Hello world, it's Wednesday, August 26, 2026, and here's what happened This Week in PHP Internals.
10 stories this week, so let's get into it. But first, You shipped a lot of code this quarter, but do you know how much of it is actually still there? Ballast reads your git history — never your code — and scores what survives. Stable velocity, plus a durability rating between 300 and 850 — like a credit score for your codebase — updated every month, and it's free. ballast.now.
This week's top story isn't really about PHP. It's about how the internals list talks to itself. Messages have been arriving that were clearly written or heavily assisted by a large language model — that flat, over-professional register that thanks you warmly and then restates your own point back at you — and somebody finally said the quiet part in plain text. 24 messages followed, about where the line actually sits. Nobody argued against translation. English is a second or third language for a good part of that list, and that is a real accessibility question rather than a laziness one. What people object to is authorship — the argument itself being handed to a machine — and the harder problem underneath, which is that there's no reliable way to tell the difference from the outside. Nothing's been written down yet. But if you want a say in what does get written down, the community Discord is at phpc.chat, and there's an internals channel there. It's a community server, not an official PHP one. There's a thread in it working on exactly these guidelines. So if you want to head there and argue with us about what should be in the guidelines, feel free to join the discussion.
A new RFC opened on Saturday, from the same contributor whose last one had just been withdrawn. Sepehr Mahmoudi proposed array_match(), which filters an array down to the values containing a substring, implemented in C so you don't have to pay for a closure on every element. Yuya Hamada replied within the hour, asking whether array_filter() isn't already enough, pointing at a GitHub search full of userland functions already named array_match that don't all behave alike, and noting that the str_icontains RFC was declined, so an ASCII-only case-insensitive flag is a hard sell. Then Tim Düsterhus pointed out that in 8.6 this is already 1 line, because partial function application lets you drop str_contains into array_filter() with a placeholder. Christian Schneider noted that preg_grep() has done basically the same job for years.
The case against it was mostly one case, made 5 ways. Ayesh Karunaratne went after the premise, writing: "I have worked on Drupal, WordPress, Silex, and bespoke code bases, spending enough time profiling them. An array string search has never been a bottleneck." Larry Garfield called it an XY problem and said huge in-memory arrays are a code smell to begin with. mickmackusa offered the alternative nobody else had — make str_contains() polymorphic, the way str_replace() already takes an array. Sepehr has since dropped the case-insensitive flag, renamed it array_str_contain(), and moved the target to 8.7, since 8.6 is frozen. Nobody's argued in favour of it so far. The proposal, so far, contains no support.
Nick Sdot brought a documentation question to the list on Thursday. He's restructuring PHP's internals book, the docs that live inside php-src, and his first pull request converts the pages from reStructuredText to Markdown — a light 1,100-line mechanical diff that nets out 86 lines shorter. Ilija Tovilo, who set the book up, called it churn, and answered that the source is neither pure Markdown nor reStructuredText but MyST, so renaming files fixes some rendering and breaks the reStructuredText-flavoured parts. His real objection was the diagnosis. He wrote: "None of this is wrong, but I don't think the docs have stalled because the syntax is too hard. Rather, more time should be designated to them." Weilin Du and Calvin Buckley both said the effort belongs in the content. Nick's clarification explains the thread. The old docs folder he's merging in is already Markdown, so a conversion happens in one direction or the other. It's still open.
Robert Chapin spent the week working through an old RFC. Deprecate Fuzzy Type Casts, by Alexandre Daubois and Nicolas Grekas, dates to January, is still under discussion, and still targets 8.6 on the wiki — which Robert would like changed, given there's been no discussion since. His objection is that the RFC's own migration guidance doesn't hold up. He points out that with the string 1.5, is_numeric() returns true, so the validate-before-casting example still triggers the deprecation, while is_int() and is_float() return false for every numeric string. He says the filter_var() alternative doesn't guarantee an integer back either. He asked: "Are developers expected to replace a sing[le] type cast with 8 lines of code for PHP 8.6?" Nobody has answered him.
A process question opened yesterday morning. Sepehr Mahmoudi noticed that some RFCs proposing new functions include a userland polyfill and some don't, and asked whether the RFC template should recommend one wherever that's feasible. His reasons are that a polyfill pins down the exact behaviour, edge cases and all, without anyone reading the C, and that it gives teams like Symfony's polyfill maintainers an accurate head start. Larry Garfield called recommended-where-relevant-but-not-required a reasonable policy, and added a second use for it, saying: "And it would also give us a target to benchmark against to see if putting it in C really has a performance benefit." Which is exactly what 2 people spent the weekend asking Sepehr for on his other proposal.
Gina P. Banyard posted a correction on Friday to an already-accepted RFC. Steven Wilton's SNMP improvements proposal states that its target is the next stable release for all currently supported PHP versions. Gina's point is that this runs against the release process policy on patch versions, so the changes will land in 8.6 only. She also asked for a design change — the new integer constants in the output-controls section would become PHP enums instead, which improves type safety in userland and translates the C enums SNMP already defines into PHP ones. The plan is to have it merged in time for 8.6.0 beta 2.
Weilin Du, who maintains the intl extension, spent part of the week handing out work. He'd like some of the extension's constants turned into enums, proper namespaces on its classes so they stop colliding, and the remaining gaps between PHP's APIs and the underlying ICU APIs filled in. He was blunt about the error handling, writing: "the error state stuff in the extension is, frankly, bad, in my opinion, and could be optimized since it has caused a lot of very stupid bugs." He suggested the enums and namespaces could be an 8.7 RFC if somebody picks it up. Yuya Hamada added the reason it matters. Feedback from right-to-left language users is thin across the whole industry, not just PHP, and PHP's internationalisation depends on getting it.
Quick hits. array_search_range(), last week's top story, is officially withdrawn — mickmackusa posted a 7-point review hours after we recorded, and Sepehr Mahmoudi pulled it on Thursday and called it a valuable learning experience. Weilin Du's extension-name case sensitivity proposal is going to a vote, eventually. A core developer objected, which under policy makes it an RFC, and he'll open that after the 8.6 branch is cut on September 22, with both behaviours on the ballot for 8.7. And Zachary DuBois introduced himself on Thursday, asked whether adding missing libsodium functions needs an RFC or just a pull request, was told to open the pull request, and had it up by Sunday. It adds X-Wing and ML-KEM-768 bindings to the sodium extension. Post-quantum key encapsulation, arriving as somebody's first pull request. Quantum-resistant, and so far review-resistant.
So that's the week: no RFC in the voting phase for the second week running; a long argument about who is actually writing the mail on the internals list; a new array_str_contain() proposal with 5 people against it and nobody for it; a documentation pull request that turned out to be a consolidation question rather than a format preference; unanswered holes in the fuzzy casts migration path; and post-quantum crypto arriving as somebody's first pull request. Links to every thread are below. The PHP Foundation bankrolls a lot of php-src development, so if you use the language, maybe consider donating at opencollective.com/phpfoundation — or try guilting your employer into it. Thanks to Ballast.now for supporting this week's episode. We're Artisan Build. See you next week.
