Improve performance by change count(array) to empty array [] comparison

[samsonasik]

@ondrejmirtes @TomasVotruba this is for increase PHPStan performance improvement by replace count($array) === 0 or count($array) > 0 to empty array comparison $array === [] or $array !== [], it utilize rector's Rector\CodingStyle\Rector\FuncCall\CountArrayToEmptyArrayComparisonRector::class.

Rector probably needed to be part of CI workflow to verify it if it is approved, @ondrejmirtes what do you think?

[samsonasik]

compiler tests notice seems unrelated

[ondrejmirtes]

Hi, thank you, but I doubt this has real world performance implications. Did you measure it?

[samsonasik]

You can check this benchmark https://phpsandbox.io/n/polished-violet-mwjd-llhvj

Count took: 0.316ms
Compare to empty array took: 0.018ms

[TomasVotruba]

@samsonasik Cool tool, thanks for sharing the link 👍

I tried to re-run it couple of times and got wide range of numbers:

• count: 160-250 ms
• compare empty array: 15-30 ms

Factor 10x for the empty array.

[ondrejmirtes]

Yeah, but it's not obvious what it means for PHPStan from these benchmarks. I'll have to compare it with Blackfire.

[orklah]

Weird, I don't reproduce the same times with 3v4l at all:
https://3v4l.org/uc9iY

Maybe a difference in opcache setup?

I had to reduce the content of the array by a factor of ten, but I doubt this would have this kind of dramatic effect. I also have a different benchmark script that show similar results: https://3v4l.org/TNFbu (ie, no dramatic effect in one way or another)

But I still prefer !== [] / === [] notation !

[ondrejmirtes]

I prefer the count version so I need some hard evidence this improves PHPStan performance by at least a few percents :)

[staabm]

AFAIR will php-src optimize code when you write \count() at OPCODE compile time.

using count without a prior use function or without a fully-qualified notation will not benefit from php-src compilation passes.
i.e. this PR as is improves performance when the native count function is not imported via use function.
(this only works for namespace'd code and depends on opcache beeing used)

[Dgame]

I prefer the count version so I need some hard evidence this improves PHPStan performance by at least a few percents :)

I doubt that there is one. In contrast to earlier PHP versions (I think from >=PHP 7.0) count only reads the length property of the internal buckets, so it's not as costly. I also like the === [] comparison much better, but if you compare both versions, you see almost no difference: https://3v4l.org/tF1WE

[staabm]

using count without a prior use function or without a fully-qualified notation will not benefit from php-src compilation passes.
i.e. this PR as is improves performance when the native count function is not imported via use function.
(this only works for namespace'd code and depends on opcache beeing used)

my assumption was wrong

https://twitter.com/nikita_ppv/status/1482096003721355266?s=20

[staabm]

in #747 (comment) we concluded that this change is not worth it