[mypyc] feat: specialize isinstance for tuple of primitive types#19949
[mypyc] feat: specialize isinstance for tuple of primitive types#19949JukkaL merged 14 commits intopython:masterfrom
Conversation
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
…obTheBuidler/mypy into isinstance-tuple-of-primitives
|
Not really sure where a run test best fits. Maybe the IR test is straightforward enough? |
JukkaL
left a comment
JukkaL
left a comment
There was a problem hiding this comment.
What about adding a run test to run-misc.test?
| from typing import Any | ||
|
|
||
| def is_instance(x: Any) -> bool: | ||
| return isinstance(x, (str, int, bytes)) |
There was a problem hiding this comment.
Test also single-item tuple as an edge case.
|
Both of those just as run tests, or IR too? |
|
Added run tests for various cases. no additional IR tests at this time. |
JukkaL
left a comment
JukkaL
left a comment
There was a problem hiding this comment.
Thanks for adding the tests! Looks mostly good now, just one thing remains.
mypyc/irbuild/specialize.py
Outdated
| is_last = i == len(descs) - 1 | ||
| next_block = fail_block if is_last else BasicBlock() | ||
| builder.add_bool_branch( | ||
| builder.primitive_op(desc, [obj], expr.line), pass_block, next_block # type: ignore [arg-type] |
There was a problem hiding this comment.
Can you avoid the # type: ignore? If there is some typing issue, I'd prefer a cast since it's more specific and tells which argument is the problematic one.
There was a problem hiding this comment.
Done. It was because descs has type list[PrimitiveDescription | None]
I'm assuming the if None in descs: return None can probably be considered as a TypeGuard of sorts by mypy but that's above my paygrade
for more information, see https://pre-commit.ci
| is_last = i == len(descs) - 1 | ||
| next_block = fail_block if is_last else BasicBlock() | ||
| builder.add_bool_branch( | ||
| builder.primitive_op(cast(PrimitiveDescription, desc), [obj], expr.line), |
There was a problem hiding this comment.
Here a slightly better way to narrow the type would be to use assert desc is not None before the call, but this is fine as well.
2 participants
This PR specializes isinstance calls where the type argument is a tuple of primitive types.
We can skip tuple creation and the associated refcounting, and daisy-chain the primitive checks with an early exit option at each step.