The logic for disabling the unused warning on unresolved imports was dependent on the visit order. Moving sustracting unresolved imports to the end of computation fixes the problem.
(cherry picked from commit adeb85e59c17261a5bf9f64dfb4a7836a2403f43)
IJ-CR-153189
GitOrigin-RevId: fac4173a0fac5b29be9301f9d5f04fed08cceff1
Check YieldType of yield expressions in PyTypeCheckerInspection
Check that (Async)Generator is used in (async) function
Check that in 'yield from' sync Generator is used
Convert PyMakeFunctionReturnTypeQuickFix into PsiUpdateModCommandAction
Infer Generator type for lambdas
When getting function type from annotation, do not convert Generator to AsyncGenerator
Introduce GeneratorTypeDescriptor to simplify working with generator annotations
Merge-request: IJ-MR-146521
Merged-by: Aleksandr Govenko <aleksandr.govenko@jetbrains.com>
(cherry picked from commit b3b8182168c5224f0e03f54d443171ccf6ca7b89)
IJ-MR-146521
GitOrigin-RevId: a95670d7e2787015bcf162637ea6d7bfb47a312a
Updated PyFunction to account for implicit 'return None' statements when inferring return statement types.
It affected return type inference of PyFunction.
Fixed a failing test related to formatted strings.
Added a quick fix to make all return statements explicit.
Updated the CFG to include PyPassStatements, enabling detection of exit points in empty functions.
Simplified PyMakeFunctionReturnTypeQuickFix to independently infer function types and handle required imports. Currently, it does not support specifying custom suggested types.
Merge-request: IJ-MR-148719
Merged-by: Aleksandr Govenko <aleksandr.govenko@jetbrains.com>
(cherry picked from commit 9f58961f9eb70e4f9dbba7359f5aafdfd392b7e2)
IJ-MR-148719
GitOrigin-RevId: 68ef5c4a1cc0fcaffd750cc0713250a106136643
__anext__ method is required to return Awaitable. When getting iterated type, unwrap that awaitable
Merge-request: IJ-MR-150432
Merged-by: Aleksandr Govenko <aleksandr.govenko@jetbrains.com>
(cherry picked from commit 97a070ead70c0ec0f2da08000517c592c5805965)
IJ-MR-150432
GitOrigin-RevId: 21dc20b164e6095e56707ea50bc93bf67564e88e
If there is an overload not followed by an implementation, which is
already an error, always resolve to the first overload, regardless
of whether it's a .py file, or a .pyi stub. It allows us to eliminate
the special RatedResolveResult#RATE_LIFTED_PY_FILE_OVERLOAD rate in .py
files, because we no longer need to duplicate the last, closest reachable
overload (normally an implementation should be reachable) with a higher
priority, and then filter it out during overload resolution.
Meanwhile, this filtering out didn't work right before
because some type inference logic, e.g., PyCallExpressionHelper.getCalleeType
used in PyReferenceExpressionImpl.getCallableType bypassed it. It should have
been done at the level of
PyCallExpressionHelper.forEveryScopeTakeOverloadsOtherwiseImplementations.
(cherry picked from commit 99a624ab85957d7a2d3c2c0ced596e472f9d615b)
IJ-MR-148398
GitOrigin-RevId: c2cdfe8c8b046118f4e6f7269dbf7848dd746e08
Previously, we mistakenly tried to resolve qualified names listed in the
`field_specifiers` argument of @dataclass_transform in the same scope
where the dataclass itself is defined, not where the actual decorator
application is located.
Thus, if in the file where the dataclass is defined, a field specifier
was imported differently than where @dataclass_transform was applied, we
couldn't recognize a field specifier call in the RHS of an assignment as such
and took it for an ordinary field default value.
In particular, this is what happened with pydantic dataclasses.
`pydantic.fields.Field` is usually imported as just `pydantic.Field` where
user dataclasses are defined, but imported with an alias and set in the
`field_specifiers` argument as `PydanticModelField` in
`pydantic._internal._model_construction` where `ModelMetaclass` is defined.
It was accidentally broken in f15a07836e7aeac7c46b489b4742e8248a0e6ef4 to
support decorating class methods with dataclass_transform
(see testData/inspections/PyDataclassInspection/DataclassTransformFieldsOrder/decorator.py).
Until PyResolveUtil.resolveQualifiedNameInScope automatically traverses through
containing scopes looking for a name, the file containing decorator application
seems like a safe trade-off for the scope, because field specifiers are normally
defined or imported somewhere at the top level.
(cherry picked from commit de9afeb0831a52f058453fe678de229d41c26a4d)
IJ-CR-151380
GitOrigin-RevId: b6576ec7b72ea1e19e93b6190372a5168003c396
Some dataclass implementations, such as Pydantic, allow declaring fields with
mutable defaults, deep-copying them under the hood.
See https://docs.pydantic.dev/latest/concepts/models/#fields-with-non-hashable-default-values
(cherry picked from commit e495621858950976226731dddbb01af4012704fa)
IJ-CR-151192
GitOrigin-RevId: 7412272584a4c26e404d3d84e6150f974027eca7
For async functions, unwrap return type from Awaitable or Coroutine
Merge-request: IJ-MR-146295
Merged-by: Aleksandr Govenko <aleksandr.govenko@jetbrains.com>
(cherry picked from commit 9fe8d02a9d8bb584b9d6972ce999912bd93875e6)
IJ-MR-146295
GitOrigin-RevId: 9bad4877a069268a2d0181cac70b9a0d399cb5e6
Docstring rendering is no longer supported for Python 2, which became obsolete after reaching its end of life in 2020. Without updates or security patches, most tools, including documentation generators like Epydoc, have shifted focus exclusively to Python 3.
(cherry picked from commit ace78ac9ad943278449d5b20bb92db9f7571b5b5)
IJ-CR-148150
GitOrigin-RevId: 75cc87e05c61c3c17c26689552080e3c3082bfdc
Also, don't query LanguageLevel for each element of each instruction,
only once per scope traversal.
This doesn't fix the problem with unreachable definition *inside* blocks under
unmatched version checks, i.e.
if sys.version_info < (3, 8):
Alias = int
expr: Alias # unresolved
but it's a more difficult problem how to handle those consistently with the idea
of unreachable version checks under the *current* interpreter version, and hopefully
it occurs rarer than, say, unresolved top-level imports of common names from typing.
(cherry picked from commit 55fd4597c6d0860d290caba15fbf4d313e985a86)
IJ-CR-149696
GitOrigin-RevId: 357ada7e10618aef75c470e6cd878f7672109e83
Because async generator methods in ABC and protocols are supposed to be declared as
plain "def" methods so as not to confuse type checkers with their AsyncIterable
return type annotations, it's better to disable such type-hint re-use if methods
async-ness don't match and not wrap anything in typing.Coroutine implicitly.
See also https://mypy.readthedocs.io/en/stable/more_types.html#asynchronous-iterators
(cherry picked from commit 6342f15a7786ec0d02ee1ab2b18fd40fd1ca1430)
IJ-CR-149694
GitOrigin-RevId: 3e19f190d9334e6c8648462ebb5b61abe931b0e6
Previously, PyTypingTypeProvider.getReferenceType returned a type from a class attribute
type hint only for assignment to instance attribute located inside a class definition.
In other words, here we inferred the expected type from the annotation
and reported incompatible types in assignment:
```python
class C:
attr: int
def m(self):
self.attr = "foo"
```
but in the following we didn't:
```python
class C:
attr: int
inst = C()
inst.attr = "foo"
```
Now we try to resolve any qualified target expression to a class
or instance attribute and then infer the type from the corresponding
annotation.
(cherry picked from commit 086dbb678a8cd89cfe332bf801631568fb6c3a4d)
IJ-MR-147382
GitOrigin-RevId: 4e3f71baa598d4caf684d0aeab23d1a9a688b94d
Previously, we reported call arguments only if either all callee
candidates have unmatched arguments or all call candidates have
unmatched parameters. When there was a mix of the two, we reported
nothing.
(cherry picked from commit 97b42faf10de74ee7cd10f934d9eb94e1c8bbb34)
IJ-CR-146869
GitOrigin-RevId: 8babfe6a8ad0152f985655eff27df4df68936594
Reword the inspection text
Co-authored-by: Mikhail Golubev <mikhail.golubev@jetbrains.com>
(cherry picked from commit cab3cb1bbb1316462ee3fd37e735765b31b8d5e8)
GitOrigin-RevId: e3288e503d714a1f76bd1a2e6a0553770e1cbad5
* Remove unnecessary descriptor type inference from `PyTargetExpressionImpl.getType`
* Refactor overload resolution for synthetic calls, minor refactorings in PyDescriptorTypeUtil
* Add additional test for __set__
* Do not substitute missing argument types with implicit `Any` in PySyntheticCallHelper
* Always take the first overload for `__set__` if more than one is present
(cherry picked from commit 9d1f9d37f3ea23ce6145f82fc8f62212f744e858)
GitOrigin-RevId: 372dc0e8ccc32095aecf3f611645adc06c2873e9
Add API for inferring the expected type of `__set__` from `value` parameter
Add corresponding logic to PyTypeCheckerInspection to check if assigned value matches the expected descriptor type
Add tests on it
(cherry picked from commit b14ab7b2e40e225b508875a778ceae8986cbb291)
GitOrigin-RevId: 2b15b2b4527a95e5912897ba256dcc73d71c3dcd
The implementation is very similar to the PyTypeShed,
it should be merged into more general solution.
GitOrigin-RevId: 1449a774e93206349f17f264c27e2a7877d79b19
Protected member should not be highlighted as a warning if it resolves to .pyi file.
We assume that everything in .pyi file is a public API.
GitOrigin-RevId: c8275f3e48e3cd69b1676de9b78606f28ea224c8
There were two underlying problems in Kotlin code:
```
run { DjangoFQNamesProvider.ACCEPTS_VIEW_NAME.isNameMatches(this) } != null
```
returned true regardless of the result of `isNameMatches(this)` call.
As a result, we used to suggest Django view names for *all* functions called
"reverse", "redirect" or "reverse_lazy".
Secondly, in FQNamesProviderExtKt.isNameMatches:
```
return getQualifiedNames().any {
return it.firstComponent == elementQualifiedName.firstComponent &&
it.lastComponent == elementQualifiedName.lastComponent
}
```
always returned the result of comparison with the first qualified name, ignoring the rest
due to non-local return from the inline `any`.
Finally, DjangoUrlViewProvider.isReverseFunction expected a PyFunctionType as the type
of reverse/redirect/reverse_lazy, but since these functions have overloads in .pyi
stubs we started to infer a union type of PyFunctionTypes for them, breaking the reference
provider. In general, it's better to perform a name resolution instead of type inference
to detect a fully qualified name for a reference.
GitOrigin-RevId: 69949b1b0e65f00557536cf16127279a83ea4f9d
