Mypy Fun of the Day! (or Pants Caching Fun) I edited files in `src/python/pants/backend/python/providers/python_build_standalone`and am now getting seemingly unrelated type errors in existing Pants code:

Partition #2 - python-default, ['CPython==3.11.*']:
src/python/pants/engine/collection.py: note: In class "Collection":
src/python/pants/engine/collection.py:55:16: error: Incompatible types in assignment (expression has type "Callable[[], int]", base class "tuple" defined the type as "Callable[[tuple[T, ...]], int]")  [assignment]
        __hash__ = Tuple.__hash__
                   ^~~~~~~~~~~~~~
src/python/pants/engine/target.py: note: In class "CoarsenedTargets":
src/python/pants/engine/target.py:939:16: error: Incompatible types in assignment (expression has type "Callable[[], int]", base class "tuple" defined the type as "Callable[[tuple[CoarsenedTarget, ...]], int]")  [assignment]
        __hash__ = Tuple.__hash__
                   ^~~~~~~~~~~~~~

Are you running

check

with

--changed-since=main

or something like that?

nope, just

./pants fmt lint check src/python/pants::

Is that on macos? I have seen issues like this sometimes, where mypy complains when I run it locally but passes in CI

Yes on macOS

Indeed!

Maybe adjacent to https://github.com/pantsbuild/pants/issues/18519 ?

seems spot on

I imagine it's weird to you also that that latent issue has suddenly started to be flagged, even without relevant code changes

seems like a "heisen-bug" at first glance

I am now getting this too: https://github.com/pantsbuild/pants/actions/runs/12171755575/job/33949786319?pr=21695

The mutability is in TargetAdaptor, AddressMap, and SyntheticAddressMap.

This has led me down a rabbit hole of cascading mutability bugs…

We've probably been fortunate that Pants has been working still even with these mutability bugs.

As far as I can tell we have been fortunate for two reasons: 1) We (mostly?) don’t use Collection subclasses as keys in dicts, so we don’t get pathological hash buckets. We still need them to be hashable because FrozenDict requires its values to be hashable, for reasons I am not totally clear on. 2) We’re relying on mutability to support the

_extend_synthetic

functionality, but it looks like we don’t actually use that anywhere. From inspection, that functionality appears to be completely broken, but the tests aren’t substantial enough to expose this, and since we don’t actually use it in non-test code, and it’s not documented anywhere, we’re getting away with it. I need to dig deeper into this, and write a test to expose the problem I believe I am seeing. But this is not just a performance issue but a correctness one.