faster TypeCombinator

[voku]

https://blackfire.io/profiles/compare/a42e1474-696b-4854-8eeb-0b5ea9869b77/graph ~18% less count calls ⇒ ~6% faster overall

[staabm]

Maybe we should replace count($xy) === 0 with $xy === [] all over the codebase via a cs fixing rule?

[staabm]

I guess the same is true for count($xy) !== 0 and $xy !== []

[voku]

Maybe we should replace count($xy) === 0 with $xy === [] all over the codebase via a cs fixing rule?

At work I always write "count($x) === 0" because its more readable, but here the code will be called very very often and it seems to make a difference if we call it 100000x so I don't know if we should change it everywhere? 🤔

[ging-dev]

There is no significant performance gain between them.

See: symfony/symfony#43779

[voku]

Yes, I also tested it and had a quick research and count() is O(1) since the number of elements is stored in the
HashTable. So, the performance benefit in this PR was maybe from not interacting with the array at all?

[ging-dev]

Exactly

[staabm]

please try to add a use function count; on the top of the class, so PHP opcache feature can fully optimize this code.

I don't have a codebase where count() is in the profiles, so I cannot check this assumption myself.

[voku]

please try to add a use function count; on the top of the class, so PHP opcache feature can fully optimize this code.

I don't have a codebase where count() is in the profiles, so I cannot check this assumption myself.

Isn't there a fixer for it? 🤔

[staabm]

Isn't there a fixer for it?

Yep in php-cs-fixer it is.

Phpstan uses slevomat and I dont know its supported with it in the same fashion

[voku]

Isn't there a fixer for it?

Yep in php-cs-fixer it is.

Phpstan uses slevomat and I dont know its supported with it in the same fashion

I think this Sniff will automatically fix it:

SlevomatCodingStandard.PHP.OptimizedFunctionsWithoutUnpacking

@ondrejmirtes should we use it? 🤔

[ondrejmirtes]

Slevomat CS definitely has a sniff to enforce using all functions, but I don't like that codestyle and I think it has negligible impact on performance. I'd accept a change that makes the PHAR compiler do it (https://github.com/phpstan/phpstan-src/tree/master/compiler) because that's where I think it belongs, but I don't want it in phpstan-src.

[ondrejmirtes]

Thank you!

[voku]

@staabm @ondrejmirtes one maybe stupid question : why can't php optimize count call in condition like "count === 1" 🤔 ... we only need to count 0,1,break but I think php will count the full array, or?

[ondrejmirtes]

@voku It probably could, raise that on https://bugs.php.net/ :)

[ondrejmirtes]

But maybe not, you have to think of edgecases like Countable implementations...

[voku]

But maybe not, you have to think of edgecases like Countable implementations...

Maybe it can be optimised for "normal" arrays?

https://bugs.php.net/bug.php?id=81561

[staabm]

within namespaced code, php can optimize count calls via opcache only when its either fully-qualified \count(..) or when you explictly use a use function count; at the top