Re: [PHP-DEV] Discussion on array_str_contains() RFC and performance benchmarking methodology

در تاریخ پنجشنبه ۱۰ سپتامبر ۲۰۲۶، ۱۵:۳۷ youkidearitai <
[email protected]> نوشت:

> ‪2026年9月10日(木) 20:40 ‫سپهر محمودی‬‎ <[email protected]>:‬
> >
> >
> >
> > در تاریخ پنجشنبه ۱۰ سپتامبر ۲۰۲۶، ۰۶:۰۶ youkidearitai <
> [email protected]> نوشت:
> >>
> >> ‪2026年9月10日(木) 10:07 ‫سپهر محمودی‬‎ <[email protected]>:‬
> >> >
> >> >
> >> > Hi everyone,
> >> >
> >> > I hope you're all having a great week.
> >> >
> >> > I’ve been following the discussions and recent feedback regarding the
> array_str_contains() RFC closely. Performance and speed comparisons have
> come up as a key point, especially around whether a dedicated internal C
> implementation provides sufficient value compared to a userland foreach
> loop.
> >> >
> >> > I'd love to open up a constructive and transparent discussion around
> the benchmarking methodology we're using here. Some of the initial numbers
> shared recently don't seem to reflect real-world scenarios or different
> data profiles (like varying array sizes, early exits vs worst-case lookups,
> UTF-8 strings, and compiler optimization flags like -O2 / -O3).
> >> >
> >> > Before drawing final conclusions on performance, I think it would be
> great if we could align on:
> >> > 1. The exact benchmark scripts and test datasets we should use to
> evaluate this fairly.
> >> > 2. The compilation environment and flags used for generating these
> metrics.
> >> > 3. Where the potential overhead in the current C implementation is
> coming from and how we can optimize it together.
> >> >
> >> > My goal has always been to make PHP more expressive and efficient for
> everyday developers. I welcome any suggestions, insights, or benchmark
> reproductions from the community so we can evaluate this RFC based on
> solid, agreed-upon data.
> >> >
> >> > Thanks for your time and feedback!
> >> >
> >> > Best regards,
> >> > Sepehr
> >> >
> >>
> >> Hi Internals
> >>
> >> For your reference, I have attached the benchmark diff files. (commit
> >> hash is 4982bf45f4786faaa198558dd619f8a4d59e25f0)
> >> - .diff
> >> - .exp
> >> - .log
> >> - .out
> >> - .php
> >> - .phpt
> >> - .sh
> >>
> >> I can not find definite improve performance.
> >> (My PC: WSL Ubuntu 24.04, Ryzen 7 7735HS, 32GB RAM)
> >>
> >> I don't familiar with Zend Engine's performance, However I think
> >> function's performance improvement is limited.
> >>
> >> Regards
> >> Yuya
> >>
> >> --
> >> ---------------------------
> >> Yuya Hamada (tekimen)
> >> - https://tekitoh-memdhoi.info
> >> - https://github.com/youkidearitai
> >> -----------------------------
> >
> >
> > Hi Internals,
> >
> > Thank you all for the time and effort invested in this discussion.
> >
> > Regarding the benchmarks provided by Yuya, I must address two critical
> issues:
> >
> > 1. The provided implementation does not even build in my environment. It
> is scientifically impossible to derive any performance conclusions from
> code that fails to compile.
> >
> > 2. Given your own admission of limited familiarity with the Zend Engine
> internals, the methodology applied here lacks the necessary technical rigor
> for a valid assessment.
> >
> > I am more than willing to engage with constructive, reproducible, and
> engine-aware performance analysis. However, as it stands, the provided
> material is effectively contentless and does not offer a basis for a
> technical evaluation.
> >
> > Best regards,
> >
> > Sepehr
> >
> >
>
> Hi
>
> PLEASE UNZIP ATTACHED FILE. YOUR CODE.
>
> Regards.
> Yuya
>
> --
> ---------------------------
> Yuya Hamada (tekimen)
> - https://tekitoh-memdhoi.info
> - https://github.com/youkidearitai
> -----------------------------
>

Mr. Hamada,

If you lack a solid grasp of Zend Engine internals and core data structures
like zval—which is the very backbone of PHP’s C runtime—it feels like you
are simply relying on an AI assistant rather than technical understanding.
Without mastering zval and Zend memory handling, working at this level in C
is effectively crippled.

To answer your question: I am testing this on high-end industrial 64-bit
hardware, and your provided files still completely fail to build.

Perhaps you should tell your AI to think a bit harder before generating
patches and benchmarks.