StubValidator::validate() accounts for ~30% of phpstan execution time for a single file

[arnaud-lb]

Feature request

StubValidator::validate() accounts for ~30% of the execution time for a command like phpstan analyze -l 9 file.php.

(I'm using phpstan-symfony and phpstan-doctrine, which have some number of stubs.)

Reducing these 30% would enhance the experience of the phpstan integration in phpstorm or vim, for example.

[ondrejmirtes]

There are some things that make it faster:

1. PHPStan only analyses stubs that are directly mentioned in your phpstan.neon. Switching to phpstan/extension-installer and removing those includes from vendor might help speed it up.
2. Recently the stub validator results have been put into the result cache, but the result cache is not used when analysing a single file. Honoring advice from this article (https://phpstan.org/blog/why-you-should-always-analyse-whole-project) might also help you. I realize you're talking about IDE integration here, but I believe these should honor this article as well... Time to open some bug reports I guess :)

[arnaud-lb]

PHPStan only analyses stubs that are directly mentioned in your phpstan.neon. Switching to phpstan/extension-installer and removing those includes from vendor might help speed it up.

Awesome! I confirm that this speeds things up :)

Recently the stub validator results have been put into the result cache, but the result cache is not used when analysing a single file. Honoring advice from this article (https://phpstan.org/blog/why-you-should-always-analyse-whole-project) might also help you. I realize you're talking about IDE integration here, but I believe these should honor this article as well... Time to open some bug reports I guess :)

I'm using this in CI, but I've found that it's not suitable for commit hooks and IDEs because it can take too long: a few minutes when the cache is invalidated or at least 7-8s on some project when cached.

So I was searching for a way to execute phpstan quickly when I save a file in my editor.

To give you an idea, analyzing the whole project when fully cached takes 7-8s, from which:

• 6.33% in BetterReflectionSourceLocatorFactory::create() and callees
• 7.42% in InceptionResult::getFiles() and callees
• 21% in MemoizingSourceLocator::locateIdentifier() and callees
• 35% in IgnoredErrorHelper::initialize() and callees (the project has a baseline.neon with 900 items)
• 20% in ResultCacheManager::process() and callees

[staabm]

@arnaud-lb

35% in IgnoredErrorHelper::initialize() and callees (the project has a baseline.neon with 900 items)

Would phpstan/phpstan-src#744 help you regarding these 35% ?

[arnaud-lb]

I think that it would :) IgnoredErrorHelper::initialize() spends most of its time in IngoredRegexValidator::validate(), so definitely yes!

[ondrejmirtes]

Oh, I just got an idea - ignored errors regexes could be validated as part of a DIC compiler extension and thus only once when they are changed, not each time when PHPStan is executed :)

[ondrejmirtes]

I've almost got it :) But I noticed that the errors from the baseline are not current validated thanks to this line: https://github.com/phpstan/phpstan-src/blob/2d52a5b077b504898fbc78874476b22d18d605d8/src/Analyser/IgnoredErrorHelper.php#L79-L81

So a big baseline doesn't matter here.

[ondrejmirtes]

Here it is: phpstan/phpstan-src@692158c

Please wait a bit so that it builds and then test phpstan/phpstan:dev-master :) Thanks.

[arnaud-lb]

But I noticed that the errors from the baseline are not current validated thanks to this line: https://github.com/phpstan/phpstan-src/blob/2d52a5b077b504898fbc78874476b22d18d605d8/src/Analyser/IgnoredErrorHelper.php#L79-L81
So a big baseline doesn't matter here.

I didn't expect that. It's weird that it was taking so much time then. Hao\Compiler\Llk\Parser::parse() appear to take the most time of IgnoredErrorHelper::initialize() / IgnoredRegexValidator::validate().

Here it is: phpstan/phpstan-src@692158c

Awesome! I confirm that this completely eliminates the IgnoredErrorHelper::initialize() time :) I've also tried removing the count: lines from the baseline (for testing) : Before 692158c194eff722167f5316d31369a188a88a72 this makes phpstan multiple times slower (as expected) and after 692158c194eff722167f5316d31369a188a88a72 this makes no different (proving the cache works).

[ondrejmirtes]

Alright, can I close this or do you see any other bottleneck that can be solved? 😊

[arnaud-lb]

Yes, I'm closing the issue now. Thank you :)

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.