[colbymchenry]

Summary

A Python class with a parameterized base, like class Leaf(Base[str]), class Repo(models.Base[int]) or class Both(Mixin, Base[str]), is now linked to that base, in extraction and in the kernel alike. A call through a module global holding such a class, like current = Leaf() then current.run(), now reaches the method the base declares.

Lands #2998 by @JosefAschauer.

Root cause

The grammar parses Base[str] in a base list as a subscript, and the base-list reader took only bare identifiers and attributes, so the base was skipped. The class had no extends edge, and everything that follows the class hierarchy missed it. The resolver's own text reader, pythonBases, also looked up Base[str] verbatim and split Pair[str, int] at its inner comma.

What came from the contributor, and what we changed

From #2998, unchanged: the subscript unwrap in src/extraction/tree-sitter.ts and in the kernel's python.rs, the bracket drop in pythonBases, and the test file.

Ours:

• The test's kernel arm now fails instead of skipping when a build expects a kernel (CODEGRAPH_KERNEL_EXPECT=1), as the other kernel-gated tests do.
• The CHANGELOG bullet moved next to the other Python bullets, with the credit added.

Validation

Stage A: every changed line read, including the Rust. Scanner clean: the only hits are the test's temp-dir writes, its CODEGRAPH_KERNEL* env toggles, which it restores, and its ../src import. No binaries; the kernel was built here from python.rs.

Reproduced on main: class Leaf(Base[str]) has no extends edge, on the wasm path and in the kernel.

Tests: the test file runs on both arms. Red on main's source and kernel (12 of 16 fail, 6 per arm), green here (16 passed).

Kernel parity with this kernel, on 3,155 Python files of flask, requests, mypy, litestar, sqlalchemy, polar and django: 0 diffs, and the same 5 files defer to wasm as on main. The kernel and wasm edge dumps of flask, litestar, mypy and sqlalchemy are identical.

A/B against main, kernel on, every arm built from its own sources:

Repo	Edges	Added	Removed	Target moved	Same target, new confidence
psf/requests @ 3c7baaa	2,976 → 2,976	0	0	0	0
pallets/flask @ 49943b5	5,287 → 5,292	5	0	0	12
litestar-org/litestar @ 716f8b0	42,289 → 42,397	108	0	12	43
python/mypy @ d8a43fb	70,754 → 70,811	57	0	0	0
sqlalchemy/sqlalchemy @ 2d8e649	228,406 → 229,431	1,025	0	82	159
polarsource/polar server/polar @ 8dd9956	62,008 → 62,305	297	0	513	9
django/django django/ @ a6a5d1e	46,228 → 46,228	0	0	0	0
expressjs/express @ 1c6f1b0 (control)	2,296 → 2,296	0	0	0	0
gin-gonic/gin @ 0f09c3a (control)	8,073 → 8,073	0	0	0	0

Both controls have byte-identical full dumps. polar and django are their package folders, to stay under the heavy-index gate; whole, they index 2,103 and 3,027 files.

Every edge class in the 7 Python repos, 457,948 edges on main:

Change	Edges	Class	Read as
Added	610	extends by import	Right
Added	291	extends by name	288 in the same file, right. The 3 across files are wrong (see below)
Added	555	calls at 0.9	Right in a sample of 20: t.insert() reaches TableClause.insert
Added	36	references at 0.9	Same walk up the hierarchy, right in the buckets read
Moved	492	calls 0.4 → 0.9	491 in polar: self.get_one_or_none() moves from the protocol stub RepositoryProtocol to RepositoryBase and its mixins
Moved	41	calls 0.7 → 0.9	litestar's GenericSyncMockRepository moves from AbstractAsyncRepository to AbstractSyncRepository, and sqlalchemy's Insert.ext from Query.ext to HasSyntaxExtensions.ext
Moved	47	calls on a receiver, 0.65 or 0.8 → 0.9	t.select() moves from FunctionElement.select to FromClause.select, and polar's repo.get_by_id() from another repository to the mixin
Moved	12	calls 0.9 → 0.4 or 0.7	Less right: sqlalchemy's self.op() moves from Operators.op to the typing-only overload SQLCoreOperations.op
Moved	15	The rest	Right, except one Enum.copy that moves to TypeEngine.copy, a base further up the method resolution order
Same target	223	calls, higher confidence	Right: the hierarchy now proves what was a guess
Removed	0

We read 44 sampled changes and every bucket under 50.

Cost, with #3001 on top: sqlalchemy, median of 3 interleaved runs: 23.7 s on main, 25.2 s here, with 1,143 more edges.

Suite: 832 files (826 passed, 6 skipped), 9,803 tests passed, 0 failed, 87 skipped, on this exact commit, with this kernel staged, 4 workers.

Not in this PR

• A base the file imports from outside the project still links by name to a same-named project class. This is not new: on main, sqlalchemy's 28 class X(Enum) with from enum import Enum link to its own sql/sqltypes.py Enum, and litestar's 21 pydantic BaseModel classes link to a docs example. This PR adds 3 more of that kind in sqlalchemy, because Sequence[_T], Tuple[int, ...] and MutableSet[_T] from typing now reach the name match like their plain forms do. None of the three brought a follow-on call edge. The fix belongs in its own PR, for plain and parameterized bases together.
• fix(python): a base list wrapped over several lines is read whole #3001 is stacked on this and lands next.

🤖 Generated with Claude Code