[mypyc] Fix allow_interpreted_subclasses not seeing subclass attrs#21013
Open
VaggelisD wants to merge 3 commits intopython:masterfrom
Open
[mypyc] Fix allow_interpreted_subclasses not seeing subclass attrs#21013VaggelisD wants to merge 3 commits intopython:masterfrom
VaggelisD wants to merge 3 commits intopython:masterfrom
Conversation
p-sawicki
reviewed
Mar 19, 2026
…e overrides When a compiled class with allow_interpreted_subclasses=True has methods that access self.ATTR via direct C struct slots, interpreted subclasses that override ATTR in their class __dict__ are ignored — the compiled method always reads the base class default from the slot. Fix: in visit_get_attr for non-property attribute access, check if the instance is a mypyc-compiled type (via a new CPy_TPFLAGS_MYPYC_COMPILED tp_flags bit). If not, fall back to PyObject_GenericGetAttr which respects the MRO and finds the subclass override. Using tp_flags rather than an exact type check ensures compiled subclasses retain fast direct struct access, while only interpreted subclasses hit the GenericGetAttr slow path. For unboxed types (bool, int), the PyObject* result is unboxed to the expected C type.
…omment - Change CPy_TPFLAGS_MYPYC_COMPILED from bit 20 (Py_TPFLAGS_IS_ABSTRACT) to bit 21, which is unused across all CPython versions (3.8-3.14) - Reword test comment to not claim direct struct access, since a runtime test cannot prove which code path was taken
VaggelisD
force-pushed
the
fix-interpreted-subclass-attr
branch
from
6ac30b0 to
96e99aa
Compare
for more information, see https://pre-commit.ci
JukkaL
reviewed
Mar 20, 2026
| // Flag bit set on all mypyc-compiled types. Used to distinguish compiled | ||
| // subclasses (safe for direct struct access) from interpreted subclasses | ||
| // (need PyObject_GenericGetAttr fallback) in allow_interpreted_subclasses mode. | ||
| #define CPy_TPFLAGS_MYPYC_COMPILED (1UL << 21) |
Collaborator
There was a problem hiding this comment.
Instead of using TPFLAGS, could we add a flags field to the mypyc vtable for this purpose? This way we wouldn't need to touch CPython internal flags.
Contributor
Author
There was a problem hiding this comment.
Do you have additional pointers on how that'd look? Would we make the first vtable slot carry the flags?
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
When a compiled class with
allow_interpreted_subclasses=Truehas methods that accessself.ATTRvia direct C struct slots, interpreted subclasses that overrideATTRin their class__dict__are ignored; the compiled method always reads the base class default from the slot.The fix: In
visit_get_attrfor non-property attribute access, check if the instance is a mypyc-compiled type (via a newCPy_TPFLAGS_MYPYC_COMPILEDtp_flagsbit). If not, fall back toPyObject_GenericGetAttrwhich respects the MRO and finds the subclass override.Using
tp_flagsrather than an exact type check ensures compiled subclasses retain fast direct struct access, while only interpreted subclasses hit theGenericGetAttrslow path.For unboxed types (bool, int), the
PyObject*result is unboxed to the expected C type.EDIT:
Benchmark:
get_x()on compiled vs interpreted subclasses (50M iters)On master, both compiled and interpreted children use direct struct access (buggy behavior)
For this PR, compiled children retain their cost while interpreted children correctly fall back to
GenericGetAttrwhich is slower, but now correct since they can see overridden attributes.