string-or-bytes-too-long (PYI053) flags and removes long strings in Literal, producing invalid type

[Avasam]

In the example reproduction below,

• List of keywords you searched for before creating this issue. Write them down here so that others can find this issue more easily and help provide feedback.
  string-or-bytes-too-long PYI053

• A minimal code snippet that reproduces the bug.

def foo(bar: typing.Literal["a", "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"]):...

becomes

import typing

def foo(bar: typing.Literal["a", ...]):...

Not only is this an invalid type form (https://github.com/microsoft/pyright/blob/main/docs/configuration.md#reportInvalidTypeForm) but even if autofixed to Literal["a"] that would fundamentally change the parameter's type.

I assume this is an oversight as it's meant to replace long default values and is not flagged by flake8-pyi .

• The command you invoked (e.g., ruff /path/to/file.py --fix), ideally including the --isolated flag.
  ruff check --isolated --select=PYI053

• The current Ruff settings (any relevant sections from your pyproject.toml).
  not relevant

• The current Ruff version (ruff --version).
  ruff 0.6.1

[AlexWaygood]

is not flagged by flake8-pyi

I can actually repro this bug with flake8-pyi as well on this snippet in a .pyi file:

from typing import Literal

x: Literal["fooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooO"]

We disable the flake8-pyi rule at typeshed for our generated stubs; maybe that's what was hiding the bug? https://github.com/python/typeshed/blob/a7ff1be4394cef089dd62090a2e2ad10cc77cfe6/.flake8#L16

[AlexWaygood]

I merged #13002, but then immediately afterwards realised that it doesn't deal properly with type aliases. We still get two false positives on this snippet, even after #13002:

from typing import TypeAlias, Literal, Annotated

x: TypeAlias = Literal["fooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooO"]
y: TypeAlias = Annotated[int, "metadataaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"]

@dylwil3, don't suppose you'd be interested in a followup PR, would you? :-) The alternative approach I suggested in #13002 (comment) might work better here after all:

• we'd want to add tracking in the semantic model for when we're entering Annotated slices, the same as we do when we're entering Literal slices
• we'd want to skip PYI053 when we're inside any Annotated or Literal slice, not just when we're inside annotations.

[dylwil3]

Sounds good I'll give it a try!

[dylwil3]

@AlexWaygood Can you not do

def f(x: int) --> "AnnotationsForClassesWithVeryLongNamesInQuotesAsReturnTypes"

?

[AlexWaygood]

@AlexWaygood Can you not do

def f(x: int) --> "AnnotationsForClassesWithVeryLongNamesInQuotesAsReturnTypes"

?

Well, there's no reason to ever use quoted annotations in stubs, since forward references can be used even with __future__ annotations enabled. We have another rule (PYI020) specifically for flagging any quoted annotation in a stub... but I suppose if you are going to use quoted annotations in stubs for some reason, and have PYI020 switched off for some reason, then maybe it makes sense for this rule to ignore veryyyyyyyyyyyyyyyyyyyy long quoted annotations.

In case it's not obvious, I changed my mind about this while writing this response out 😆 So it's definitely worth adding a comment about this to the code for the rule!

[dylwil3]

Hmm... for some reason in_typing_literal has no effect - it doesn't trip for either the argument annotation or typevar. Maybe I've done something wrong. I'll try to investigate later today.

[AlexWaygood]

Hmm... for some reason in_typing_literal has no effect - it doesn't trip for either the argument annotation or typevar. Maybe I've done something wrong. I'll try to investigate later today.

That is odd. I can't immediately see why that would be... don't worry too much about fixing it if you can't figure it out; feel free to ping me if you're stuck!

[Avasam]

We disable the flake8-pyi rule at typeshed for our generated stubs; maybe that's what was hiding the bug?

Yep, you're right ! That's exactly what happened.

[dylwil3]

Sorry for the delay - I was derailed by my baby learning to crawl all of a sudden 🚀 🚀 🚀

Hmm... for some reason in_typing_literal has no effect - it doesn't trip for either the argument annotation or typevar. Maybe I've done something wrong. I'll try to investigate later today.

That is odd. I can't immediately see why that would be... don't worry too much about fixing it if you can't figure it out; feel free to ping me if you're stuck!

Mystery solved - I had an invalid .pyi file since I did not actually import Literal or Annotated 🤦