[PropertyInfo] Adds static cache to PhpStanExtractor

[mvhirsch]

Q	A
Branch?	7.2
Bug fix?	no (performance)
New feature?	no
Deprecations?	no
Issues
License	MIT

I was able to detect a performance penalty when using dozens of traits in even more classes. The PhpStanExtractor creates a NameScope every time, but afaik this can be cached like in PhpDocExtractor (see #32188).

The performance impact is impressive, as it reduces the time needed for my TestCase by 30%. See Blackfire profile comparison.

[derrabus]

Please target 7.1. We usually don't merge performance improvements into LTS branches.

[xabbuh]

Shouldn't we now also implement the ResetInterface to prevent memory leaks?

[mvhirsch]

Shouldn't we now also implement the ResetInterface to prevent memory leaks?

If so, the PhpDocExtractor should implement it too, I guess.

[nicolas-grekas]

implement the ResetInterface to prevent memory leaks?

If all we store is class metadata, we don't need a reset IMHO as the leak is upper bounded, and resetting might effect performance of long running kernels for little memory savings in the end.

[nicolas-grekas]

Suggested change

	$nameScope = $this->contexts[$class.$declaringClass] ??= $this->nameScopeFactory->create($class, $declaringClass);
	$nameScope = $this->contexts[$class.'/'.$declaringClass] ??= $this->nameScopeFactory->create($class, $declaringClass);

[nicolas-grekas]

would it make sense to move this inside the loop, right before the nested foreach?:

$nameScope ??= $this->contexts[$class.'/'.$declaringClass] ??= $this->nameScopeFactory->create($class, $declaringClass);

[mvhirsch]

would it make sense to move this inside the loop, right before the nested foreach?:

You mean here?

symfony/src/Symfony/Component/PropertyInfo/Extractor/PhpStanExtractor.php

Line 127 in de14980

Sounds reasonable to me. It may make sense to skip creating one unless really needed (and since my change is all about re-using a created one, it hardly influences performance).