TestSlide

Fixes #193. The issue is from 2020 but still reproduces on current (555422d) — repro below. Cause and wrap the decorated generator function with , which copies over verbatim. So for the recorded return type is — what the generator yields — while calling actually returns an . reads that annotation straight off the template, so mocking the function rejects every legitimate return value: There is no return value that could satisfy it — the annotation describes something the caller never receives. Reproduction Exactly the files from #193; fails before this change and passes after. foo.py / test_foo.py Approach Detect the decoration by comparing the template against its : contextlib turns a generator function into one that is no longer a generator function, which is what distinguishes it from in general. The expected type then becomes / . I used the unparameterized ABCs deliberately. The yielded type is produced later by , so it is not present in the value being validated and cannot be checked at this point; claiming would be checking something we cannot see. Happy to change if you would rather keep the parameter for readability. The check is redirected, not disabled — a non-context-manager return value is still rejected. asserts that, so this cannot silently rot into "no validation". Tests Adds a fixture (implements both the sync and async protocols) plus decorated factories to , and three examples to the context. The two positive ones fail without the change with exactly the error from the issue. Verified on Windows / Python 3.13: , , , , , : 2865 successful, 4 failed, 62 skipped** — identical baseline of 4 pre-existing failures before and after; the 3 new successes are mine. unittest suites: 141 tests, only erroring, which is being Unix-only, and reproduces on a clean checkout. back to the same 22 pre-existing findings; clean. I left the pre-existing deviation in (a lambda in ) untouched so the diff stays on topic. 🤖 Generated with Claude Code

Summary: The StrictMock type stubs across strictmock and testslide defined , making it a type alias for . Pyrefly treats bare as non-instantiable, causing ~14,700 errors () across ~3,500 test files in fbcode. Additionally, when the stub was changed to a plain , Pyrefly correctly flagged ~13,500 errors where instances were passed to functions expecting specific types — which is the intended usage of StrictMock as a mock stand-in. The fix changes all four stub files from: to: This makes StrictMock an instantiable class that inherits from Any, so it is both constructible and assignable to any type — matching its real-world usage as a universal mock object. Affected stubs: fbcode/strictmock/strict_mock.pyi fbcode/testslide/testslide/strict_mock.pyi fbcode/python/typeshed_internal/testslide/core/strict_mock.pyi fbcode/python/typeshed_experimental/testslide/core/strict_mock.pyi Differential Revision: D97783042

What: closes #195 Why: feature was requested, and I think it should be implemented How: just a simple warning tracker class in Risks: n/a, backward compatible, non breaking changes Checklist**: [x] Added tests, if you've added code that should be tested [x] Updated the documentation, if you've changed APIs [x] Ensured the test suite passes [x] Made sure your code lints [x] Completed the Contributor License Agreement ("CLA")