AST optimizer skips PEP 696 type parameter defaults

[JelleZijlstra]

Bug report

Bug description:

>>> dis.dis("def f[T = 1 + 2](): pass")
... snip ...
      L1:     LOAD_CONST               0 (1)
              LOAD_CONST               1 (2)
              BINARY_OP                0 (+)

... snip ...

cc @Eclips4. This is easy to fix but it might be better to wait for #122667 so we can test the fix better.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-123344: Add missing ast optimizations for PEP 696 #123377
• [3.13] gh-123344: Add missing ast optimizations for PEP 696 (GH-123377) #123427

[sobolevn]

diff --git Python/ast_opt.c Python/ast_opt.c
index d7a26e64150..8907c6daec7 100644
--- Python/ast_opt.c
+++ Python/ast_opt.c
@@ -1087,6 +1087,7 @@ astfold_type_param(type_param_ty node_, PyArena *ctx_, _PyASTOptimizeState *stat
     switch (node_->kind) {
         case TypeVar_kind:
             CALL_OPT(astfold_expr, expr_ty, node_->v.TypeVar.bound);
+            CALL_OPT(astfold_expr, expr_ty, node_->v.TypeVar.default_value);
             break;
         case ParamSpec_kind:
             break;

This might help someone in the future, I tested it locally: echo 'def f[T = 1 + 2](): pass' | ./python.exe -m dis

LOAD_CONST               3 (<code object f at 0x1019cab60, file "<stdin>", line 1>)

[JelleZijlstra]

The same change is also needed for the ParamSpec and TypeVarTuple cases though.

[Eclips4]

If no one complains, I would like to work with @wrongnull on this

[picnixz]

Is this one completed or is there something else to do?