications.rs

@@ -1,9 +1,11 @@
+mod did_change;
 mod did_close;
 mod did_close_notebook;
 mod did_open;
 mod did_open_notebook;
 mod set_trace;
 
+pub(super) use did_change::DidChangeTextDocumentHandler;
 pub(super) use did_close::DidCloseTextDocumentHandler;
 pub(super) use did_close_notebook::DidCloseNotebookHandler;
 pub(super) use did_open::DidOpenTextDocumentHandler;

crates/red_knot_server/src/server/api/notifications/did_change.rs

@@ -0,0 +1,55 @@
+use lsp_server::ErrorCode;
+use lsp_types::notification::DidChangeTextDocument;
+use lsp_types::DidChangeTextDocumentParams;
+
+use red_knot_workspace::watch::ChangeEvent;
+
+use crate::server::api::traits::{NotificationHandler, SyncNotificationHandler};
+use crate::server::api::LSPResult;
+use crate::server::client::{Notifier, Requester};
+use crate::server::Result;
+use crate::session::Session;
+use crate::system::{url_to_any_system_path, AnySystemPath};
+
+pub(crate) struct DidChangeTextDocumentHandler;
+
+impl NotificationHandler for DidChangeTextDocumentHandler {
+    type NotificationType = DidChangeTextDocument;
+}
+
+impl SyncNotificationHandler for DidChangeTextDocumentHandler {
+    fn run(
+        session: &mut Session,
+        _notifier: Notifier,
+        _requester: &mut Requester,
+        params: DidChangeTextDocumentParams,
+    ) -> Result<()> {
+        let Ok(path) = url_to_any_system_path(&params.text_document.uri) else {
+            return Ok(());
+        };
+
+        let key = session.key_from_url(params.text_document.uri);
+
+        session
+            .update_text_document(&key, params.content_changes, params.text_document.version)
+            .with_failure_code(ErrorCode::InternalError)?;
+
+        match path {
+            AnySystemPath::System(path) => {
+                let db = match session.workspace_db_for_path_mut(path.as_std_path()) {
+                    Some(db) => db,
+                    None => session.default_workspace_db_mut(),
+                };
+                db.apply_changes(vec![ChangeEvent::file_content_changed(path)], None);
+            }
+            AnySystemPath::SystemVirtual(virtual_path) => {
+                let db = session.default_workspace_db_mut();
+                db.apply_changes(vec![ChangeEvent::ChangedVirtual(virtual_path)], None);
+            }
+        }
+
+        // TODO(dhruvmanila): Publish diagnostics if the client doesnt support pull diagnostics
+
+        Ok(())
+    }
+}

crates/red_knot_server/src/server/api/notifications/did_close.rs

@@ -1,8 +1,7 @@
 use lsp_server::ErrorCode;
 use lsp_types::notification::DidCloseTextDocument;
 use lsp_types::DidCloseTextDocumentParams;
-
-use ruff_db::files::File;
+use red_knot_workspace::watch::ChangeEvent;
 
 use crate::server::api::diagnostics::clear_diagnostics;
 use crate::server::api::traits::{NotificationHandler, SyncNotificationHandler};
@@ -10,7 +9,7 @@ use crate::server::api::LSPResult;
 use crate::server::client::{Notifier, Requester};
 use crate::server::Result;
 use crate::session::Session;
-use crate::system::url_to_system_path;
+use crate::system::{url_to_any_system_path, AnySystemPath};
 
 pub(crate) struct DidCloseTextDocumentHandler;
 
@@ -25,7 +24,7 @@ impl SyncNotificationHandler for DidCloseTextDocumentHandler {
         _requester: &mut Requester,
         params: DidCloseTextDocumentParams,
     ) -> Result<()> {
-        let Ok(path) = url_to_system_path(&params.text_document.uri) else {
+        let Ok(path) = url_to_any_system_path(&params.text_document.uri) else {
             return Ok(());
         };
 
@@ -34,8 +33,9 @@ impl SyncNotificationHandler for DidCloseTextDocumentHandler {
             .close_document(&key)
             .with_failure_code(ErrorCode::InternalError)?;
 
-        if let Some(db) = session.workspace_db_for_path_mut(path.as_std_path()) {
-            File::sync_path(db, &path);
+        if let AnySystemPath::SystemVirtual(virtual_path) = path {
+            let db = session.default_workspace_db_mut();
+            db.apply_changes(vec![ChangeEvent::DeletedVirtual(virtual_path)], None);
         }
 
         clear_diagnostics(key.url(), &notifier)?;

crates/red_knot_server/src/server/api/notifications/did_close_notebook.rs

@@ -1,14 +1,14 @@
 use lsp_types::notification::DidCloseNotebookDocument;
 use lsp_types::DidCloseNotebookDocumentParams;
 
-use ruff_db::files::File;
+use red_knot_workspace::watch::ChangeEvent;
 
 use crate::server::api::traits::{NotificationHandler, SyncNotificationHandler};
 use crate::server::api::LSPResult;
 use crate::server::client::{Notifier, Requester};
 use crate::server::Result;
 use crate::session::Session;
-use crate::system::url_to_system_path;
+use crate::system::{url_to_any_system_path, AnySystemPath};
 
 pub(crate) struct DidCloseNotebookHandler;
 
@@ -23,7 +23,7 @@ impl SyncNotificationHandler for DidCloseNotebookHandler {
         _requester: &mut Requester,
         params: DidCloseNotebookDocumentParams,
     ) -> Result<()> {
-        let Ok(path) = url_to_system_path(&params.notebook_document.uri) else {
+        let Ok(path) = url_to_any_system_path(&params.notebook_document.uri) else {
             return Ok(());
         };
 
@@ -32,8 +32,9 @@ impl SyncNotificationHandler for DidCloseNotebookHandler {
             .close_document(&key)
             .with_failure_code(lsp_server::ErrorCode::InternalError)?;
 
-        if let Some(db) = session.workspace_db_for_path_mut(path.as_std_path()) {
-            File::sync_path(db, &path);
+        if let AnySystemPath::SystemVirtual(virtual_path) = path {
+            let db = session.default_workspace_db_mut();
+            db.apply_changes(vec![ChangeEvent::DeletedVirtual(virtual_path)], None);
         }
 
         Ok(())

crates/red_knot_server/src/server/api/notifications/did_open.rs

@@ -1,13 +1,14 @@
 use lsp_types::notification::DidOpenTextDocument;
 use lsp_types::DidOpenTextDocumentParams;
 
-use ruff_db::files::system_path_to_file;
+use red_knot_workspace::watch::ChangeEvent;
+use ruff_db::Db;
 
 use crate::server::api::traits::{NotificationHandler, SyncNotificationHandler};
 use crate::server::client::{Notifier, Requester};
 use crate::server::Result;
 use crate::session::Session;
-use crate::system::url_to_system_path;
+use crate::system::{url_to_any_system_path, AnySystemPath};
 use crate::TextDocument;
 
 pub(crate) struct DidOpenTextDocumentHandler;
@@ -23,17 +24,25 @@ impl SyncNotificationHandler for DidOpenTextDocumentHandler {
         _requester: &mut Requester,
         params: DidOpenTextDocumentParams,
     ) -> Result<()> {
-        let Ok(path) = url_to_system_path(&params.text_document.uri) else {
+        let Ok(path) = url_to_any_system_path(&params.text_document.uri) else {
             return Ok(());
         };
 
         let document = TextDocument::new(params.text_document.text, params.text_document.version);
         session.open_text_document(params.text_document.uri, document);
 
-        if let Some(db) = session.workspace_db_for_path_mut(path.as_std_path()) {
-            // TODO(dhruvmanila): Store the `file` in `DocumentController`
-            let file = system_path_to_file(db, &path).unwrap();
-            file.sync(db);
+        match path {
+            AnySystemPath::System(path) => {
+                let db = match session.workspace_db_for_path_mut(path.as_std_path()) {
+                    Some(db) => db,
+                    None => session.default_workspace_db_mut(),
+                };
+                db.apply_changes(vec![ChangeEvent::Opened(path)], None);
+            }
+            AnySystemPath::SystemVirtual(virtual_path) => {
+                let db = session.default_workspace_db_mut();
+                db.files().virtual_file(db, &virtual_path);
+            }
         }
 
         // TODO(dhruvmanila): Publish diagnostics if the client doesn't support pull diagnostics

crates/red_knot_server/src/server/api/notifications/did_open_notebook.rs

@@ -2,7 +2,8 @@ use lsp_server::ErrorCode;
 use lsp_types::notification::DidOpenNotebookDocument;
 use lsp_types::DidOpenNotebookDocumentParams;
 
-use ruff_db::files::system_path_to_file;
+use red_knot_workspace::watch::ChangeEvent;
+use ruff_db::Db;
 
 use crate::edit::NotebookDocument;
 use crate::server::api::traits::{NotificationHandler, SyncNotificationHandler};
@@ -10,7 +11,7 @@ use crate::server::api::LSPResult;
 use crate::server::client::{Notifier, Requester};
 use crate::server::Result;
 use crate::session::Session;
-use crate::system::url_to_system_path;
+use crate::system::{url_to_any_system_path, AnySystemPath};
 
 pub(crate) struct DidOpenNotebookHandler;
 
@@ -25,7 +26,7 @@ impl SyncNotificationHandler for DidOpenNotebookHandler {
         _requester: &mut Requester,
         params: DidOpenNotebookDocumentParams,
     ) -> Result<()> {
-        let Ok(path) = url_to_system_path(&params.notebook_document.uri) else {
+        let Ok(path) = url_to_any_system_path(&params.notebook_document.uri) else {
             return Ok(());
         };
 
@@ -38,10 +39,18 @@ impl SyncNotificationHandler for DidOpenNotebookHandler {
         .with_failure_code(ErrorCode::InternalError)?;
         session.open_notebook_document(params.notebook_document.uri.clone(), notebook);
 
-        if let Some(db) = session.workspace_db_for_path_mut(path.as_std_path()) {
-            // TODO(dhruvmanila): Store the `file` in `DocumentController`
-            let file = system_path_to_file(db, &path).unwrap();
-            file.sync(db);
+        match path {
+            AnySystemPath::System(path) => {
+                let db = match session.workspace_db_for_path_mut(path.as_std_path()) {
+                    Some(db) => db,
+                    None => session.default_workspace_db_mut(),
+                };
+                db.apply_changes(vec![ChangeEvent::Opened(path)], None);
+            }
+            AnySystemPath::SystemVirtual(virtual_path) => {
+                let db = session.default_workspace_db_mut();
+                db.files().virtual_file(db, &virtual_path);
+            }
         }
 
         // TODO(dhruvmanila): Publish diagnostics if the client doesn't support pull diagnostics

crates/red_knot_server/src/server/api/requests/diagnostic.rs

@@ -2,8 +2,9 @@ use std::borrow::Cow;
 
 use lsp_types::request::DocumentDiagnosticRequest;
 use lsp_types::{
-    Diagnostic, DocumentDiagnosticParams, DocumentDiagnosticReport, DocumentDiagnosticReportResult,
-    FullDocumentDiagnosticReport, Range, RelatedFullDocumentDiagnosticReport, Url,
+    Diagnostic, DiagnosticSeverity, DocumentDiagnosticParams, DocumentDiagnosticReport,
+    DocumentDiagnosticReportResult, FullDocumentDiagnosticReport, Position, Range,
+    RelatedFullDocumentDiagnosticReport, Url,
 };
 
 use red_knot_workspace::db::RootDatabase;
@@ -25,13 +26,11 @@ impl BackgroundDocumentRequestHandler for DocumentDiagnosticRequestHandler {
 
     fn run_with_snapshot(
         snapshot: DocumentSnapshot,
-        db: Option<RootDatabase>,
+        db: RootDatabase,
         _notifier: Notifier,
         _params: DocumentDiagnosticParams,
     ) -> Result<DocumentDiagnosticReportResult> {
-        let diagnostics = db
-            .map(|db| compute_diagnostics(&snapshot, &db))
-            .unwrap_or_default();
+        let diagnostics = compute_diagnostics(&snapshot, &db);
 
         Ok(DocumentDiagnosticReportResult::Report(
             DocumentDiagnosticReport::Full(RelatedFullDocumentDiagnosticReport {
@@ -47,25 +46,55 @@ impl BackgroundDocumentRequestHandler for DocumentDiagnosticRequestHandler {
 
 fn compute_diagnostics(snapshot: &DocumentSnapshot, db: &RootDatabase) -> Vec<Diagnostic> {
     let Some(file) = snapshot.file(db) else {
+        tracing::info!(
+            "No file found for snapshot for '{}'",
+            snapshot.query().file_url()
+        );
         return vec![];
     };
-    let Ok(diagnostics) = db.check_file(file) else {
-        return vec![];
+
+    let diagnostics = match db.check_file(file) {
+        Ok(diagnostics) => diagnostics,
+        Err(cancelled) => {
+            tracing::info!("Diagnostics computation {cancelled}");
+            return vec![];
+        }
     };
 
     diagnostics
         .as_slice()
         .iter()
-        .map(|message| Diagnostic {
-            range: Range::default(),
-            severity: None,
-            tags: None,
-            code: None,
-            code_description: None,
-            source: Some("red-knot".into()),
-            message: message.to_string(),
-            related_information: None,
-            data: None,
-        })
+        .map(|message| to_lsp_diagnostic(message))
         .collect()
 }
+
+fn to_lsp_diagnostic(message: &str) -> Diagnostic {
+    let words = message.split(':').collect::<Vec<_>>();
+
+    let (range, message) = match words.as_slice() {
+        [_, _, line, column, message] | [_, line, column, message] => {
+            let line = line.parse::<u32>().unwrap_or_default().saturating_sub(1);
+            let column = column.parse::<u32>().unwrap_or_default();
+            (
+                Range::new(
+                    Position::new(line, column.saturating_sub(1)),
+                    Position::new(line, column),
+                ),
+                message.trim(),
+            )
+        }
+        _ => (Range::default(), message),
+    };
+
+    Diagnostic {
+        range,
+        severity: Some(DiagnosticSeverity::ERROR),
+        tags: None,
+        code: None,
+        code_description: None,
+        source: Some("red-knot".into()),
+        message: message.to_string(),
+        related_information: None,
+        data: None,
+    }
+}

crates/red_knot_server/src/server/api/traits.rs

@@ -34,7 +34,7 @@ pub(super) trait BackgroundDocumentRequestHandler: RequestHandler {
 
     fn run_with_snapshot(
         snapshot: DocumentSnapshot,
-        db: Option<RootDatabase>,
+        db: RootDatabase,
         notifier: Notifier,
         params: <<Self as RequestHandler>::RequestType as Request>::Params,
     ) -> super::Result<<<Self as RequestHandler>::RequestType as Request>::Result>;

crates/red_knot_server/src/session.rs

@@ -6,16 +6,16 @@ use std::path::{Path, PathBuf};
 use std::sync::Arc;
 
 use anyhow::anyhow;
-use lsp_types::{ClientCapabilities, Url};
+use lsp_types::{ClientCapabilities, TextDocumentContentChangeEvent, Url};
 
-use red_knot_python_semantic::{ProgramSettings, PythonVersion, SearchPathSettings};
 use red_knot_workspace::db::RootDatabase;
 use red_knot_workspace::workspace::WorkspaceMetadata;
 use ruff_db::files::{system_path_to_file, File};
 use ruff_db::system::SystemPath;
+use ruff_db::Db;
 
-use crate::edit::{DocumentKey, NotebookDocument};
-use crate::system::{url_to_system_path, LSPSystem};
+use crate::edit::{DocumentKey, DocumentVersion, NotebookDocument};
+use crate::system::{url_to_any_system_path, AnySystemPath, LSPSystem};
 use crate::{PositionEncoding, TextDocument};
 
 pub(crate) use self::capabilities::ResolvedClientCapabilities;
@@ -67,18 +67,10 @@ impl Session {
                 .ok_or_else(|| anyhow!("Workspace path is not a valid UTF-8 path: {:?}", path))?;
             let system = LSPSystem::new(index.clone());
 
-            let metadata = WorkspaceMetadata::from_path(system_path, &system)?;
             // TODO(dhruvmanila): Get the values from the client settings
-            let program_settings = ProgramSettings {
-                target_version: PythonVersion::default(),
-                search_paths: SearchPathSettings {
-                    extra_paths: vec![],
-                    src_root: system_path.to_path_buf(),
-                    site_packages: vec![],
-                    custom_typeshed: None,
-                },
-            };
-            workspaces.insert(path, RootDatabase::new(metadata, program_settings, system));
+            let metadata = WorkspaceMetadata::from_path(system_path, &system, None)?;
+            // TODO(micha): Handle the case where the program settings are incorrect more gracefully.
+            workspaces.insert(path, RootDatabase::new(metadata, system)?);
         }
 
         Ok(Self {
@@ -91,6 +83,12 @@ impl Session {
         })
     }
 
+    // TODO(dhruvmanila): Ideally, we should have a single method for `workspace_db_for_path_mut`
+    // and `default_workspace_db_mut` but the borrow checker doesn't allow that.
+    // https://github.com/astral-sh/ruff/pull/13041#discussion_r1726725437
+
+    /// Returns a reference to the workspace [`RootDatabase`] corresponding to the given path, if
+    /// any.
     pub(crate) fn workspace_db_for_path(&self, path: impl AsRef<Path>) -> Option<&RootDatabase> {
         self.workspaces
             .range(..=path.as_ref().to_path_buf())
@@ -98,6 +96,8 @@ impl Session {
             .map(|(_, db)| db)
     }
 
+    /// Returns a mutable reference to the workspace [`RootDatabase`] corresponding to the given
+    /// path, if any.
     pub(crate) fn workspace_db_for_path_mut(
         &mut self,
         path: impl AsRef<Path>,
@@ -108,6 +108,19 @@ impl Session {
             .map(|(_, db)| db)
     }
 
+    /// Returns a reference to the default workspace [`RootDatabase`]. The default workspace is the
+    /// minimum root path in the workspace map.
+    pub(crate) fn default_workspace_db(&self) -> &RootDatabase {
+        // SAFETY: Currently, red knot only support a single workspace.
+        self.workspaces.values().next().unwrap()
+    }
+
+    /// Returns a mutable reference to the default workspace [`RootDatabase`].
+    pub(crate) fn default_workspace_db_mut(&mut self) -> &mut RootDatabase {
+        // SAFETY: Currently, red knot only support a single workspace.
+        self.workspaces.values_mut().next().unwrap()
+    }
+
     pub fn key_from_url(&self, url: Url) -> DocumentKey {
         self.index().key_from_url(url)
     }
@@ -134,6 +147,20 @@ impl Session {
         self.index_mut().open_text_document(url, document);
     }
 
+    /// Updates a text document at the associated `key`.
+    ///
+    /// The document key must point to a text document, or this will throw an error.
+    pub(crate) fn update_text_document(
+        &mut self,
+        key: &DocumentKey,
+        content_changes: Vec<TextDocumentContentChangeEvent>,
+        new_version: DocumentVersion,
+    ) -> crate::Result<()> {
+        let position_encoding = self.position_encoding;
+        self.index_mut()
+            .update_text_document(key, content_changes, new_version, position_encoding)
+    }
+
     /// De-registers a document, specified by its key.
     /// Calling this multiple times for the same document is a logic error.
     pub(crate) fn close_document(&mut self, key: &DocumentKey) -> crate::Result<()> {
@@ -220,6 +247,7 @@ impl Drop for MutIndexGuard<'_> {
 
 /// An immutable snapshot of `Session` that references
 /// a specific document.
+#[derive(Debug)]
 pub struct DocumentSnapshot {
     resolved_client_capabilities: Arc<ResolvedClientCapabilities>,
     document_ref: index::DocumentQuery,
@@ -240,7 +268,12 @@ impl DocumentSnapshot {
     }
 
     pub(crate) fn file(&self, db: &RootDatabase) -> Option<File> {
-        let path = url_to_system_path(self.document_ref.file_url()).ok()?;
-        system_path_to_file(db, path).ok()
+        match url_to_any_system_path(self.document_ref.file_url()).ok()? {
+            AnySystemPath::System(path) => system_path_to_file(db, path).ok(),
+            AnySystemPath::SystemVirtual(virtual_path) => db
+                .files()
+                .try_virtual_file(&virtual_path)
+                .map(|virtual_file| virtual_file.file()),
+        }
     }
 }

crates/red_knot_server/src/session/capabilities.rs

@@ -1,5 +1,4 @@
 use lsp_types::ClientCapabilities;
-use ruff_linter::display_settings;
 
 #[derive(Debug, Clone, PartialEq, Eq, Default)]
 #[allow(clippy::struct_excessive_bools)]
@@ -66,20 +65,3 @@ impl ResolvedClientCapabilities {
         }
     }
 }
-
-impl std::fmt::Display for ResolvedClientCapabilities {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        display_settings! {
-            formatter = f,
-            namespace = "capabilities",
-            fields = [
-                self.code_action_deferred_edit_resolution,
-                self.apply_edit,
-                self.document_changes,
-                self.workspace_refresh,
-                self.pull_diagnostics,
-            ]
-        };
-        Ok(())
-    }
-}

crates/red_knot_server/src/session/index.rs

@@ -278,18 +278,6 @@ impl DocumentQuery {
         }
     }
 
-    /// Generate a source kind used by the linter.
-    pub(crate) fn make_source_kind(&self) -> ruff_linter::source_kind::SourceKind {
-        match self {
-            Self::Text { document, .. } => {
-                ruff_linter::source_kind::SourceKind::Python(document.contents().to_string())
-            }
-            Self::Notebook { notebook, .. } => {
-                ruff_linter::source_kind::SourceKind::IpyNotebook(notebook.make_ruff_notebook())
-            }
-        }
-    }
-
     /// Attempts to access the underlying notebook document that this query is selecting.
     pub fn as_notebook(&self) -> Option<&NotebookDocument> {
         match self {

crates/red_knot_server/src/system.rs

@@ -8,27 +8,40 @@ use ruff_db::file_revision::FileRevision;
 use ruff_db::system::walk_directory::WalkDirectoryBuilder;
 use ruff_db::system::{
     DirectoryEntry, FileType, Metadata, OsSystem, Result, System, SystemPath, SystemPathBuf,
-    SystemVirtualPath,
+    SystemVirtualPath, SystemVirtualPathBuf,
 };
 use ruff_notebook::{Notebook, NotebookError};
 
 use crate::session::index::Index;
 use crate::DocumentQuery;
 
-/// Converts the given [`Url`] to a [`SystemPathBuf`].
+/// Converts the given [`Url`] to an [`AnySystemPath`].
+///
+/// If the URL scheme is `file`, then the path is converted to a [`SystemPathBuf`]. Otherwise, the
+/// URL is converted to a [`SystemVirtualPathBuf`].
 ///
 /// This fails in the following cases:
-/// * The URL scheme is not `file`.
 /// * The URL cannot be converted to a file path (refer to [`Url::to_file_path`]).
 /// * If the URL is not a valid UTF-8 string.
-pub(crate) fn url_to_system_path(url: &Url) -> std::result::Result<SystemPathBuf, ()> {
+pub(crate) fn url_to_any_system_path(url: &Url) -> std::result::Result<AnySystemPath, ()> {
     if url.scheme() == "file" {
-        Ok(SystemPathBuf::from_path_buf(url.to_file_path()?).map_err(|_| ())?)
+        Ok(AnySystemPath::System(
+            SystemPathBuf::from_path_buf(url.to_file_path()?).map_err(|_| ())?,
+        ))
     } else {
-        Err(())
+        Ok(AnySystemPath::SystemVirtual(
+            SystemVirtualPath::new(url.as_str()).to_path_buf(),
+        ))
     }
 }
 
+/// Represents either a [`SystemPath`] or a [`SystemVirtualPath`].
+#[derive(Debug)]
+pub(crate) enum AnySystemPath {
+    System(SystemPathBuf),
+    SystemVirtual(SystemVirtualPathBuf),
+}
+
 #[derive(Debug)]
 pub(crate) struct LSPSystem {
     /// A read-only copy of the index where the server stores all the open documents and settings.
@@ -144,19 +157,6 @@ impl System for LSPSystem {
         }
     }
 
-    fn virtual_path_metadata(&self, path: &SystemVirtualPath) -> Result<Metadata> {
-        // Virtual paths only exists in the LSP system, so we don't need to check the OS system.
-        let document = self
-            .system_virtual_path_to_document_ref(path)?
-            .ok_or_else(|| virtual_path_not_found(path))?;
-
-        Ok(Metadata::new(
-            document_revision(&document),
-            None,
-            FileType::File,
-        ))
-    }
-
     fn read_virtual_path_to_string(&self, path: &SystemVirtualPath) -> Result<String> {
         let document = self
             .system_virtual_path_to_document_ref(path)?

crates/red_knot_wasm/Cargo.toml

@@ -20,9 +20,9 @@ default = ["console_error_panic_hook"]
 
 [dependencies]
 red_knot_python_semantic = { workspace = true }
-red_knot_workspace = { workspace = true }
+red_knot_workspace = { workspace = true, default-features = false, features = ["deflate"] }
 
-ruff_db = { workspace = true }
+ruff_db = { workspace = true, features = [] }
 ruff_notebook = { workspace = true }
 
 console_error_panic_hook = { workspace = true, optional = true }

crates/red_knot_wasm/src/lib.rs

@@ -3,8 +3,8 @@ use std::any::Any;
 use js_sys::Error;
 use wasm_bindgen::prelude::*;
 
-use red_knot_python_semantic::{ProgramSettings, SearchPathSettings};
 use red_knot_workspace::db::RootDatabase;
+use red_knot_workspace::workspace::settings::Configuration;
 use red_knot_workspace::workspace::WorkspaceMetadata;
 use ruff_db::files::{system_path_to_file, File};
 use ruff_db::system::walk_directory::WalkDirectoryBuilder;
@@ -41,15 +41,17 @@ impl Workspace {
     #[wasm_bindgen(constructor)]
     pub fn new(root: &str, settings: &Settings) -> Result<Workspace, Error> {
         let system = WasmSystem::new(SystemPath::new(root));
-        let workspace =
-            WorkspaceMetadata::from_path(SystemPath::new(root), &system).map_err(into_error)?;
+        let workspace = WorkspaceMetadata::from_path(
+            SystemPath::new(root),
+            &system,
+            Some(Configuration {
+                target_version: Some(settings.target_version.into()),
+                ..Configuration::default()
+            }),
+        )
+        .map_err(into_error)?;
 
-        let program_settings = ProgramSettings {
-            target_version: settings.target_version.into(),
-            search_paths: SearchPathSettings::default(),
-        };
-
-        let db = RootDatabase::new(workspace, program_settings, system.clone());
+        let db = RootDatabase::new(workspace, system.clone()).map_err(into_error)?;
 
         Ok(Self { db, system })
     }
@@ -108,7 +110,7 @@ impl Workspace {
     pub fn check_file(&self, file_id: &FileHandle) -> Result<Vec<String>, Error> {
         let result = self.db.check_file(file_id.file).map_err(into_error)?;
 
-        Ok(result.to_vec())
+        Ok(result.clone())
     }
 
     /// Checks all open files
@@ -232,13 +234,6 @@ impl System for WasmSystem {
         Notebook::from_source_code(&content)
     }
 
-    fn virtual_path_metadata(
-        &self,
-        _path: &SystemVirtualPath,
-    ) -> ruff_db::system::Result<Metadata> {
-        Err(not_found())
-    }
-
     fn read_virtual_path_to_string(
         &self,
         _path: &SystemVirtualPath,

crates/red_knot_wasm/tests/api.rs

@@ -11,11 +11,14 @@ fn check() {
     };
     let mut workspace = Workspace::new("/", &settings).expect("Workspace to be created");
 
-    let test = workspace
+    workspace
         .open_file("test.py", "import random22\n")
         .expect("File to be opened");
 
-    let result = workspace.check_file(&test).expect("Check to succeed");
+    let result = workspace.check().expect("Check to succeed");
 
-    assert_eq!(result, vec!["Unresolved import 'random22'"]);
+    assert_eq!(
+        result,
+        vec!["/test.py:1:8: Cannot resolve import 'random22'."]
+    );
 }

crates/red_knot_workspace/Cargo.toml

@@ -17,16 +17,25 @@ red_knot_python_semantic = { workspace = true }
 ruff_cache = { workspace = true }
 ruff_db = { workspace = true, features = ["os", "cache"] }
 ruff_python_ast = { workspace = true }
+ruff_text_size = { workspace = true }
+ruff_vendored = { workspace = true }
 
 anyhow = { workspace = true }
 crossbeam = { workspace = true }
 notify = { workspace = true }
+rayon = { workspace = true }
 rustc-hash = { workspace = true }
 salsa = { workspace = true }
-thiserror = { workspace = true }
 tracing = { workspace = true }
 
 [dev-dependencies]
+ruff_db = { workspace = true, features = ["testing"] }
+tempfile = { workspace = true }
+
+[features]
+default = ["zstd"]
+zstd = ["ruff_vendored/zstd"]
+deflate = ["ruff_vendored/deflate"]
 
 [lints]
 workspace = true

crates/red_knot_workspace/resources/test/corpus/04_assign_named_expr.py

@@ -0,0 +1,2 @@
+x = 0
+(x := x + 1)

crates/red_knot_workspace/resources/test/corpus/10_if_with_named_expr.py

@@ -0,0 +1,3 @@
+x = 0
+if x := x + 1:
+    pass

crates/red_knot_workspace/resources/test/corpus/25_func_annotations_same_name.py

@@ -0,0 +1,11 @@
+def bool(x) -> bool:
+    return True
+
+
+class MyClass: ...
+
+
+def MyClass() -> MyClass: ...
+
+
+def x(self) -> x: ...

crates/red_knot_workspace/resources/test/corpus/26_func_defaults_same_name.py

@@ -0,0 +1,2 @@
+def bool(x=bool):
+    return x

crates/red_knot_workspace/resources/test/corpus/67_with_non_name_target.py

@@ -0,0 +1,2 @@
+with foo() as self.bar:
+    pass

crates/red_knot_workspace/resources/test/corpus/85_match_guard_with_named_expr.py

@@ -0,0 +1,3 @@
+match x:
+    case [1, 0] if x := x[:0]:
+        y = 1

crates/red_knot_workspace/src/db.rs

@@ -1,17 +1,15 @@
 use std::panic::RefUnwindSafe;
 use std::sync::Arc;
 
-use red_knot_python_semantic::{
-    vendored_typeshed_stubs, Db as SemanticDb, Program, ProgramSettings,
-};
+use salsa::plumbing::ZalsaDatabase;
+use salsa::{Cancelled, Event};
+
+use red_knot_python_semantic::{Db as SemanticDb, Program};
 use ruff_db::files::{File, Files};
 use ruff_db::system::System;
 use ruff_db::vendored::VendoredFileSystem;
 use ruff_db::{Db as SourceDb, Upcast};
-use salsa::plumbing::ZalsaDatabase;
-use salsa::{Cancelled, Event};
 
-use crate::lint::Diagnostics;
 use crate::workspace::{check_file, Workspace, WorkspaceMetadata};
 
 mod changes;
@@ -28,7 +26,7 @@ pub struct RootDatabase {
 }
 
 impl RootDatabase {
-    pub fn new<S>(workspace: WorkspaceMetadata, settings: ProgramSettings, system: S) -> Self
+    pub fn new<S>(workspace: WorkspaceMetadata, system: S) -> anyhow::Result<Self>
     where
         S: System + 'static + Send + Sync + RefUnwindSafe,
     {
@@ -39,12 +37,12 @@ impl RootDatabase {
             system: Arc::new(system),
         };
 
-        let workspace = Workspace::from_metadata(&db, workspace);
         // Initialize the `Program` singleton
-        Program::from_settings(&db, settings);
+        Program::from_settings(&db, workspace.settings().program())?;
 
-        db.workspace = Some(workspace);
-        db
+        db.workspace = Some(Workspace::from_metadata(&db, workspace));
+
+        Ok(db)
     }
 
     pub fn workspace(&self) -> Workspace {
@@ -57,7 +55,9 @@ impl RootDatabase {
         self.with_db(|db| db.workspace().check(db))
     }
 
-    pub fn check_file(&self, file: File) -> Result<Diagnostics, Cancelled> {
+    pub fn check_file(&self, file: File) -> Result<Vec<String>, Cancelled> {
+        let _span = tracing::debug_span!("check_file", file=%file.path(self)).entered();
+
         self.with_db(|db| check_file(db, file))
     }
 
@@ -111,12 +111,20 @@ impl Upcast<dyn SourceDb> for RootDatabase {
 }
 
 #[salsa::db]
-impl SemanticDb for RootDatabase {}
+impl SemanticDb for RootDatabase {
+    fn is_file_open(&self, file: File) -> bool {
+        let Some(workspace) = &self.workspace else {
+            return false;
+        };
+
+        workspace.is_file_open(self, file)
+    }
+}
 
 #[salsa::db]
 impl SourceDb for RootDatabase {
     fn vendored(&self) -> &VendoredFileSystem {
-        vendored_typeshed_stubs()
+        ruff_vendored::file_system()
     }
 
     fn system(&self) -> &dyn System {
@@ -149,9 +157,11 @@ impl Db for RootDatabase {}
 
 #[cfg(test)]
 pub(crate) mod tests {
+    use std::sync::Arc;
+
     use salsa::Event;
 
-    use red_knot_python_semantic::{vendored_typeshed_stubs, Db as SemanticDb};
+    use red_knot_python_semantic::Db as SemanticDb;
     use ruff_db::files::Files;
     use ruff_db::system::{DbWithTestSystem, System, TestSystem};
     use ruff_db::vendored::VendoredFileSystem;
@@ -162,6 +172,7 @@ pub(crate) mod tests {
     #[salsa::db]
     pub(crate) struct TestDb {
         storage: salsa::Storage<Self>,
+        events: std::sync::Arc<std::sync::Mutex<Vec<salsa::Event>>>,
         files: Files,
         system: TestSystem,
         vendored: VendoredFileSystem,
@@ -172,12 +183,26 @@ pub(crate) mod tests {
             Self {
                 storage: salsa::Storage::default(),
                 system: TestSystem::default(),
-                vendored: vendored_typeshed_stubs().clone(),
+                vendored: ruff_vendored::file_system().clone(),
                 files: Files::default(),
+                events: Arc::default(),
             }
         }
     }
 
+    impl TestDb {
+        /// Takes the salsa events.
+        ///
+        /// ## Panics
+        /// If there are any pending salsa snapshots.
+        pub(crate) fn take_salsa_events(&mut self) -> Vec<salsa::Event> {
+            let inner = Arc::get_mut(&mut self.events).expect("no pending salsa snapshots");
+
+            let events = inner.get_mut().unwrap();
+            std::mem::take(&mut *events)
+        }
+    }
+
     impl DbWithTestSystem for TestDb {
         fn test_system(&self) -> &TestSystem {
             &self.system
@@ -222,12 +247,20 @@ pub(crate) mod tests {
     }
 
     #[salsa::db]
-    impl red_knot_python_semantic::Db for TestDb {}
+    impl red_knot_python_semantic::Db for TestDb {
+        fn is_file_open(&self, file: ruff_db::files::File) -> bool {
+            !file.path(self).is_vendored_path()
+        }
+    }
+
     #[salsa::db]
     impl Db for TestDb {}
 
     #[salsa::db]
     impl salsa::Database for TestDb {
-        fn salsa_event(&self, _event: &dyn Fn() -> Event) {}
+        fn salsa_event(&self, event: &dyn Fn() -> Event) {
+            let mut events = self.events.lock().unwrap();
+            events.push(event());
+        }
     }
 }

crates/red_knot_workspace/src/db/changes.rs

@@ -1,22 +1,33 @@
-use rustc_hash::FxHashSet;
-
+use red_knot_python_semantic::Program;
 use ruff_db::files::{system_path_to_file, File, Files};
 use ruff_db::system::walk_directory::WalkState;
 use ruff_db::system::SystemPath;
 use ruff_db::Db;
+use rustc_hash::FxHashSet;
 
 use crate::db::RootDatabase;
 use crate::watch;
 use crate::watch::{CreatedKind, DeletedKind};
+use crate::workspace::settings::Configuration;
 use crate::workspace::WorkspaceMetadata;
 
 impl RootDatabase {
-    #[tracing::instrument(level = "debug", skip(self, changes))]
-    pub fn apply_changes(&mut self, changes: Vec<watch::ChangeEvent>) {
+    #[tracing::instrument(level = "debug", skip(self, changes, base_configuration))]
+    pub fn apply_changes(
+        &mut self,
+        changes: Vec<watch::ChangeEvent>,
+        base_configuration: Option<&Configuration>,
+    ) {
         let workspace = self.workspace();
         let workspace_path = workspace.root(self).to_path_buf();
+        let program = Program::get(self);
+        let custom_stdlib_versions_path = program
+            .custom_stdlib_search_path(self)
+            .map(|path| path.join("VERSIONS"));
 
         let mut workspace_change = false;
+        // Changes to a custom stdlib path's VERSIONS
+        let mut custom_stdlib_change = false;
         // Packages that need reloading
         let mut changed_packages = FxHashSet::default();
         // Paths that were added
@@ -39,7 +50,7 @@ impl RootDatabase {
         };
 
         for change in changes {
-            if let Some(path) = change.path() {
+            if let Some(path) = change.system_path() {
                 if matches!(
                     path.file_name(),
                     Some(".gitignore" | ".ignore" | "ruff.toml" | ".ruff.toml" | "pyproject.toml")
@@ -54,10 +65,15 @@ impl RootDatabase {
 
                     continue;
                 }
+
+                if Some(path) == custom_stdlib_versions_path.as_deref() {
+                    custom_stdlib_change = true;
+                }
             }
 
             match change {
-                watch::ChangeEvent::Changed { path, kind: _ } => sync_path(self, &path),
+                watch::ChangeEvent::Changed { path, kind: _ }
+                | watch::ChangeEvent::Opened(path) => sync_path(self, &path),
 
                 watch::ChangeEvent::Created { kind, path } => {
                     match kind {
@@ -100,7 +116,13 @@ impl RootDatabase {
                     } else {
                         sync_recursively(self, &path);
 
-                        // TODO: Remove after converting `package.files()` to a salsa query.
+                        if custom_stdlib_versions_path
+                            .as_ref()
+                            .is_some_and(|versions_path| versions_path.starts_with(&path))
+                        {
+                            custom_stdlib_change = true;
+                        }
+
                         if let Some(package) = workspace.package(self, &path) {
                             changed_packages.insert(package);
                         } else {
@@ -109,6 +131,17 @@ impl RootDatabase {
                     }
                 }
 
+                watch::ChangeEvent::CreatedVirtual(path)
+                | watch::ChangeEvent::ChangedVirtual(path) => {
+                    File::sync_virtual_path(self, &path);
+                }
+
+                watch::ChangeEvent::DeletedVirtual(path) => {
+                    if let Some(virtual_file) = self.files().try_virtual_file(&path) {
+                        virtual_file.close(self);
+                    }
+                }
+
                 watch::ChangeEvent::Rescan => {
                     workspace_change = true;
                     Files::sync_all(self);
@@ -118,9 +151,13 @@ impl RootDatabase {
         }
 
         if workspace_change {
-            match WorkspaceMetadata::from_path(&workspace_path, self.system()) {
+            match WorkspaceMetadata::from_path(
+                &workspace_path,
+                self.system(),
+                base_configuration.cloned(),
+            ) {
                 Ok(metadata) => {
-                    tracing::debug!("Reload workspace after structural change.");
+                    tracing::debug!("Reloading workspace after structural change.");
                     // TODO: Handle changes in the program settings.
                     workspace.reload(self, metadata);
                 }
@@ -130,6 +167,11 @@ impl RootDatabase {
             }
 
             return;
+        } else if custom_stdlib_change {
+            let search_paths = workspace.search_path_settings(self).clone();
+            if let Err(error) = program.update_search_paths(self, &search_paths) {
+                tracing::error!("Failed to set the new search paths: {error}");
+            }
         }
 
         let mut added_paths = added_paths.into_iter().filter(|path| {

crates/red_knot_workspace/src/lib.rs

@@ -1,5 +1,4 @@
 pub mod db;
 pub mod lint;
-pub mod site_packages;
 pub mod watch;
 pub mod workspace;

crates/red_knot_workspace/src/lint.rs

@@ -1,5 +1,4 @@
 use std::cell::RefCell;
-use std::ops::Deref;
 use std::time::Duration;
 
 use tracing::debug_span;
@@ -11,6 +10,7 @@ use ruff_db::parsed::{parsed_module, ParsedModule};
 use ruff_db::source::{source_text, SourceText};
 use ruff_python_ast as ast;
 use ruff_python_ast::visitor::{walk_expr, walk_stmt, Visitor};
+use ruff_text_size::{Ranged, TextSize};
 
 use crate::db::Db;
 
@@ -21,7 +21,7 @@ use crate::db::Db;
 pub(crate) fn unwind_if_cancelled(db: &dyn Db) {}
 
 #[salsa::tracked(return_ref)]
-pub(crate) fn lint_syntax(db: &dyn Db, file_id: File) -> Diagnostics {
+pub(crate) fn lint_syntax(db: &dyn Db, file_id: File) -> Vec<String> {
     #[allow(clippy::print_stdout)]
     if std::env::var("RED_KNOT_SLOW_LINT").is_ok() {
         for i in 0..10 {
@@ -48,11 +48,9 @@ pub(crate) fn lint_syntax(db: &dyn Db, file_id: File) -> Diagnostics {
         };
         visitor.visit_body(&ast.body);
         diagnostics = visitor.diagnostics;
-    } else {
-        diagnostics.extend(parsed.errors().iter().map(ToString::to_string));
     }
 
-    Diagnostics::from(diagnostics)
+    diagnostics
 }
 
 fn lint_lines(source: &str, diagnostics: &mut Vec<String>) {
@@ -74,7 +72,7 @@ fn lint_lines(source: &str, diagnostics: &mut Vec<String>) {
 
 #[allow(unreachable_pub)]
 #[salsa::tracked(return_ref)]
-pub fn lint_semantic(db: &dyn Db, file_id: File) -> Diagnostics {
+pub fn lint_semantic(db: &dyn Db, file_id: File) -> Vec<String> {
     let _span = debug_span!("lint_semantic", file=%file_id.path(db)).entered();
 
     let source = source_text(db.upcast(), file_id);
@@ -82,7 +80,7 @@ pub fn lint_semantic(db: &dyn Db, file_id: File) -> Diagnostics {
     let semantic = SemanticModel::new(db.upcast(), file_id);
 
     if !parsed.is_valid() {
-        return Diagnostics::Empty;
+        return vec![];
     }
 
     let context = SemanticLintContext {
@@ -94,30 +92,21 @@ pub fn lint_semantic(db: &dyn Db, file_id: File) -> Diagnostics {
 
     SemanticVisitor { context: &context }.visit_body(parsed.suite());
 
-    Diagnostics::from(context.diagnostics.take())
+    context.diagnostics.take()
 }
 
-fn lint_unresolved_imports(context: &SemanticLintContext, import: AnyImportRef) {
-    match import {
-        AnyImportRef::Import(import) => {
-            for alias in &import.names {
-                let ty = alias.ty(&context.semantic);
-
-                if ty.is_unbound() {
-                    context.push_diagnostic(format!("Unresolved import '{}'", &alias.name));
-                }
-            }
-        }
-        AnyImportRef::ImportFrom(import) => {
-            for alias in &import.names {
-                let ty = alias.ty(&context.semantic);
-
-                if ty.is_unbound() {
-                    context.push_diagnostic(format!("Unresolved import '{}'", &alias.name));
-                }
-            }
-        }
-    }
+fn format_diagnostic(context: &SemanticLintContext, message: &str, start: TextSize) -> String {
+    let source_location = context
+        .semantic
+        .line_index()
+        .source_location(start, context.source_text());
+    format!(
+        "{}:{}:{}: {}",
+        context.semantic.file_path(),
+        source_location.row,
+        source_location.column,
+        message,
+    )
 }
 
 fn lint_maybe_undefined(context: &SemanticLintContext, name: &ast::ExprName) {
@@ -125,17 +114,19 @@ fn lint_maybe_undefined(context: &SemanticLintContext, name: &ast::ExprName) {
         return;
     }
     let semantic = &context.semantic;
-    match name.ty(semantic) {
-        Type::Unbound => {
-            context.push_diagnostic(format!("Name '{}' used when not defined.", &name.id));
-        }
-        Type::Union(union) if union.contains(semantic.db(), Type::Unbound) => {
-            context.push_diagnostic(format!(
-                "Name '{}' used when possibly not defined.",
-                &name.id
-            ));
-        }
-        _ => {}
+    let ty = name.ty(semantic);
+    if ty.is_unbound() {
+        context.push_diagnostic(format_diagnostic(
+            context,
+            &format!("Name '{}' used when not defined.", &name.id),
+            name.start(),
+        ));
+    } else if ty.may_be_unbound(semantic.db()) {
+        context.push_diagnostic(format_diagnostic(
+            context,
+            &format!("Name '{}' used when possibly not defined.", &name.id),
+            name.start(),
+        ));
     }
 }
 
@@ -170,7 +161,7 @@ fn lint_bad_override(context: &SemanticLintContext, class: &ast::StmtClassDef) {
         if ty.has_decorator(db, override_ty) {
             let method_name = ty.name(db);
             if class_ty
-                .inherited_class_member(db, &method_name)
+                .inherited_class_member(db, method_name)
                 .is_unbound()
             {
                 // TODO should have a qualname() method to support nested classes
@@ -237,17 +228,8 @@ struct SemanticVisitor<'a> {
 
 impl Visitor<'_> for SemanticVisitor<'_> {
     fn visit_stmt(&mut self, stmt: &ast::Stmt) {
-        match stmt {
-            ast::Stmt::ClassDef(class) => {
-                lint_bad_override(self.context, class);
-            }
-            ast::Stmt::Import(import) => {
-                lint_unresolved_imports(self.context, AnyImportRef::Import(import));
-            }
-            ast::Stmt::ImportFrom(import) => {
-                lint_unresolved_imports(self.context, AnyImportRef::ImportFrom(import));
-            }
-            _ => {}
+        if let ast::Stmt::ClassDef(class) = stmt {
+            lint_bad_override(self.context, class);
         }
 
         walk_stmt(self, stmt);
@@ -265,53 +247,15 @@ impl Visitor<'_> for SemanticVisitor<'_> {
     }
 }
 
-#[derive(Debug, Clone, PartialEq, Eq)]
-pub enum Diagnostics {
-    Empty,
-    List(Vec<String>),
-}
-
-impl Diagnostics {
-    pub fn as_slice(&self) -> &[String] {
-        match self {
-            Diagnostics::Empty => &[],
-            Diagnostics::List(list) => list.as_slice(),
-        }
-    }
-}
-
-impl Deref for Diagnostics {
-    type Target = [String];
-    fn deref(&self) -> &Self::Target {
-        self.as_slice()
-    }
-}
-
-impl From<Vec<String>> for Diagnostics {
-    fn from(value: Vec<String>) -> Self {
-        if value.is_empty() {
-            Diagnostics::Empty
-        } else {
-            Diagnostics::List(value)
-        }
-    }
-}
-
-#[derive(Copy, Clone, Debug)]
-enum AnyImportRef<'a> {
-    Import(&'a ast::StmtImport),
-    ImportFrom(&'a ast::StmtImportFrom),
-}
-
 #[cfg(test)]
 mod tests {
-    use red_knot_python_semantic::{Program, PythonVersion, SearchPathSettings};
+    use red_knot_python_semantic::{Program, ProgramSettings, PythonVersion, SearchPathSettings};
     use ruff_db::files::system_path_to_file;
     use ruff_db::system::{DbWithTestSystem, SystemPathBuf};
 
     use crate::db::tests::TestDb;
 
-    use super::{lint_semantic, Diagnostics};
+    use super::lint_semantic;
 
     fn setup_db() -> TestDb {
         setup_db_with_root(SystemPathBuf::from("/src"))
@@ -320,16 +264,18 @@ mod tests {
     fn setup_db_with_root(src_root: SystemPathBuf) -> TestDb {
         let db = TestDb::new();
 
-        Program::new(
+        db.memory_file_system()
+            .create_directory_all(&src_root)
+            .unwrap();
+
+        Program::from_settings(
             &db,
-            PythonVersion::default(),
-            SearchPathSettings {
-                extra_paths: Vec::new(),
-                src_root,
-                site_packages: vec![],
-                custom_typeshed: None,
+            &ProgramSettings {
+                target_version: PythonVersion::default(),
+                search_paths: SearchPathSettings::new(src_root),
             },
-        );
+        )
+        .expect("Valid program settings");
 
         db
     }
@@ -350,16 +296,23 @@ mod tests {
         .unwrap();
 
         let file = system_path_to_file(&db, "/src/a.py").expect("file to exist");
-        let Diagnostics::List(messages) = lint_semantic(&db, file) else {
-            panic!("expected some diagnostics");
-        };
+        let messages = lint_semantic(&db, file);
+
+        assert_ne!(messages, &[] as &[String], "expected some diagnostics");
 
         assert_eq!(
             *messages,
-            vec![
-                "Name 'flag' used when not defined.",
-                "Name 'y' used when possibly not defined."
-            ]
+            if cfg!(windows) {
+                vec![
+                    "\\src\\a.py:3:4: Name 'flag' used when not defined.",
+                    "\\src\\a.py:5:1: Name 'y' used when possibly not defined.",
+                ]
+            } else {
+                vec![
+                    "/src/a.py:3:4: Name 'flag' used when not defined.",
+                    "/src/a.py:5:1: Name 'y' used when possibly not defined.",
+                ]
+            }
         );
     }
 }

crates/red_knot_workspace/src/watch.rs

@@ -1,4 +1,4 @@
-use ruff_db::system::{SystemPath, SystemPathBuf};
+use ruff_db::system::{SystemPath, SystemPathBuf, SystemVirtualPathBuf};
 pub use watcher::{directory_watcher, EventHandler, Watcher};
 pub use workspace_watcher::WorkspaceWatcher;
 
@@ -20,6 +20,9 @@ mod workspace_watcher;
 /// event instead of emitting an event for each file or subdirectory in that path.
 #[derive(Debug, PartialEq, Eq)]
 pub enum ChangeEvent {
+    /// The file corresponding to the given path was opened in an editor.
+    Opened(SystemPathBuf),
+
     /// A new path was created
     Created {
         path: SystemPathBuf,
@@ -38,6 +41,15 @@ pub enum ChangeEvent {
         kind: DeletedKind,
     },
 
+    /// A new virtual path was created.
+    CreatedVirtual(SystemVirtualPathBuf),
+
+    /// The content of a virtual path was changed.
+    ChangedVirtual(SystemVirtualPathBuf),
+
+    /// A virtual path was deleted.
+    DeletedVirtual(SystemVirtualPathBuf),
+
     /// The file watcher failed to observe some changes and now is out of sync with the file system.
     ///
     /// This can happen if many files are changed at once. The consumer should rescan all files to catch up
@@ -46,16 +58,27 @@ pub enum ChangeEvent {
 }
 
 impl ChangeEvent {
-    pub fn file_name(&self) -> Option<&str> {
-        self.path().and_then(|path| path.file_name())
+    /// Creates a new [`Changed`] event for the file content at the given path.
+    ///
+    /// [`Changed`]: ChangeEvent::Changed
+    pub fn file_content_changed(path: SystemPathBuf) -> ChangeEvent {
+        ChangeEvent::Changed {
+            path,
+            kind: ChangedKind::FileContent,
+        }
     }
 
-    pub fn path(&self) -> Option<&SystemPath> {
+    pub fn file_name(&self) -> Option<&str> {
+        self.system_path().and_then(|path| path.file_name())
+    }
+
+    pub fn system_path(&self) -> Option<&SystemPath> {
         match self {
-            ChangeEvent::Created { path, .. }
+            ChangeEvent::Opened(path)
+            | ChangeEvent::Created { path, .. }
             | ChangeEvent::Changed { path, .. }
             | ChangeEvent::Deleted { path, .. } => Some(path),
-            ChangeEvent::Rescan => None,
+            _ => None,
         }
     }
 }

crates/red_knot_workspace/src/watch/watcher.rs

@@ -109,7 +109,7 @@ struct WatcherInner {
 impl Watcher {
     /// Sets up file watching for `path`.
     pub fn watch(&mut self, path: &SystemPath) -> notify::Result<()> {
-        tracing::debug!("Watching path: {path}.");
+        tracing::debug!("Watching path: '{path}'.");
 
         self.inner_mut()
             .watcher
@@ -118,7 +118,7 @@ impl Watcher {
 
     /// Stops file watching for `path`.
     pub fn unwatch(&mut self, path: &SystemPath) -> notify::Result<()> {
-        tracing::debug!("Unwatching path: {path}.");
+        tracing::debug!("Unwatching path: '{path}'.");
 
         self.inner_mut().watcher.unwatch(path.as_std_path())
     }
@@ -351,7 +351,7 @@ impl Debouncer {
             }
 
             EventKind::Any => {
-                tracing::debug!("Skip any FS event for {path}.");
+                tracing::debug!("Skipping any FS event for '{path}'.");
                 return;
             }
         };

crates/red_knot_workspace/src/workspace.rs

@@ -1,23 +1,30 @@
-use salsa::{Durability, Setter as _};
 use std::{collections::BTreeMap, sync::Arc};
 
 use rustc_hash::{FxBuildHasher, FxHashSet};
+use salsa::{Durability, Setter as _};
 
 pub use metadata::{PackageMetadata, WorkspaceMetadata};
+use red_knot_python_semantic::types::check_types;
+use red_knot_python_semantic::SearchPathSettings;
+use ruff_db::parsed::parsed_module;
+use ruff_db::source::{line_index, source_text, SourceDiagnostic};
 use ruff_db::{
     files::{system_path_to_file, File},
     system::{walk_directory::WalkState, SystemPath, SystemPathBuf},
 };
 use ruff_python_ast::{name::Name, PySourceType};
+use ruff_text_size::Ranged;
 
-use crate::workspace::files::{Index, IndexedFiles, PackageFiles};
+use crate::db::RootDatabase;
+use crate::workspace::files::{Index, Indexed, IndexedIter, PackageFiles};
 use crate::{
     db::Db,
-    lint::{lint_semantic, lint_syntax, Diagnostics},
+    lint::{lint_semantic, lint_syntax},
 };
 
 mod files;
 mod metadata;
+pub mod settings;
 
 /// The project workspace as a Salsa ingredient.
 ///
@@ -78,6 +85,10 @@ pub struct Workspace {
     /// The (first-party) packages in this workspace.
     #[return_ref]
     package_tree: BTreeMap<SystemPathBuf, Package>,
+
+    /// The unresolved search path configuration.
+    #[return_ref]
+    pub search_path_settings: SearchPathSettings,
 }
 
 /// A first-party package in a workspace.
@@ -91,8 +102,8 @@ pub struct Package {
     root_buf: SystemPathBuf,
 
     /// The files that are part of this package.
-    #[return_ref]
     #[default]
+    #[return_ref]
     file_set: PackageFiles,
     // TODO: Add the loaded settings.
 }
@@ -106,10 +117,14 @@ impl Workspace {
             packages.insert(package.root.clone(), Package::from_metadata(db, package));
         }
 
-        Workspace::builder(metadata.root, packages)
-            .durability(Durability::MEDIUM)
-            .open_fileset_durability(Durability::LOW)
-            .new(db)
+        Workspace::builder(
+            metadata.root,
+            packages,
+            metadata.settings.program.search_paths,
+        )
+        .durability(Durability::MEDIUM)
+        .open_fileset_durability(Durability::LOW)
+        .new(db)
     }
 
     pub fn root(self, db: &dyn Db) -> &SystemPath {
@@ -140,9 +155,12 @@ impl Workspace {
             new_packages.insert(path, package);
         }
 
-        self.set_package_tree(db)
-            .with_durability(Durability::MEDIUM)
-            .to(new_packages);
+        if &metadata.settings.program.search_paths != self.search_path_settings(db) {
+            self.set_search_path_settings(db)
+                .to(metadata.settings.program.search_paths);
+        }
+
+        self.set_package_tree(db).to(new_packages);
     }
 
     pub fn update_package(self, db: &mut dyn Db, metadata: PackageMetadata) -> anyhow::Result<()> {
@@ -173,30 +191,42 @@ impl Workspace {
     }
 
     /// Checks all open files in the workspace and its dependencies.
-    #[tracing::instrument(level = "debug", skip_all)]
-    pub fn check(self, db: &dyn Db) -> Vec<String> {
+    pub fn check(self, db: &RootDatabase) -> Vec<String> {
+        let workspace_span = tracing::debug_span!("check_workspace");
+        let _span = workspace_span.enter();
+
         tracing::debug!("Checking workspace");
+        let files = WorkspaceFiles::new(db, self);
+        let result = Arc::new(std::sync::Mutex::new(Vec::new()));
+        let inner_result = Arc::clone(&result);
 
-        let mut result = Vec::new();
+        let db = db.snapshot();
+        let workspace_span = workspace_span.clone();
 
-        if let Some(open_files) = self.open_files(db) {
-            for file in open_files {
-                result.extend_from_slice(&check_file(db, *file));
+        rayon::scope(move |scope| {
+            for file in &files {
+                let result = inner_result.clone();
+                let db = db.snapshot();
+                let workspace_span = workspace_span.clone();
+
+                scope.spawn(move |_| {
+                    let check_file_span = tracing::debug_span!(parent: &workspace_span, "check_file", file=%file.path(&db));
+                    let _entered = check_file_span.entered();
+
+                    let file_diagnostics = check_file(&db, file);
+                    result.lock().unwrap().extend(file_diagnostics);
+                });
             }
-        } else {
-            for package in self.packages(db) {
-                result.extend(package.check(db));
-            }
-        }
+        });
 
-        result
+        Arc::into_inner(result).unwrap().into_inner().unwrap()
     }
 
     /// Opens a file in the workspace.
     ///
     /// This changes the behavior of `check` to only check the open files rather than all files in the workspace.
     pub fn open_file(self, db: &mut dyn Db, file: File) {
-        tracing::debug!("Opening file {}", file.path(db));
+        tracing::debug!("Opening file '{}'", file.path(db));
 
         let mut open_files = self.take_open_files(db);
         open_files.insert(file);
@@ -205,7 +235,7 @@ impl Workspace {
 
     /// Closes a file in the workspace.
     pub fn close_file(self, db: &mut dyn Db, file: File) -> bool {
-        tracing::debug!("Closing file {}", file.path(db));
+        tracing::debug!("Closing file '{}'", file.path(db));
 
         let mut open_files = self.take_open_files(db);
         let removed = open_files.remove(&file);
@@ -248,9 +278,25 @@ impl Workspace {
             FxHashSet::default()
         }
     }
+
+    /// Returns `true` if the file is open in the workspace.
+    ///
+    /// A file is considered open when:
+    /// * explicitly set as an open file using [`open_file`](Self::open_file)
+    /// * It has a [`SystemPath`] and belongs to a package's `src` files
+    /// * It has a [`SystemVirtualPath`](ruff_db::system::SystemVirtualPath)
+    pub fn is_file_open(self, db: &dyn Db, file: File) -> bool {
+        if let Some(open_files) = self.open_files(db) {
+            open_files.contains(&file)
+        } else if let Some(system_path) = file.path(db).as_system_path() {
+            self.package(db, system_path)
+                .map_or(false, |package| package.contains_file(db, file))
+        } else {
+            file.path(db).is_system_virtual_path()
+        }
+    }
 }
 
-#[salsa::tracked]
 impl Package {
     pub fn root(self, db: &dyn Db) -> &SystemPath {
         self.root_buf(db)
@@ -258,13 +304,13 @@ impl Package {
 
     /// Returns `true` if `file` is a first-party file part of this package.
     pub fn contains_file(self, db: &dyn Db, file: File) -> bool {
-        self.files(db).read().contains(&file)
+        self.files(db).contains(&file)
     }
 
     #[tracing::instrument(level = "debug", skip(db))]
     pub fn remove_file(self, db: &mut dyn Db, file: File) {
         tracing::debug!(
-            "Remove file {} from package {}",
+            "Removing file '{}' from package '{}'",
             file.path(db),
             self.name(db)
         );
@@ -277,7 +323,11 @@ impl Package {
     }
 
     pub fn add_file(self, db: &mut dyn Db, file: File) {
-        tracing::debug!("Add file {} to package {}", file.path(db), self.name(db));
+        tracing::debug!(
+            "Adding file '{}' to package '{}'",
+            file.path(db),
+            self.name(db)
+        );
 
         let Some(mut index) = PackageFiles::indexed_mut(db, self) else {
             return;
@@ -286,29 +336,17 @@ impl Package {
         index.insert(file);
     }
 
-    #[tracing::instrument(level = "debug", skip(db))]
-    pub(crate) fn check(self, db: &dyn Db) -> Vec<String> {
-        tracing::debug!("Checking package {}", self.root(db));
-
-        let mut result = Vec::new();
-        for file in &self.files(db).read() {
-            let diagnostics = check_file(db, file);
-            result.extend_from_slice(&diagnostics);
-        }
-
-        result
-    }
-
     /// Returns the files belonging to this package.
-    #[salsa::tracked]
-    pub fn files(self, db: &dyn Db) -> IndexedFiles {
-        let _entered = tracing::debug_span!("files").entered();
+    pub fn files(self, db: &dyn Db) -> Indexed<'_> {
         let files = self.file_set(db);
 
         let indexed = match files.get() {
             Index::Lazy(vacant) => {
-                tracing::debug!("Indexing files for package {}", self.name(db));
+                let _entered =
+                    tracing::debug_span!("index_package_files", package = %self.name(db)).entered();
+
                 let files = discover_package_files(db, self.root(db));
+                tracing::info!("Found {} files in package '{}'", files.len(), self.name(db));
                 vacant.set(files)
             }
             Index::Indexed(indexed) => indexed,
@@ -329,14 +367,12 @@ impl Package {
         assert_eq!(root, metadata.root());
 
         if self.name(db) != metadata.name() {
-            self.set_name(db)
-                .with_durability(Durability::MEDIUM)
-                .to(metadata.name);
+            self.set_name(db).to(metadata.name);
         }
     }
 
     pub fn reload_files(self, db: &mut dyn Db) {
-        tracing::debug!("Reload files for package {}", self.name(db));
+        tracing::debug!("Reloading files for package '{}'", self.name(db));
 
         if !self.file_set(db).is_lazy() {
             // Force a re-index of the files in the next revision.
@@ -345,15 +381,51 @@ impl Package {
     }
 }
 
-pub(super) fn check_file(db: &dyn Db, file: File) -> Diagnostics {
-    let path = file.path(db);
-    let _span = tracing::debug_span!("check_file", file=%path).entered();
-    tracing::debug!("Checking file {path}");
+#[salsa::tracked]
+pub(super) fn check_file(db: &dyn Db, file: File) -> Vec<String> {
+    tracing::debug!("Checking file '{path}'", path = file.path(db));
 
     let mut diagnostics = Vec::new();
+
+    let source_diagnostics = source_text::accumulated::<SourceDiagnostic>(db.upcast(), file);
+    // TODO(micha): Consider using a single accumulator for all diagnostics
+    diagnostics.extend(
+        source_diagnostics
+            .iter()
+            .map(std::string::ToString::to_string),
+    );
+
+    // Abort checking if there are IO errors.
+    let source = source_text(db.upcast(), file);
+
+    if source.has_read_error() {
+        return diagnostics;
+    }
+
+    let parsed = parsed_module(db.upcast(), file);
+
+    if !parsed.errors().is_empty() {
+        let path = file.path(db);
+        let line_index = line_index(db.upcast(), file);
+        diagnostics.extend(parsed.errors().iter().map(|err| {
+            let source_location = line_index.source_location(err.location.start(), source.as_str());
+            format!("{path}:{source_location}: {message}", message = err.error)
+        }));
+    }
+
+    for diagnostic in check_types(db.upcast(), file) {
+        let index = line_index(db.upcast(), diagnostic.file());
+        let location = index.source_location(diagnostic.start(), source.as_str());
+        diagnostics.push(format!(
+            "{path}:{location}: {message}",
+            path = file.path(db),
+            message = diagnostic.message()
+        ));
+    }
+
     diagnostics.extend_from_slice(lint_syntax(db, file));
     diagnostics.extend_from_slice(lint_semantic(db, file));
-    Diagnostics::from(diagnostics)
+    diagnostics
 }
 
 fn discover_package_files(db: &dyn Db, path: &SystemPath) -> FxHashSet<File> {
@@ -398,3 +470,113 @@ fn discover_package_files(db: &dyn Db, path: &SystemPath) -> FxHashSet<File> {
 
     files
 }
+
+#[derive(Debug)]
+enum WorkspaceFiles<'a> {
+    OpenFiles(&'a FxHashSet<File>),
+    PackageFiles(Vec<Indexed<'a>>),
+}
+
+impl<'a> WorkspaceFiles<'a> {
+    fn new(db: &'a dyn Db, workspace: Workspace) -> Self {
+        if let Some(open_files) = workspace.open_files(db) {
+            WorkspaceFiles::OpenFiles(open_files)
+        } else {
+            WorkspaceFiles::PackageFiles(
+                workspace
+                    .packages(db)
+                    .map(|package| package.files(db))
+                    .collect(),
+            )
+        }
+    }
+}
+
+impl<'a> IntoIterator for &'a WorkspaceFiles<'a> {
+    type Item = File;
+    type IntoIter = WorkspaceFilesIter<'a>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        match self {
+            WorkspaceFiles::OpenFiles(files) => WorkspaceFilesIter::OpenFiles(files.iter()),
+            WorkspaceFiles::PackageFiles(package_files) => {
+                let mut package_files = package_files.iter();
+                WorkspaceFilesIter::PackageFiles {
+                    current: package_files.next().map(IntoIterator::into_iter),
+                    package_files,
+                }
+            }
+        }
+    }
+}
+
+enum WorkspaceFilesIter<'db> {
+    OpenFiles(std::collections::hash_set::Iter<'db, File>),
+    PackageFiles {
+        package_files: std::slice::Iter<'db, Indexed<'db>>,
+        current: Option<IndexedIter<'db>>,
+    },
+}
+
+impl Iterator for WorkspaceFilesIter<'_> {
+    type Item = File;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        match self {
+            WorkspaceFilesIter::OpenFiles(files) => files.next().copied(),
+            WorkspaceFilesIter::PackageFiles {
+                package_files,
+                current,
+            } => loop {
+                if let Some(file) = current.as_mut().and_then(Iterator::next) {
+                    return Some(file);
+                }
+
+                *current = Some(package_files.next()?.into_iter());
+            },
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use ruff_db::files::system_path_to_file;
+    use ruff_db::source::source_text;
+    use ruff_db::system::{DbWithTestSystem, SystemPath};
+    use ruff_db::testing::assert_function_query_was_not_run;
+
+    use crate::db::tests::TestDb;
+    use crate::lint::lint_syntax;
+    use crate::workspace::check_file;
+
+    #[test]
+    fn check_file_skips_linting_when_file_cant_be_read() -> ruff_db::system::Result<()> {
+        let mut db = TestDb::new();
+        let path = SystemPath::new("test.py");
+
+        db.write_file(path, "x = 10")?;
+        let file = system_path_to_file(&db, path).unwrap();
+
+        // Now the file gets deleted before we had a chance to read its source text.
+        db.memory_file_system().remove_file(path)?;
+        file.sync(&mut db);
+
+        assert_eq!(source_text(&db, file).as_str(), "");
+        assert_eq!(
+            check_file(&db, file),
+            vec!["Failed to read file: No such file or directory".to_string()]
+        );
+
+        let events = db.take_salsa_events();
+        assert_function_query_was_not_run(&db, lint_syntax, file, &events);
+
+        // The user now creates a new file with an empty text. The source text
+        // content returned by `source_text` remains unchanged, but the diagnostics should get updated.
+        db.write_file(path, "").unwrap();
+
+        assert_eq!(source_text(&db, file).as_str(), "");
+        assert_eq!(check_file(&db, file), vec![] as Vec<String>);
+
+        Ok(())
+    }
+}

crates/red_knot_workspace/src/workspace/files.rs

@@ -1,4 +1,4 @@
-use std::iter::FusedIterator;
+use std::marker::PhantomData;
 use std::ops::Deref;
 use std::sync::Arc;
 
@@ -10,6 +10,9 @@ use ruff_db::files::File;
 use crate::db::Db;
 use crate::workspace::Package;
 
+/// Cheap cloneable hash set of files.
+type FileSet = Arc<FxHashSet<File>>;
+
 /// The indexed files of a package.
 ///
 /// The indexing happens lazily, but the files are then cached for subsequent reads.
@@ -18,7 +21,7 @@ use crate::workspace::Package;
 /// The implementation uses internal mutability to transition between the lazy and indexed state
 /// without triggering a new salsa revision. This is safe because the initial indexing happens on first access,
 /// so no query can be depending on the contents of the indexed files before that. All subsequent mutations to
-/// the indexed files must go through `IndexedFilesMut`, which uses the Salsa setter `package.set_file_set` to
+/// the indexed files must go through `IndexedMut`, which uses the Salsa setter `package.set_file_set` to
 /// ensure that Salsa always knows when the set of indexed files have changed.
 #[derive(Debug)]
 pub struct PackageFiles {
@@ -32,46 +35,67 @@ impl PackageFiles {
         }
     }
 
-    fn indexed(indexed_files: IndexedFiles) -> Self {
+    fn indexed(files: FileSet) -> Self {
         Self {
-            state: std::sync::Mutex::new(State::Indexed(indexed_files)),
+            state: std::sync::Mutex::new(State::Indexed(files)),
         }
     }
 
-    pub fn get(&self) -> Index {
+    pub(super) fn get(&self) -> Index {
         let state = self.state.lock().unwrap();
 
         match &*state {
             State::Lazy => Index::Lazy(LazyFiles { files: state }),
-            State::Indexed(files) => Index::Indexed(files.clone()),
+            State::Indexed(files) => Index::Indexed(Indexed {
+                files: Arc::clone(files),
+                _lifetime: PhantomData,
+            }),
         }
     }
 
-    pub fn is_lazy(&self) -> bool {
+    pub(super) fn is_lazy(&self) -> bool {
         matches!(*self.state.lock().unwrap(), State::Lazy)
     }
 
     /// Returns a mutable view on the index that allows cheap in-place mutations.
     ///
     /// The changes are automatically written back to the database once the view is dropped.
-    pub fn indexed_mut(db: &mut dyn Db, package: Package) -> Option<IndexedFilesMut> {
+    pub(super) fn indexed_mut(db: &mut dyn Db, package: Package) -> Option<IndexedMut> {
         // Calling `zalsa_mut` cancels all pending salsa queries. This ensures that there are no pending
         // reads to the file set.
         // TODO: Use a non-internal API instead https://salsa.zulipchat.com/#narrow/stream/333573-salsa-3.2E0/topic/Expose.20an.20API.20to.20cancel.20other.20queries
         let _ = db.as_dyn_database_mut().zalsa_mut();
 
-        let files = package.file_set(db);
-
-        let indexed = match &*files.state.lock().unwrap() {
-            State::Lazy => return None,
-            State::Indexed(indexed) => indexed.clone(),
+        // Replace the state with lazy. The `IndexedMut` guard restores the state
+        // to `State::Indexed`  or sets a new `PackageFiles` when it gets dropped to ensure the state
+        // is restored to how it has been before replacing the value.
+        //
+        // It isn't necessary to hold on to the lock after this point:
+        // * The above call to `zalsa_mut` guarantees that there's exactly **one** DB reference.
+        // * `Indexed` has a `'db` lifetime, and this method requires a `&mut db`.
+        //   This means that there can't be any pending reference to `Indexed` because Rust
+        //   doesn't allow borrowing `db` as mutable (to call this method) and immutable (`Indexed<'db>`) at the same time.
+        //   There can't be any other `Indexed<'db>` references created by clones of this DB because
+        //   all clones must have been dropped at this point and the `Indexed`
+        //   can't outlive the database (constrained by the `db` lifetime).
+        let state = {
+            let files = package.file_set(db);
+            let mut locked = files.state.lock().unwrap();
+            std::mem::replace(&mut *locked, State::Lazy)
         };
 
-        Some(IndexedFilesMut {
+        let indexed = match state {
+            // If it's already lazy, just return. We also don't need to restore anything because the
+            // replace above was a no-op.
+            State::Lazy => return None,
+            State::Indexed(indexed) => indexed,
+        };
+
+        Some(IndexedMut {
             db: Some(db),
             package,
-            new_revision: indexed.revision,
-            indexed,
+            files: indexed,
+            did_change: false,
         })
     }
 }
@@ -88,152 +112,95 @@ enum State {
     Lazy,
 
     /// The files are indexed. Stores the known files of a package.
-    Indexed(IndexedFiles),
+    Indexed(FileSet),
 }
 
-pub enum Index<'a> {
+pub(super) enum Index<'db> {
     /// The index has not yet been computed. Allows inserting the files.
-    Lazy(LazyFiles<'a>),
+    Lazy(LazyFiles<'db>),
 
-    Indexed(IndexedFiles),
+    Indexed(Indexed<'db>),
 }
 
 /// Package files that have not been indexed yet.
-pub struct LazyFiles<'a> {
-    files: std::sync::MutexGuard<'a, State>,
+pub(super) struct LazyFiles<'db> {
+    files: std::sync::MutexGuard<'db, State>,
 }
 
-impl<'a> LazyFiles<'a> {
+impl<'db> LazyFiles<'db> {
     /// Sets the indexed files of a package to `files`.
-    pub fn set(mut self, files: FxHashSet<File>) -> IndexedFiles {
-        let files = IndexedFiles::new(files);
-        *self.files = State::Indexed(files.clone());
+    pub(super) fn set(mut self, files: FxHashSet<File>) -> Indexed<'db> {
+        let files = Indexed {
+            files: Arc::new(files),
+            _lifetime: PhantomData,
+        };
+        *self.files = State::Indexed(Arc::clone(&files.files));
         files
     }
 }
 
 /// The indexed files of a package.
 ///
-/// # Salsa integration
-/// The type is cheap clonable and allows for in-place mutation of the files. The in-place mutation requires
-/// extra care because the type is used as the result of Salsa queries and Salsa relies on a type's equality
-/// to determine if the output has changed. This is accomplished by using a `revision` that gets incremented
-/// whenever the files are changed. The revision ensures that salsa's comparison of the
-/// previous [`IndexedFiles`] with the next [`IndexedFiles`] returns false even though they both
-/// point to the same underlying hash set.
-///
-/// # Equality
-/// Two [`IndexedFiles`] are only equal if they have the same revision and point to the **same** (identity) hash set.
-#[derive(Debug, Clone)]
-pub struct IndexedFiles {
-    revision: u64,
-    files: Arc<std::sync::Mutex<FxHashSet<File>>>,
+/// Note: This type is intentionally non-cloneable. Making it cloneable requires
+/// revisiting the locking behavior in [`PackageFiles::indexed_mut`].
+#[derive(Debug, PartialEq, Eq)]
+pub struct Indexed<'db> {
+    files: FileSet,
+    // Preserve the lifetime of `PackageFiles`.
+    _lifetime: PhantomData<&'db ()>,
 }
 
-impl IndexedFiles {
-    fn new(files: FxHashSet<File>) -> Self {
-        Self {
-            files: Arc::new(std::sync::Mutex::new(files)),
-            revision: 0,
-        }
-    }
-
-    /// Locks the file index for reading.
-    pub fn read(&self) -> IndexedFilesGuard {
-        IndexedFilesGuard {
-            guard: self.files.lock().unwrap(),
-        }
-    }
-}
-
-impl PartialEq for IndexedFiles {
-    fn eq(&self, other: &Self) -> bool {
-        self.revision == other.revision && Arc::ptr_eq(&self.files, &other.files)
-    }
-}
-
-impl Eq for IndexedFiles {}
-
-pub struct IndexedFilesGuard<'a> {
-    guard: std::sync::MutexGuard<'a, FxHashSet<File>>,
-}
-
-impl Deref for IndexedFilesGuard<'_> {
+impl Deref for Indexed<'_> {
     type Target = FxHashSet<File>;
 
     fn deref(&self) -> &Self::Target {
-        &self.guard
+        &self.files
     }
 }
 
-impl<'a> IntoIterator for &'a IndexedFilesGuard<'a> {
+pub(super) type IndexedIter<'a> = std::iter::Copied<std::collections::hash_set::Iter<'a, File>>;
+
+impl<'a> IntoIterator for &'a Indexed<'_> {
     type Item = File;
-    type IntoIter = IndexedFilesIter<'a>;
+    type IntoIter = IndexedIter<'a>;
 
     fn into_iter(self) -> Self::IntoIter {
-        IndexedFilesIter {
-            inner: self.guard.iter(),
-        }
+        self.files.iter().copied()
     }
 }
 
-/// Iterator over the indexed files.
-///
-/// # Locks
-/// Holding on to the iterator locks the file index for reading.
-pub struct IndexedFilesIter<'a> {
-    inner: std::collections::hash_set::Iter<'a, File>,
-}
-
-impl<'a> Iterator for IndexedFilesIter<'a> {
-    type Item = File;
-
-    fn next(&mut self) -> Option<Self::Item> {
-        self.inner.next().copied()
-    }
-
-    fn size_hint(&self) -> (usize, Option<usize>) {
-        self.inner.size_hint()
-    }
-}
-
-impl FusedIterator for IndexedFilesIter<'_> {}
-
-impl ExactSizeIterator for IndexedFilesIter<'_> {}
-
 /// A Mutable view of a package's indexed files.
 ///
 /// Allows in-place mutation of the files without deep cloning the hash set.
 /// The changes are written back when the mutable view is dropped or by calling [`Self::set`] manually.
-pub struct IndexedFilesMut<'db> {
+pub(super) struct IndexedMut<'db> {
     db: Option<&'db mut dyn Db>,
     package: Package,
-    indexed: IndexedFiles,
-    new_revision: u64,
+    files: FileSet,
+    did_change: bool,
 }
 
-impl IndexedFilesMut<'_> {
-    pub fn insert(&mut self, file: File) -> bool {
-        if self.indexed.files.lock().unwrap().insert(file) {
-            self.new_revision += 1;
+impl IndexedMut<'_> {
+    pub(super) fn insert(&mut self, file: File) -> bool {
+        if self.files_mut().insert(file) {
+            self.did_change = true;
             true
         } else {
             false
         }
     }
 
-    pub fn remove(&mut self, file: File) -> bool {
-        if self.indexed.files.lock().unwrap().remove(&file) {
-            self.new_revision += 1;
+    pub(super) fn remove(&mut self, file: File) -> bool {
+        if self.files_mut().remove(&file) {
+            self.did_change = true;
             true
         } else {
             false
         }
     }
 
-    /// Writes the changes back to the database.
-    pub fn set(mut self) {
-        self.set_impl();
+    fn files_mut(&mut self) -> &mut FxHashSet<File> {
+        Arc::get_mut(&mut self.files).expect("All references to `FilesSet` to have been dropped")
     }
 
     fn set_impl(&mut self) {
@@ -241,19 +208,70 @@ impl IndexedFilesMut<'_> {
             return;
         };
 
-        if self.indexed.revision != self.new_revision {
+        let files = Arc::clone(&self.files);
+
+        if self.did_change {
+            // If there are changes, set the new file_set to trigger a salsa revision change.
             self.package
                 .set_file_set(db)
-                .to(PackageFiles::indexed(IndexedFiles {
-                    revision: self.new_revision,
-                    files: self.indexed.files.clone(),
-                }));
+                .to(PackageFiles::indexed(files));
+        } else {
+            // The `indexed_mut` replaced the `state` with Lazy. Restore it back to the indexed state.
+            *self.package.file_set(db).state.lock().unwrap() = State::Indexed(files);
         }
     }
 }
 
-impl Drop for IndexedFilesMut<'_> {
+impl Drop for IndexedMut<'_> {
     fn drop(&mut self) {
         self.set_impl();
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use rustc_hash::FxHashSet;
+
+    use ruff_db::files::system_path_to_file;
+    use ruff_db::system::{DbWithTestSystem, SystemPathBuf};
+    use ruff_python_ast::name::Name;
+
+    use crate::db::tests::TestDb;
+    use crate::workspace::files::Index;
+    use crate::workspace::Package;
+
+    #[test]
+    fn re_entrance() -> anyhow::Result<()> {
+        let mut db = TestDb::new();
+
+        db.write_file("test.py", "")?;
+
+        let package = Package::new(&db, Name::new("test"), SystemPathBuf::from("/test"));
+
+        let file = system_path_to_file(&db, "test.py").unwrap();
+
+        let files = match package.file_set(&db).get() {
+            Index::Lazy(lazy) => lazy.set(FxHashSet::from_iter([file])),
+            Index::Indexed(files) => files,
+        };
+
+        // Calling files a second time should not dead-lock.
+        // This can e.g. happen when `check_file` iterates over all files and
+        // `is_file_open` queries the open files.
+        let files_2 = package.file_set(&db).get();
+
+        match files_2 {
+            Index::Lazy(_) => {
+                panic!("Expected indexed files, got lazy files");
+            }
+            Index::Indexed(files_2) => {
+                assert_eq!(
+                    files_2.iter().collect::<Vec<_>>(),
+                    files.iter().collect::<Vec<_>>()
+                );
+            }
+        }
+
+        Ok(())
+    }
+}

crates/red_knot_workspace/src/workspace/metadata.rs

@@ -1,3 +1,4 @@
+use crate::workspace::settings::{Configuration, WorkspaceSettings};
 use ruff_db::system::{System, SystemPath, SystemPathBuf};
 use ruff_python_ast::name::Name;
 
@@ -7,6 +8,8 @@ pub struct WorkspaceMetadata {
 
     /// The (first-party) packages in this workspace.
     pub(super) packages: Vec<PackageMetadata>,
+
+    pub(super) settings: WorkspaceSettings,
 }
 
 /// A first-party package in a workspace.
@@ -21,7 +24,11 @@ pub struct PackageMetadata {
 
 impl WorkspaceMetadata {
     /// Discovers the closest workspace at `path` and returns its metadata.
-    pub fn from_path(path: &SystemPath, system: &dyn System) -> anyhow::Result<WorkspaceMetadata> {
+    pub fn from_path(
+        path: &SystemPath,
+        system: &dyn System,
+        base_configuration: Option<Configuration>,
+    ) -> anyhow::Result<WorkspaceMetadata> {
         assert!(
             system.is_directory(path),
             "Workspace root path must be a directory"
@@ -38,9 +45,20 @@ impl WorkspaceMetadata {
             root: root.clone(),
         };
 
+        // TODO: Load the configuration from disk.
+        let mut configuration = Configuration::default();
+
+        if let Some(base_configuration) = base_configuration {
+            configuration.extend(base_configuration);
+        }
+
+        // TODO: Respect the package configurations when resolving settings (e.g. for the target version).
+        let settings = configuration.into_workspace_settings(&root);
+
         let workspace = WorkspaceMetadata {
             root,
             packages: vec![package],
+            settings,
         };
 
         Ok(workspace)
@@ -53,6 +71,10 @@ impl WorkspaceMetadata {
     pub fn packages(&self) -> &[PackageMetadata] {
         &self.packages
     }
+
+    pub fn settings(&self) -> &WorkspaceSettings {
+        &self.settings
+    }
 }
 
 impl PackageMetadata {

crates/red_knot_workspace/src/workspace/settings.rs

@@ -0,0 +1,89 @@
+use red_knot_python_semantic::{ProgramSettings, PythonVersion, SearchPathSettings, SitePackages};
+use ruff_db::system::{SystemPath, SystemPathBuf};
+
+/// The resolved configurations.
+///
+/// The main difference to [`Configuration`] is that default values are filled in.
+#[derive(Debug, Clone)]
+pub struct WorkspaceSettings {
+    pub(super) program: ProgramSettings,
+}
+
+impl WorkspaceSettings {
+    pub fn program(&self) -> &ProgramSettings {
+        &self.program
+    }
+}
+
+/// The configuration for the workspace or a package.
+#[derive(Debug, Default, Clone)]
+pub struct Configuration {
+    pub target_version: Option<PythonVersion>,
+    pub search_paths: SearchPathConfiguration,
+}
+
+impl Configuration {
+    /// Extends this configuration by using the values from `with` for all values that are absent in `self`.
+    pub fn extend(&mut self, with: Configuration) {
+        self.target_version = self.target_version.or(with.target_version);
+        self.search_paths.extend(with.search_paths);
+    }
+
+    pub fn into_workspace_settings(self, workspace_root: &SystemPath) -> WorkspaceSettings {
+        WorkspaceSettings {
+            program: ProgramSettings {
+                target_version: self.target_version.unwrap_or_default(),
+                search_paths: self.search_paths.into_settings(workspace_root),
+            },
+        }
+    }
+}
+
+#[derive(Debug, Default, Clone, Eq, PartialEq)]
+pub struct SearchPathConfiguration {
+    /// List of user-provided paths that should take first priority in the module resolution.
+    /// Examples in other type checkers are mypy's MYPYPATH environment variable,
+    /// or pyright's stubPath configuration setting.
+    pub extra_paths: Option<Vec<SystemPathBuf>>,
+
+    /// The root of the workspace, used for finding first-party modules.
+    pub src_root: Option<SystemPathBuf>,
+
+    /// Optional path to a "custom typeshed" directory on disk for us to use for standard-library types.
+    /// If this is not provided, we will fallback to our vendored typeshed stubs for the stdlib,
+    /// bundled as a zip file in the binary
+    pub custom_typeshed: Option<SystemPathBuf>,
+
+    /// The path to the user's `site-packages` directory, where third-party packages from ``PyPI`` are installed.
+    pub site_packages: Option<SitePackages>,
+}
+
+impl SearchPathConfiguration {
+    pub fn into_settings(self, workspace_root: &SystemPath) -> SearchPathSettings {
+        let site_packages = self.site_packages.unwrap_or(SitePackages::Known(vec![]));
+
+        SearchPathSettings {
+            extra_paths: self.extra_paths.unwrap_or_default(),
+            src_root: self
+                .src_root
+                .unwrap_or_else(|| workspace_root.to_path_buf()),
+            custom_typeshed: self.custom_typeshed,
+            site_packages,
+        }
+    }
+
+    pub fn extend(&mut self, with: SearchPathConfiguration) {
+        if let Some(extra_paths) = with.extra_paths {
+            self.extra_paths.get_or_insert(extra_paths);
+        }
+        if let Some(src_root) = with.src_root {
+            self.src_root.get_or_insert(src_root);
+        }
+        if let Some(custom_typeshed) = with.custom_typeshed {
+            self.custom_typeshed.get_or_insert(custom_typeshed);
+        }
+        if let Some(site_packages) = with.site_packages {
+            self.site_packages.get_or_insert(site_packages);
+        }
+    }
+}

crates/red_knot_workspace/tests/check.rs

@@ -1,45 +1,130 @@
-use red_knot_python_semantic::{ProgramSettings, PythonVersion, SearchPathSettings};
-use red_knot_workspace::db::RootDatabase;
-use red_knot_workspace::lint::lint_semantic;
-use red_knot_workspace::workspace::WorkspaceMetadata;
-use ruff_db::files::system_path_to_file;
-use ruff_db::system::{OsSystem, SystemPathBuf};
 use std::fs;
 use std::path::PathBuf;
 
-fn setup_db(workspace_root: SystemPathBuf) -> anyhow::Result<RootDatabase> {
-    let system = OsSystem::new(&workspace_root);
-    let workspace = WorkspaceMetadata::from_path(&workspace_root, &system)?;
-    let search_paths = SearchPathSettings {
-        extra_paths: vec![],
-        src_root: workspace_root,
-        custom_typeshed: None,
-        site_packages: vec![],
-    };
-    let settings = ProgramSettings {
-        target_version: PythonVersion::default(),
-        search_paths,
-    };
-    let db = RootDatabase::new(workspace, settings, system);
-    Ok(db)
+use red_knot_python_semantic::{HasTy, SemanticModel};
+use red_knot_workspace::db::RootDatabase;
+use red_knot_workspace::workspace::WorkspaceMetadata;
+use ruff_db::files::{system_path_to_file, File};
+use ruff_db::parsed::parsed_module;
+use ruff_db::system::{OsSystem, SystemPath, SystemPathBuf};
+use ruff_python_ast::visitor::source_order;
+use ruff_python_ast::visitor::source_order::SourceOrderVisitor;
+use ruff_python_ast::{Alias, Expr, Parameter, ParameterWithDefault, Stmt};
+
+fn setup_db(workspace_root: &SystemPath) -> anyhow::Result<RootDatabase> {
+    let system = OsSystem::new(workspace_root);
+    let workspace = WorkspaceMetadata::from_path(workspace_root, &system, None)?;
+    RootDatabase::new(workspace, system)
 }
 
 /// Test that all snippets in testcorpus can be checked without panic
 #[test]
 #[allow(clippy::print_stdout)]
 fn corpus_no_panic() -> anyhow::Result<()> {
-    let corpus = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("resources/test/corpus");
-    let system_corpus =
-        SystemPathBuf::from_path_buf(corpus.clone()).expect("corpus path to be UTF8");
-    let db = setup_db(system_corpus.clone())?;
+    let root = SystemPathBuf::from_path_buf(tempfile::TempDir::new()?.into_path()).unwrap();
+    let db = setup_db(&root)?;
 
-    for path in fs::read_dir(&corpus).expect("corpus to be a directory") {
-        let path = path.expect("path to not be an error").path();
-        println!("checking {path:?}");
-        let path = SystemPathBuf::from_path_buf(path.clone()).expect("path to be UTF-8");
-        // this test is only asserting that we can run the lint without a panic
-        let file = system_path_to_file(&db, path).expect("file to exist");
-        lint_semantic(&db, file);
+    let corpus = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("resources/test/corpus");
+
+    for path in fs::read_dir(&corpus)? {
+        let source = path?.path();
+        println!("checking {source:?}");
+        let source_fn = source.file_name().unwrap().to_str().unwrap();
+        let py_dest = root.join(source_fn);
+        fs::copy(&source, py_dest.as_std_path())?;
+        // this test is only asserting that we can pull every expression type without a panic
+        // (and some non-expressions that clearly define a single type)
+        let file = system_path_to_file(&db, py_dest).unwrap();
+        pull_types(&db, file);
+
+        // try the file as a stub also
+        println!("re-checking as .pyi");
+        let pyi_dest = root.join(format!("{source_fn}i"));
+        std::fs::copy(source, pyi_dest.as_std_path())?;
+        let file = system_path_to_file(&db, pyi_dest).unwrap();
+        pull_types(&db, file);
     }
     Ok(())
 }
+
+fn pull_types(db: &RootDatabase, file: File) {
+    let mut visitor = PullTypesVisitor::new(db, file);
+
+    let ast = parsed_module(db, file);
+
+    visitor.visit_body(ast.suite());
+}
+
+struct PullTypesVisitor<'db> {
+    model: SemanticModel<'db>,
+}
+
+impl<'db> PullTypesVisitor<'db> {
+    fn new(db: &'db RootDatabase, file: File) -> Self {
+        Self {
+            model: SemanticModel::new(db, file),
+        }
+    }
+}
+
+impl SourceOrderVisitor<'_> for PullTypesVisitor<'_> {
+    fn visit_stmt(&mut self, stmt: &Stmt) {
+        match stmt {
+            Stmt::FunctionDef(function) => {
+                let _ty = function.ty(&self.model);
+            }
+            Stmt::ClassDef(class) => {
+                let _ty = class.ty(&self.model);
+            }
+            Stmt::AnnAssign(_)
+            | Stmt::Return(_)
+            | Stmt::Delete(_)
+            | Stmt::Assign(_)
+            | Stmt::AugAssign(_)
+            | Stmt::TypeAlias(_)
+            | Stmt::For(_)
+            | Stmt::While(_)
+            | Stmt::If(_)
+            | Stmt::With(_)
+            | Stmt::Match(_)
+            | Stmt::Raise(_)
+            | Stmt::Try(_)
+            | Stmt::Assert(_)
+            | Stmt::Import(_)
+            | Stmt::ImportFrom(_)
+            | Stmt::Global(_)
+            | Stmt::Nonlocal(_)
+            | Stmt::Expr(_)
+            | Stmt::Pass(_)
+            | Stmt::Break(_)
+            | Stmt::Continue(_)
+            | Stmt::IpyEscapeCommand(_) => {}
+        }
+
+        source_order::walk_stmt(self, stmt);
+    }
+
+    fn visit_expr(&mut self, expr: &Expr) {
+        let _ty = expr.ty(&self.model);
+
+        source_order::walk_expr(self, expr);
+    }
+
+    fn visit_parameter(&mut self, parameter: &Parameter) {
+        let _ty = parameter.ty(&self.model);
+
+        source_order::walk_parameter(self, parameter);
+    }
+
+    fn visit_parameter_with_default(&mut self, parameter_with_default: &ParameterWithDefault) {
+        let _ty = parameter_with_default.ty(&self.model);
+
+        source_order::walk_parameter_with_default(self, parameter_with_default);
+    }
+
+    fn visit_alias(&mut self, alias: &Alias) {
+        let _ty = alias.ty(&self.model);
+
+        source_order::walk_alias(self, alias);
+    }
+}

crates/ruff/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "ruff"
-version = "0.5.7"
+version = "0.6.7"
 publish = true
 authors = { workspace = true }
 edition = { workspace = true }
@@ -14,7 +14,9 @@ default-run = "ruff"
 
 [dependencies]
 ruff_cache = { workspace = true }
+ruff_db = { workspace = true, default-features = false, features = ["os"] }
 ruff_diagnostics = { workspace = true }
+ruff_graph = { workspace = true, features = ["serde", "clap"] }
 ruff_linter = { workspace = true, features = ["clap"] }
 ruff_macros = { workspace = true }
 ruff_notebook = { workspace = true }
@@ -36,6 +38,7 @@ clap_complete_command = { workspace = true }
 clearscreen = { workspace = true }
 colored = { workspace = true }
 filetime = { workspace = true }
+globwalk = { workspace = true }
 ignore = { workspace = true }
 is-macro = { workspace = true }
 itertools = { workspace = true }
@@ -59,8 +62,11 @@ wild = { workspace = true }
 [dev-dependencies]
 # Enable test rules during development
 ruff_linter = { workspace = true, features = ["clap", "test-rules"] }
+
+assert_fs = { workspace = true }
 # Avoid writing colored snapshots when running tests from the terminal
 colored = { workspace = true, features = ["no-color"] }
+indoc = { workspace = true }
 insta = { workspace = true, features = ["filters", "json"] }
 insta-cmd = { workspace = true }
 tempfile = { workspace = true }