When `a` has type `Any`, after `assert a is None`, the rest of the block is considered dead code

[benkuhn]

Steps to reproduce:

1. Check out this repo
2. Install the packages from requirements.txt
3. Run mypy repro.py from the repo root.

Expected output:

repro.py:4: error: Revealed type is 'Any'
repro.py:6: error: Revealed type is 'builtins.None'

Actual output:

repro.py:4: error: Revealed type is 'Any'

The issue goes away if you disable strict Optional checking.

This is preventing a significant amount of our codebase from being checked right now, so I would appreciate any suggestions for a workaround!

[gvanrossum]

So the crux of the repro is simply (with --strict-optional):

from typing import Any, Optional

def foo(a: Any) -> None:
    reveal_type(a)
    assert a is None
    reveal_type(None)  # This line is not analyzed

Thanks for reporting. This definitely is a bug. It seems related to (but is not a duplicate of) #2776.

[gvanrossum]

Adding a strict-optional label since this is only a problem with strict-optional. @ddfisher any idea?

[benkuhn]

As a workaround/general safety measure, is there a way to force mypy to analyze all "dead" code after asserts? Our codebase has very little actual dead code, and comparatively many places where I'm worried that mypy would erroneously declare the code dead due to an assert like this.

[gvanrossum]

I'm not aware of any. Note that it's not every assert, only very specific asserts. You could also refrain from using strict-optional for now (it's still very much experimental).

[benkuhn]

OK, I couldn't help poking around to see what the issue was, and I think I may have found the right place to fix it--let me know what you think!

[benkuhn]

Any thoughts on the pull request I submitted?

[gvanrossum]

Yes, that I'm way behind on the queue.