[ilevkivskyi]

On one hand, this will likely degrade inference quality in some cases, but on the other hand this makes some core numeric libraries like pandas ~2x faster to type-check. Reasons why I think this should rarely affect users:

• This only has effect when there is a redundant union item in a function annotation itself (as opposite to some type that we inferred, which may be redundant quite often).
• This only affects subtyping for generic callables, which is rare in user code (unlike 3rd party libraries, where we don't show errors anyway).

Let's see how the primer looks.

[ilevkivskyi]

Hm, primer output looks good, but performance improvement in primer for numerical libraries is much more modest that I see locally, only ~20% FWIW I think it is still worth it. cc @JukkaL @hauntsaninja

[JukkaL]

Thanks! Type checking some numeric code is definitely quite slow currently, so optimizations like these are valuable.

[JukkaL]

Wrap this around try/finally so that the new value won't stick around?

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

hydra-zen (https://github.com/mit-ll-responsible-ai/hydra-zen)
- src/hydra_zen/typing/_builds_overloads.py:229: error: Overloaded function implementation does not accept all possible parameters of signature 3  [misc]
- src/hydra_zen/typing/_builds_overloads.py:229: error: Overloaded function implementation does not accept all possible parameters of signature 5  [misc]
- src/hydra_zen/typing/_builds_overloads.py:229: error: Overloaded function implementation does not accept all possible parameters of signature 7  [misc]
- src/hydra_zen/typing/_builds_overloads.py:229: error: Overloaded function implementation does not accept all possible parameters of signature 9  [misc]
- src/hydra_zen/typing/_builds_overloads.py:468: error: Overloaded function implementation does not accept all possible parameters of signature 2  [misc]
- src/hydra_zen/typing/_builds_overloads.py:468: error: Overloaded function implementation does not accept all possible parameters of signature 4  [misc]
- src/hydra_zen/typing/_builds_overloads.py:468: error: Overloaded function implementation does not accept all possible parameters of signature 6  [misc]
- src/hydra_zen/typing/_builds_overloads.py:468: error: Overloaded function implementation does not accept all possible parameters of signature 8  [misc]
- src/hydra_zen/typing/_builds_overloads.py:468: error: Overloaded function implementation does not accept all possible parameters of signature 10  [misc]
- src/hydra_zen/typing/_builds_overloads.py:654: error: Overloaded function implementation does not accept all possible parameters of signature 1  [misc]
- src/hydra_zen/typing/_builds_overloads.py:654: error: Overloaded function implementation does not accept all possible parameters of signature 5  [misc]

[ilevkivskyi]

Btw OoC I tried to compare simple mypy -c "import pandas", and it went down from 3.2 sec on v2.3.1 (compiled) to 2.4 sec on current master (compiled), so I think it was definitely worth it after all.