Cache PCRE named group names

[mvorisek]

Description

I have discovered PCRE preg_match is at least twice as slow when using named groups.

PCRE internally remaps named groups to indexes, so the slowdown is comming from php-src.

I belive the reason is the named group names zvals are created each time the preg_match is run.

I propose to improve the performance by caching the names zvals in the regex cache entry - https://github.com/php/php-src/blob/php-8.3.7/ext/pcre/php_pcre.c#L55

[nielsdos]

Do you have a test case for this?

[mvorisek]

Here is a testcase:

with 5 named groups - https://3v4l.org/HDdlX/rfc#vgit.master_jit - the performance difference is 2x
with 50 named groups - https://3v4l.org/PFe2A/rfc#vgit.master_jit - the performance difference is 8x

So the optimization potential is huge and I belive when the names zvals will be cached, the performance difference can be cut by 10x at least.

[nielsdos]

Thanks, can confirm. The idea probably makes sense. Let me try something.

[nielsdos]

I opened a draft PR.
With your code for 50 groups:

Before:

unnamed: 6.39 ms
named: 64.39 ms

After:

unnamed: 6.49 ms
named: 6.55 ms

So yeah, there's a huge potential here.