Summary
--
This is the last main difference between the `OldDiagnostic` and
`Message`
types, so attaching a `SourceFile` to `OldDiagnostic` should make
combining the
two types almost trivial.
Initially I updated the remaining rules without access to a `Checker` to
take a
`&SourceFile` directly, but after Micha's suggestion in
https://github.com/astral-sh/ruff/pull/18356#discussion_r2113281552, I
updated all of these calls to take a
`LintContext` instead. This new type is a thin wrapper around a
`RefCell<Vec<OldDiagnostic>>`
and a `SourceFile` and now has the `report_diagnostic` method returning
a `DiagnosticGuard` instead of `Checker`.
This allows the same `Drop`-based implementation to be used in cases
without a `Checker` and also avoids a lot of intermediate allocations of
`Vec<OldDiagnostic>`s.
`Checker` now also contains a `LintContext`, which it defers to for its
`report_diagnostic` methods, which I preserved for convenience.
Test Plan
--
Existing tests
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
<!-- What's the purpose of the change? What does it do, and why? -->
Follow up on https://github.com/astral-sh/ruff/pull/18093 and apply it
to AIR311
---
Rules fixed
* `airflow.models.datasets.expand_alias_to_datasets` →
`airflow.models.asset.expand_alias_to_assets`
* `airflow.models.baseoperatorlink.BaseOperatorLink` →
`airflow.sdk.BaseOperatorLink`
## Test Plan
<!-- How was it tested? -->
The existing test fixtures have been updated
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
<!-- What's the purpose of the change? What does it do, and why? -->
Follow up on https://github.com/astral-sh/ruff/pull/18093 and apply it
to AIR301
## Test Plan
<!-- How was it tested? -->
The existing test fixtures have been updated
## Summary
- Convert tests demonstrating our resilience to malformed/absent
`version` fields in `pyvenf.cfg` files to mdtests. Also make them more
expansive.
- Convert the regression test I added in
https://github.com/astral-sh/ruff/pull/18157 to an mdtest
- Add comments next to unit tests that cannot be converted to mdtests
(but where it's not obvious why they can't) so I don't have to do this
exercise again 😄
- In `site_packages.rs`, factor out the logic for figuring out where we
expect the system-installation `site-packages` to be. Currently we have
the same logic twice.
## Test Plan
`cargo test -p ty_python_semantic`
## Summary
This change was based on a mis-reading of a comment in typeshed, and a
wrong assumption about what was causing a test failure in a prior PR.
Reverting it doesn't cause any tests to fail.
## Test Plan
Existing tests.
## Summary
Resolves [#513](https://github.com/astral-sh/ty/issues/513).
Callable types are now considered to be disjoint from nominal instance
types where:
* The class is `@final`, and
* Its `__call__` either does not exist or is not assignable to `(...) ->
Unknown`.
## Test Plan
Markdown tests.
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
## Summary
Partially implement https://github.com/astral-sh/ty/issues/538,
```py
from pathlib import Path
def setup_test_project(registry_name: str, registry_url: str, project_dir: str) -> Path:
pyproject_file = Path(project_dir) / "pyproject.toml"
pyproject_file.write_text("...", encoding="utf-8")
```
As no return statement is defined in the function `setup_test_project`
with annotated return type `Path`, we provide the following diagnosis :
- error[invalid-return-type]: Function **always** implicitly returns
`None`, which is not assignable to return type `Path`
with a subdiagnostic :
- note: Consider changing your return annotation to `-> None` or adding a `return` statement
## Test Plan
mdtests with snapshots to capture the subdiagnostic. I have to mention
that existing snapshots were modified since they now fall in this
category.
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
<!-- What's the purpose of the change? What does it do, and why? -->
Add utility functions `generate_import_edit` and
`generate_remove_and_runtime_import_edit` to generate the fix needed for
the airflow rules.
1. `generate_import_edit` is for the cases where the member name has
changed. (e.g., `airflow.datasts.Dataset` to `airflow.sdk.Asset`) It's
just extracted from the original logic
2. `generate_remove_and_runtime_import_edit` is for cases where the
member name has not changed. (e.g.,
`airflow.operators.pig_operator.PigOperator` to
`airflow.providers.apache.pig.hooks.pig.PigCliHook`) This is newly
introduced. As it introduced runtime import, I mark it as an unsafe fix.
Under the hook, it tried to find the original import statement, remove
it, and add a new import fix
---
* rules fix
* `airflow.sensors.external_task_sensor.ExternalTaskSensorLink` →
`airflow.providers.standard.sensors.external_task.ExternalDagLink`
## Test Plan
<!-- How was it tested? -->
The existing test fixtures have been updated
Summary
--
It's a bit late in the refactoring process, but I think there are still
a couple of PRs left before getting rid of this type entirely, so I
thought it would still be worth doing.
This PR is just a quick rename with no other changes.
Test Plan
--
Existing tests
## Summary
Adds coverage of using set(...) in addition to `{...} in
SingleItemMembershipTest.
Fixes#15792
(and replaces the old PR #15793)
<!-- What's the purpose of the change? What does it do, and why? -->
## Test Plan
Updated unit test and snapshot.
Steps to reproduce are in the issue linked above.
<!-- How was it tested? -->
Previously, completions were based on just returning every identifier
parsed in the current Python file. In this commit, we change it to
identify an expression under the cursor and then return all symbols
available to the scope containing that expression.
This is still returning too much, and also, in some cases, not enough.
Namely, it doesn't really take the specific context into account other
than scope. But this does improve on the status quo. For example:
def foo(): ...
def bar():
def fast(): ...
def foofoo(): ...
f<CURSOR>
When asking for completions here, the LSP will no longer include `fast`
as a possible completion in this context.
Ref https://github.com/astral-sh/ty/issues/86
This is analogous to the existing `Tokens::after` method. Its
implementation is almost identical.
We plan to use this for looking at the tokens immediately before the
cursor when fetching completions.
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
Fixes#18231
<!-- What's the purpose of the change? What does it do, and why? -->
## Test Plan
Snapshot tests
<!-- How was it tested? -->
## Summary
Implements `use-maxsplit-arg` (`PLC0207`)
https://pylint.readthedocs.io/en/latest/user_guide/messages/convention/use-maxsplit-arg.html
> Emitted when accessing only the first or last element of str.split().
The first and last element can be accessed by using str.split(sep,
maxsplit=1)[0] or str.rsplit(sep, maxsplit=1)[-1] instead.
This is part of https://github.com/astral-sh/ruff/issues/970
## Test Plan
`cargo test`
Additionally compared Ruff output to Pylint:
```
pylint --disable=all --enable=use-maxsplit-arg crates/ruff_linter/resources/test/fixtures/pylint/missing_maxsplit_arg.py
cargo run -p ruff -- check crates/ruff_linter/resources/test/fixtures/pylint/missing_maxsplit_arg.py --no-cache --select PLC0207
```
---------
Co-authored-by: Brent Westbrook <brentrwestbrook@gmail.com>
## Summary
Allow classes with `__init__` to be subtypes of `Callable`
Fixes https://github.com/astral-sh/ty/issues/358
## Test Plan
Update is_subtype_of.md
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
## Summary
This PR add the `fix safety` section for rule `B006` in
`mutable_argument_default.rs` for #15584
When applying this rule for fixes, certain changes may alter the
original logical behavior. For example:
before:
```python
def cache(x, storage=[]):
storage.append(x)
return storage
print(cache(1)) # [1]
print(cache(2)) # [1, 2]
```
after:
```python
def cache(x, storage=[]):
storage.append(x)
return storage
print(cache(1)) # [1]
print(cache(2)) # [2]
```
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
Fixes#18353
<!-- What's the purpose of the change? What does it do, and why? -->
## Test Plan
<!-- How was it tested? -->
Snapshot tests
There were many fields in `Signature` and friends that really had more
to do with how a signature was being _used_ — how it was looked up,
details about an individual call site, etc. Those fields more properly
belong in `Bindings` and friends.
This is a pure refactoring, and should not affect any tests or ecosystem
projects.
I started on this journey in support of
https://github.com/astral-sh/ty/issues/462. It seemed worth pulling out
as a separate PR.
One major concrete benefit of this refactoring is that we can now use
`CallableSignature` directly in `CallableType`. (We can't use
`CallableSignature` directly in that `Type` variant because signatures
are not currently interned.)
Summary
--
This PR adds a `DiagnosticGuard` type to ruff that is adapted from the
`DiagnosticGuard` and `LintDiagnosticGuard` types from ty. This guard is
returned by `Checker::report_diagnostic` and derefs to a
`ruff_diagnostics::Diagnostic` (`OldDiagnostic`), allowing methods like
`OldDiagnostic::set_fix` to be called on the result. On `Drop` the
`DiagnosticGuard` pushes its contained `OldDiagnostic` to the `Checker`.
The main motivation for this is to make a following PR adding a
`SourceFile` to each diagnostic easier. For every rule where a `Checker`
is available, this will now only require modifying
`Checker::report_diagnostic` rather than all the rules.
In the few cases where we need to create a diagnostic before we know if
we actually want to emit it, there is a `DiagnosticGuard::defuse`
method, which consumes the guard without emitting the diagnostic. I was
able to restructure about half of the rules that naively called this to
avoid calling it, but a handful of rules still need it.
One of the fairly common patterns where `defuse` was needed initially
was something like
```rust
let diagnostic = Diagnostic::new(DiagnosticKind, range);
if !checker.enabled(diagnostic.rule()) {
return;
}
```
So I also added a `Checker::checked_report_diagnostic` method that
handles this check internally. That helped to avoid some additional
`defuse` calls. The name is a bit repetitive, so I'm definitely open to
suggestions there. I included a warning against using it in the docs
since, as we've seen, the conversion from a diagnostic to a rule is
actually pretty expensive.
Test Plan
--
Existing tests
## Summary
We create `Callable` types for synthesized functions like the `__init__`
method of a dataclass. These generated functions are real functions
though, with descriptor-like behavior. That is, they can bind `self`
when accessed on an instance. This was modeled incorrectly so far.
## Test Plan
Updated tests
## Summary
I don't think we're ever going to add any `KnownInstanceType` variants
that evaluate to `False` in a boolean context; the
`KnownInstanceType::bool()` method just seems like unnecessary
complexity.
## Test Plan
`cargo test -p ty_python_semantic`
# Summary
Adds a subdiagnostic hint in the following scenario where a
synchronous `with` is used with an async context manager:
```py
class Manager:
async def __aenter__(self): ...
async def __aexit__(self, *args): ...
# error: [invalid-context-manager] "Object of type `Manager` cannot be used with `with` because it does not implement `__enter__` and `__exit__`"
# note: Objects of type `Manager` *can* be used as async context managers
# note: Consider using `async with` here
with Manager():
...
```
closes https://github.com/astral-sh/ty/issues/508
## Test Plan
New MD snapshot tests
---------
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
fixesastral-sh/ty#366
## Test Plan
* Added panic corpus regression tests
* I also wrote a hover regression test (see below), but decided not to
include it. The corpus tests are much more "effective" at finding these
types of errors, since they exhaustively check all expressions for
types.
<details>
```rs
#[test]
fn hover_regression_test_366() {
let test = cursor_test(
r#"
from ty_extensions import Intersection
class A: ...
class B: ...
def _(x: Intersection[A,<CURSOR> B]):
pass
"#,
);
assert_snapshot!(test.hover(), @r"
A & B
---------------------------------------------
```text
A & B
```
---------------------------------------------
info[hover]: Hovered content is
--> main.py:7:31
|
5 | class B: ...
6 |
7 | def _(x: Intersection[A, B]):
| ^^-^
| | |
| | Cursor offset
| source
8 | pass
|
");
}
```
</details>
## Summary
The previous `try_call_dunder_with_policy` API was a bit of a footgun
since you needed to pass `NO_INSTANCE_FALLBACK` in *addition* to other
policies that you wanted for the member lookup. Implicit calls to dunder
methods never access instance members though, so we can do this
implicitly in `try_call_dunder_with_policy`.
No functional changes.
## Summary
`Type::member_lookup_with_policy` now falls back to calling
`__getattribute__` when a member cannot be found as a second fallback
after `__getattr__`.
closes https://github.com/astral-sh/ty/issues/441
## Test Plan
Added markdown tests.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
This should address a problem that came up while working on
https://github.com/astral-sh/ruff/pull/18280. When looking up an
attribute (typically a dunder method) with the `MRO_NO_OBJECT_FALLBACK`
policy, the attribute is first looked up on the meta type. If the meta
type happens to be `type`, we go through the following branch in
`find_name_in_mro_with_policy`:
97ff015c88/crates/ty_python_semantic/src/types.rs (L2565-L2573)
The problem is that we now look up the attribute on `object` *directly*
(instead of just having `object` in the MRO). In this case,
`MRO_NO_OBJECT_FALLBACK` has no effect in `class_member_from_mro`:
c3feb8ce27/crates/ty_python_semantic/src/types/class.rs (L1081-L1082)
So instead, we need to explicitly respect the `MRO_NO_OBJECT_FALLBACK`
policy here by returning `Symbol::Unbound`.
## Test Plan
Added new Markdown tests that explain the ecosystem changes that we
observe.
## Summary
Fix a bug that involved writes to attributes on union/intersection types
that included modules as elements.
This is a prerequisite to avoid some ecosystem false positives in
https://github.com/astral-sh/ruff/pull/18312
## Test Plan
Added regression test
## Summary
This PR moves the diagnostics API for the language server out from the
request handler module to the diagnostics API module.
This is in preparation to add support for publishing diagnostics.
## Summary
Resolves https://github.com/astral-sh/ty/issues/485.
`infer_binary_intersection_type_comparison()` now checks for all
positive members before concluding that an operation is unsupported for
a given intersection type.
## Test Plan
Markdown tests.
---------
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
This is a practice I followed on previous projects. Should hopefully
further help developers who want to update the documentation.
The big downside is that it's annoying to see this *as a user of the
documentation* if you don't open the Markdown file in the browser. But
I'd argue that those files don't really follow the original Markdown
spirit anyway with all the inline HTML.
## Summary
This is something I wrote a few months ago, and continued to update from
time to time. It was mostly written for my own education. I found a few
bugs while writing it at the time (there are still one or two TODOs in
the test assertions that are probably bugs). Our other tests are fairly
comprehensive, but they are usually structured around a certain
functionality or operation (subtyping, assignability, narrowing). The
idea here was to focus on individual *types and their properties*.
closes#197 (added `JustFloat` and `JustComplex` to `ty_extensions`).
## Summary
Fix remaining `knot.toml` reference and replace it with `ty.toml`. This
change was probably still in flight while we renamed things.
## Test Plan
Added a second assertion which ensures that the config file has any
effect.
## Summary
It doesn't seem to be necessary for our generics implementation to carry
the `GenericContext` in the `ClassBase` variants. Removing it simplifies
the code, fixes many TODOs about `Generic` or `Protocol` appearing
multiple times in MROs when each should only appear at most once, and
allows us to more accurately detect runtime errors that occur due to
`Generic` or `Protocol` appearing multiple times in a class's bases.
In order to remove the `GenericContext` from the `ClassBase` variant, it
turns out to be necessary to emulate
`typing._GenericAlias.__mro_entries__`, or we end up with a large number
of false-positive `inconsistent-mro` errors. This PR therefore also does
that.
Lastly, this PR fixes the inferred MROs of PEP-695 generic classes,
which implicitly inherit from `Generic` even if they have no explicit
bases.
## Test Plan
mdtests
## Summary
Fix some issues with subtying/assignability for instances vs callables.
We need to look up dunders on the class, not the instance, and we should
limit our logic here to delegating to the type of `__call__`, so it
doesn't get out of sync with the calls we allow.
Also, we were just entirely missing assignability handling for
`__call__` implemented as anything other than a normal bound method
(though we had it for subtyping.)
A first step towards considering what else we want to change in
https://github.com/astral-sh/ty/issues/491
## Test Plan
mdtests
---------
Co-authored-by: med <medioqrity@gmail.com>
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Division works differently in Python than in Rust. If the result is
negative and there is a remainder, the division rounds down (instead of
towards zero). The remainder needs to be adjusted to compensate so that
`(lhs // rhs) * rhs + (lhs % rhs) == lhs`.
Fixesastral-sh/ty#481.
## Summary
https://github.com/astral-sh/ty/issues/111
This PR adds support for `frozen` dataclasses. It will emit a diagnostic
with a similar message to mypy
Note: This does not include emitting a diagnostic if `__setattr__` or
`__delattr__` are defined on the object as per the
[spec](https://docs.python.org/3/library/dataclasses.html#module-contents)
## Test Plan
mdtest
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Co-authored-by: Carl Meyer <carl@astral.sh>
## Summary
Make sure that the following definitions all lead to the same outcome
(bug originally noticed by @AlexWaygood)
```py
from typing import ClassVar
class Descriptor:
def __get__(self, instance, owner) -> int:
return 42
class C:
a: ClassVar[Descriptor]
b: Descriptor = Descriptor()
c: ClassVar[Descriptor] = Descriptor()
reveal_type(C().a) # revealed: int (previously: int | Descriptor)
reveal_type(C().b) # revealed: int
reveal_type(C().c) # revealed: int
```
## Test Plan
New Markdown tests
## Summary
I think `division-by-zero` is a low-value diagnostic in general; most
real division-by-zero errors (especially those that are less obvious to
the human eye) will occur on values typed as `int`, in which case we
don't issue the diagnostic anyway. Mypy and pyright do not emit this
diagnostic.
Currently the diagnostic is prone to false positives because a) we do
not silence it in unreachable code, and b) we do not implement narrowing
of literals from inequality checks. We will probably fix (a) regardless,
but (b) is low priority apart from division-by-zero.
I think we have many more important things to do and should not allow
false positives on a low-value diagnostic to be a distraction. Not
opposed to re-enabling this diagnostic in future when we can prioritize
reducing its false positives.
References https://github.com/astral-sh/ty/issues/443
## Test Plan
Existing tests.
## Summary
This PR updates the language server to avoid panicking when there are
multiple workspace folders passed during initialization. The server
currently picks up the first workspace folder and provides a warning and
a log message.
## Test Plan
<img width="1724" alt="Screenshot 2025-05-17 at 11 43 09"
src="https://github.com/user-attachments/assets/1a7ddbc3-198d-4191-a28f-9b69321e8f99"
/>
## Summary
Resolves [#461](https://github.com/astral-sh/ty/issues/461).
ty was hardcoded to infer `BytesLiteral` types for integer indexing into
`BytesLiteral`. It will now infer `IntLiteral` types instead.
## Test Plan
Markdown tests.
Summary
--
I thought that emitting multiple diagnostics at once would be difficult
to port to a diagnostic construction model closer to ty's
`InferContext::report_lint`, so as a first step toward that, this PR
removes `Checker::report_diagnostics`.
In many cases I was able to do some related refactoring to avoid
allocating a `Vec<Diagnostic>` at all, often by adding a `Checker` field
to a `Visitor` or by passing a `Checker` instead of a `&mut
Vec<Diagnostic>`.
In other cases, I had to fall back on something like
```rust
for diagnostic in diagnostics {
checker.report_diagnostic(diagnostic);
}
```
which I guess is a bit worse than the `extend` call in
`report_diagnostics`, but hopefully it won't make too much of a
difference.
I'm still not quite sure what to do with the remaining loop cases. The
two main use cases for collecting a sequence of diagnostics before
emitting any of them are:
1. Applying a single `Fix` to a group of diagnostics
2. Avoiding an earlier diagnostic if something goes wrong later
I was hoping we could get away with just a `DiagnosticGuard` that
reported a `Diagnostic` on drop, but I guess we will still need a
`DiagnosticGuardBuilder` that can be collected in these cases and
produce a `DiagnosticGuard` once we know we actually want the
diagnostics.
Test Plan
--
Existing tests
Closes https://github.com/astral-sh/ty/issues/453.
## Summary
Add an additional info diagnostic to `unresolved-import` check to hint
to users that they should make sure their Python environment is properly
configured for ty, linking them to the corresponding doc. This
diagnostic is only shown when an import is not relative, e.g., `import
maturin` not `import .maturin`.
## Test Plan
Updated snapshots with new info message and reran tests.
The PR add the `fix safety` section for rule `SIM110` (#15584 )
### Unsafe Fix Example
```python
def predicate(item):
global called
called += 1
if called == 1:
# after first call we change the method
def new_predicate(_): return False
globals()['predicate'] = new_predicate
return True
def foo():
for item in range(10):
if predicate(item):
return True
return False
def foo_gen():
return any(predicate(item) for item in range(10))
called = 0
print(foo()) # true – returns immediately on first call
called = 0
print(foo_gen()) # false – second call uses new `predicate`
```
### Note
I notice that
[here](46be305ad2/crates/ruff_linter/src/rules/flake8_simplify/rules/reimplemented_builtin.rs (L60))
we have two rules, `SIM110` & `SIM111`. The second one seems not anymore
active. Should I delete `SIM111`?
This implements the stopgap approach described in
https://github.com/astral-sh/ty/issues/336#issuecomment-2880532213 for
handling literal types in generic class specializations.
With this approach, we will promote any literal to its instance type,
but _only_ when inferring a generic class specialization from a
constructor call:
```py
class C[T]:
def __init__(self, x: T) -> None: ...
reveal_type(C("string")) # revealed: C[str]
```
If you specialize the class explicitly, we still use whatever type you
provide, even if it's a literal:
```py
from typing import Literal
reveal_type(C[Literal[5]](5)) # revealed: C[Literal[5]]
```
And this doesn't apply at all to generic functions:
```py
def f[T](x: T) -> T:
return x
reveal_type(f(5)) # revealed: Literal[5]
```
---
As part of making this happen, we also generalize the `TypeMapping`
machinery. This provides a way to apply a function to type, returning a
new type. Complicating matters is that for function literals, we have to
apply the mapping lazily, since the function's signature is not created
until (and if) someone calls its `signature` method. That means we have
to stash away the mappings that we want to apply to the signatures
parameter/return annotations once we do create it. This requires some
minor `Cow` shenanigans to continue working for partial specializations.
This is a follow-on to #18155. For the example raised in
https://github.com/astral-sh/ty/issues/370:
```py
import tempfile
with tempfile.TemporaryDirectory() as tmp: ...
```
the new logic would notice that both overloads of `TemporaryDirectory`
match, and combine their specializations, resulting in an inferred type
of `str | bytes`.
This PR updates the logic to match our other handling of other calls,
where we only keep the _first_ matching overload. The result for this
example then becomes `str`, matching the runtime behavior. (We still do
not implement the full [overload resolution
algorithm](https://typing.python.org/en/latest/spec/overload.html#overload-call-evaluation)
from the spec.)
## Summary
Add a new diagnostic hint if you try to use PEP 604 `X | Y` union syntax
in a non-type-expression before 3.10.
closes https://github.com/astral-sh/ty/issues/437
## Test Plan
New snapshot test
Brent Westbrook
Micha Reiser
Wei Lee
Alex Waygood
Carl Meyer
InSync
lipefree
Marcus Näslund
Andrew Gallant
Victor Hugo Gomes
vjurczenia
Matthew Mckee
Hans
Douglas Creager
Viktor Merkurev
David Peter
Dhruv Manilawala
Max Mynter
chiri
हिमांशु
justin
Maddy Guthridge
Felix Scherz
Vasanth
David Sherret
Jo
Charlie Marsh
Dylan
Brandt Bucher
Sumana Harihareswara
Denys Kyslytsyn
Ramil Aleskerov
Vasco Schiavo
Adam Aaronson
Emily B. Zhang