[ty] Use RHS inferred type for bare Final symbols
#19142
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sharkdp
commented
Jul 4, 2025
|
|
||
| `mod.py`: | ||
|
|
||
| ```py | ||
| from typing import Final, Annotated | ||
|
|
||
| FINAL_A: int = 1 |
There was a problem hiding this comment.
Pretty sure I intended to write FINAL_A: Final[int] = 1 here originally :-)
|
I think our behaviour is correct here. Even if we put aside the issue of variance, the module still shouldn't satisfy For the module to satisfy the protocol, either the members on the module would need to be made mutable (and declared as - Z_FULL_FLUSH: Final = 3
- Z_SYNC_FLUSH: Final = 2
+ Z_FULL_FLUSH: int = 3
+ Z_SYNC_FLUSH: int = 2or the protocol would need to be changed so that it uses immutable, covariant members rather than mutable, invariant members. The only way to do that on a protocol currently (at least until https://peps.python.org/pep-0767/ is approved) is to use class ZLibBackendProtocol(Protocol):
- Z_FULL_FLUSH: int
- Z_SYNC_FLUSH: int
+ @property
+ def Z_FULL_FLUSH(self) -> int: ...
+ @property
+ def Z_SYNC_FLUSH(self) -> int: ... |
yeah, that does sound like a true positive. Possibly a typeshed bug? |
This comment was marked as resolved.
This comment was marked as resolved.
sharkdp
force-pushed
the
david/typing-final-type-inference
branch
from
7d591b2 to
9e1657d
Compare
sharkdp
changed the title
Final symbolsFinal symbols
sharkdp
force-pushed
the
david/typing-final-type-inference
branch
from
9e1657d to
bf1ddb2
Compare
AlexWaygood
approved these changes
Jul 7, 2025
Comment on lines
+527
to
+540
| /// Returns `Some(…)` if the place is qualified with `typing.Final` without a specified type. | ||
| pub(crate) fn is_bare_final(&self) -> Option<TypeQualifiers> { | ||
| match self { | ||
| PlaceAndQualifiers { place, qualifiers } | ||
| if (qualifiers.contains(TypeQualifiers::FINAL) | ||
| && place | ||
| .ignore_possibly_unbound() | ||
| .is_some_and(|ty| ty.is_unknown())) => | ||
| { | ||
| Some(*qualifiers) | ||
| } | ||
| _ => None, | ||
| } | ||
| } |
There was a problem hiding this comment.
Does this mean that we would consider x to be "qualified with bare Final" here?
from ty_extensions import Unknown
from typing import Final
x: Final[Unknown]I think that's okay if so; people just shouldn't do that ;)
There was a problem hiding this comment.
Yes, well spotted 😄. I agree that it's not as clean as it could be. The alternative is much more complicated as it requires us to return Option<Type> and TypeQualifiers instead of just Type and TypeQualifiers from infer_annotation_expression and related methods (such that we could return None for a bare Final or ClassVar), which has a lot of downstream consequences.
UnboundVariable
pushed a commit
to UnboundVariable/ruff
that referenced
this pull request
Jul 7, 2025
…c_tokens * 'main' of https://github.com/astral-sh/ruff: (27 commits) [ty] First cut at semantic token provider (astral-sh#19108) [`flake8-simplify`] Make example error out-of-the-box (`SIM116`) (astral-sh#19111) [`flake8-use-pathlib`] Make example error out-of-the-box (`PTH210`) (astral-sh#19189) [`flake8-use-pathlib`] Add autofixes for `PTH203`, `PTH204`, `PTH205` (astral-sh#18922) [`flake8-type-checking`] Fix syntax error introduced by fix (`TC008`) (astral-sh#19150) [`flake8-pyi`] Make example error out-of-the-box (`PYI007`, `PYI008`) (astral-sh#19103) Update Rust crate indicatif to 0.18.0 (astral-sh#19165) [ty] Add separate CI job for memory usage stats (astral-sh#19134) [ty] Add documentation for server traits (astral-sh#19137) Rename to `SessionSnapshot`, move unwind assertion closer (astral-sh#19177) [`flake8-type-checking`] Make example error out-of-the-box (`TC001`) (astral-sh#19151) [ty] Bare `ClassVar` annotations (astral-sh#15768) [ty] Re-enable multithreaded pydantic benchmark (astral-sh#19176) [ty] Implement equivalence for protocols with method members (astral-sh#18659) [ty] Use RHS inferred type for bare `Final` symbols (astral-sh#19142) [ty] Support declaration-only attributes (astral-sh#19048) [ty] Sync vendored typeshed stubs (astral-sh#19174) Update dependency pyodide to ^0.28.0 (astral-sh#19164) Update NPM Development dependencies (astral-sh#19170) Update taiki-e/install-action action to v2.56.7 (astral-sh#19169) ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Infer the type of symbols with a
Finalqualifier as their right-hand-side inferred type:Part of astral-sh/ty#158
Ecosystem analysis
aiohttp
aiohttp (https://github.com/aio-libs/aiohttp) + error[invalid-argument-type] aiohttp/compression_utils.py:131:54: Argument to bound method `__init__` is incorrect: Expected `ZLibBackendProtocol`, found `<module 'zlib'>`This code creates a protocol that looks like
It then tries to assign the module literal
zlibto that protocol. Howefer, in typeshed, thesezlibmembers are annotated like this:With the proposed change here, we now infer these as
Literal[3]/Literal[2]. Since protocol members have to be assignable both ways (invariance), we do not considerzlibassignable to this protocol anymore.That seems rather unfortunate. Not sure who is to blame here? That
ZLibBackendProtocolprotocol should probably not annotate the members withint, given thattypesheddoesn't use an explicit annotation here either? But what should they do instead? Annotate those fields withAny?Or is it another case where we should consider literal-widening?
FYI @AlexWaygood
cloud-init
New false positives on expressions like
oct(os.stat(legacy_script_f)[stat.ST_MODE]). We now correctly inferstat.ST_MODEasLiteral[1], because in typeshed, it is annotated asST_MODE: Final = 0.os.statreturns astat_resultwhich is a tuple subclass. Accessing it at index 0 should return anint, but we currently returnint | float, presumably due to missing support for tuple subclasses (FYI @AlexWaygood):In terms of
typing.Final, things are working as expected here.pywin-32
Many new false positives similar to:
pywin32 (https://github.com/mhammond/pywin32) + error[invalid-argument-type] Pythonwin/pywin/docking/DockingBar.py:288:55: Argument to function `LoadCursor` is incorrect: Expected `PyResourceId`, found `Literal[32645]`The line in question calls
win32api.LoadCursor(0, win32con.IDC_ARROW). Thewin32con.IDC_ARROWsymbol is annotated asIDC_ARROW: Final = 32512in typeshed, butLoadCursorexpects aPyResourceId, which is an empty class. So.. this seems like a true positive to me, unless that typeshed annotation ofIDC_ARROWis meant to imply that the type should beUnknown/Any?streamlit
streamlit (https://github.com/streamlit/streamlit) + error[invalid-argument-type] lib/streamlit/string_util.py:163:37: Argument to bound method `translate` is incorrect: Expected `bytes`, found `bytearray`This looks like a true positive? The code calls
inp.translate(None, TEXTCHARS).inpisbytes, andTEXTCHARSis:We now infer this as(Edit: this is now fixed in typeshed)bytearray, butbytes.translateexpectsbytesfor itsdeleteparameter. This seems to work at runtime, so maybe the typeshed annotation is wrong?rotki
+ error[invalid-return-type] rotkehlchen/chain/ethereum/modules/yearn/decoder.py:412:13: Return type does not match returned value: expected `dict[Unknown, str]`, found `dict[Unknown, Literal["yearn-v1", "yearn-v2"]]`The code in question looks like
where
CPT_BEEFY_FINANCE: Final = 'beefy_finance'. We previously inferred the value type of the returneddictasUnknown, and now we infer it asLiteral["beefy_finance"], which does not match the annotated return type becausedict` is invariant in the value type.+ error[invalid-argument-type] rotkehlchen/tests/unit/decoders/test_curve.py:249:9: Argument is incorrect: Expected `int`, found `FVal`There are true positives that were previously silenced through the
Unknown.Test Plan
New Markdown tests