[WIP] Fix issue #4975

[theodoretliu]

Addresses issue #4975. See three test cases, first is passing, unclear on what the behavior should be on second two

[theodoretliu]

@ilevkivskyi @Michael0x2a appears to be working if I hardcode a file test.py and run mypy on it

from typing import Tuple

x: Tuple[str, float]
y: Tuple[int, ...]

lst = [x, y]
reveal_type(lst[0])

produces

test.py:7: note: Revealed type is 'builtins.tuple*[builtins.object]'

with no errors.

However, running the same in the test suite seems to cause issues.

Similar with the following

from typing import Tuple

x: Tuple[int, str]
y: Tuple[int, ...]

lst = [x, y]
reveal_type(lst[0])

produces

test.py:7: note: Revealed type is 'builtins.tuple*[builtins.object]'

[ilevkivskyi]

Thanks for PR!

The tests failed because the list.pyi test fixture defines tuple as invariant, you can update it. Alternatively you can use tuple.pyi fixture.

I don't think this covers all the cases in the original issue and the issues that were closed as duplicates. In particular, these test cases are missing:

• Join two tuples of different length: Tuple[str, str] vs Tuple[str, str, str]
• Ditto with different types
• Joins with an empty tuple (fixed length and variable length)
• It would be great to have a test with ternary expressions instead of lists.

You can browse through linked issues, maybe there are more.

[ilevkivskyi]

Sorry, I will not have time for this, hopefully @JukkaL or @msullivan can help.

[JukkaL]

I can finish the review. The next steps would be fixing the merge conflict and the failing test cases. If you need help with some of these, just let me know.

[intgr]

Doh, sorry :( I was under the impression that nobody was working on this issue (made that assumption when JelleZijlstra commented in #4975). It didn't help that this PR is titled "Initial attempt"

So meanwhile I submitted two pull requests for the two behaviors implemented here: #8333 and #8335.

The good news is that I happened to find a cleaner implementation, my approach can join Tuple[x, x] + List[x] into a Sequence[x] type, which seems to not be the case here.

I now tested my code with the test cases in this PR and my code passes these cases.

What do you think is the best way forward?

[theodoretliu]

@intgr well, you were quick on the uptake, so I don't have a problem if they merge yours instead of mine. In the future, I can title my PR better, but I did reference the issue in this PR and it does show up on the Issue tracker... Nevertheless, taking a quick look at your PR shows that we had approximately the same idea, it's up the maintainers @JukkaL @ilevkivskyi to make the final call.

There was an issue with my PR that was raised regarding types that inherit from Tuple: if we go with yours, we might try to fix that too?

[intgr]

Thanks, for the heads-up, I hadn't considered Tuple subclasses. I'll have to test that behavior.

[JukkaL]

Hmm, I think I prefer #8333 and #8335, since they are more recent and have no conflicts. Supporting tuple subclasses would be a nice addition.

[theodoretliu]

Closing in favor of @intgr work in #8333 and #8335