Apply class purity tags to annotated methods

[truongbo17]

Summary

@method declarations currently have unknown purity even when their class is marked @phpstan-all-methods-pure or @phpstan-all-methods-impure. This makes a call to a pure annotated method report a possibly impure call from pure code.

Propagate the class-level purity tag when creating an annotation method reflection and use it for both isPure() and hasSideEffects(). Void-returning methods keep their existing side-effect behavior, matching native methods.

Closes phpstan/phpstan#15322

Tests

• Added a failing regression test for the reflection and a rule-level reproduction of the reported behavior.
• php vendor/bin/phpunit tests/PHPStan/Reflection/Annotations/AnnotationsMethodsClassReflectionExtensionTest.php
• php vendor/bin/phpunit tests/PHPStan/Rules/Pure/PureMethodRuleTest.php
• php -d memory_limit=450M bin/phpstan analyse src/Reflection/Annotations/AnnotationMethodReflection.php src/Reflection/Annotations/AnnotationsMethodsClassReflectionExtension.php tests/PHPStan/Reflection/Annotations/AnnotationsMethodsClassReflectionExtensionTest.php tests/PHPStan/Rules/Pure/PureMethodRuleTest.php --no-progress --debug
• php vendor/bin/parallel-lint on the six changed PHP files

[staabm]

my gut feeling is, that magic behaviour and purity kind of contradict each other.

phpstan cannot give guarantes for a method it will not see/know and I think people will shoot into their own foot by doing impure stuff in such magic context without seeing it

[VincentLanglet]

Now you point it out, it's indeed annoying that

• PHPStan won't report impurity in magic
• People cannot annotate a magic method as impure even when all are annotated as pure (which is doable with a well defined method).

[VincentLanglet]

With your change, do we still report the following example
https://phpstan.org/r/fb9d4b25-5cbf-4335-b566-11fdc26a1f0d

?

.

[truongbo17]

Thanks for the counterexample. The previous revision suppressed that diagnostic. I updated the annotated method reflection so a native __call() or __callStatic() known to be impure takes precedence over the class-wide pure tag. Running your linked example on the updated branch now reports Impure call to method Foo::magicFoo() in pure method Bar::magicBar(). at line 36.

The regression test covers both instance and static dispatchers, while the original issue case (no impure dispatcher) remains pure. The focused purity and annotation tests pass locally.

[VincentLanglet]

I'm not sure about the order of those condition.

As soon as nativeCallMethod is not null shouldn't we use the purity from it ?

[truongbo17]

Agreed. I changed the precedence in 70fe6a2: when a native __call() or __callStatic() exists, the annotated method now takes its yes/no/maybe purity from that dispatcher. The class-wide tag is used only when there is no native dispatcher. I added a regression case where an explicitly pure __call() overrides @phpstan-all-methods-impure, alongside the impure dispatcher cases. The focused tests and PHPStan analysis pass locally.