This Week In PHP Internals - June 17, 2026

Hello world, it's Wednesday, June 17, 2026, and here's what happened This Week in PHP Internals.

We start where we left off last week: generics. Last week, this was the most traction generics had seen in years. This week, Seifeddine Gmati took his Bound-Erased Generics RFC to a vote — and he did it over Larry Garfield's explicit objection.

Here's the thing about that vote: as of this recording, it's losing. 7 in favor, 18 against, 5 abstaining. It needs a two-thirds majority, and it's sitting around a quarter. Rowan Tommins summed up the No camp in one line: "I want PHP to have generic types, but I want to be able to trust those types. This RFC does not deliver that." Erased generics let you write a type the engine won't enforce — and 18 voters decided that's a step they're not willing to take.

Larry had warned them. He reposted this into the vote thread for visibility: "you don't call the vote until you know you have the votes." Right now, that looks prescient.

So what happens to generics now? Two camps are already sketching the next move. Larry's pitch: don't cram this into 8.6. Make 8.6 the last of the 8.x line, make PHP 2027 a 9.0, and ship enforced generics there as the headline — "PHP 9, now with generics." Levi Morrison surprised himself by voting yes anyway, as a "stepping stone" you tighten later. And a non-voter writing as wheakerd made the plea of the week: don't let this collapse into "full reified generics now, or PHPDoc forever" — ship a smaller, enforceable subset first, interfaces and abstract classes, and grow from there. The vote runs through June 28. There's also a secondary vote on syntax — +T/-T versus in T/out T — and the C#-style in-out spelling is winning 18 to 4.

And honestly, Seifeddine had a week. While his generics RFC was going down, he opened a brand-new one: Literal Scalar Types. The idea — let a parameter be typed as the literal values it actually accepts. array_filter's $mode really only takes 0, 1, or 2, but it's typed int, because that's all we can say today. Literal types let the signature say what it means. David Gebler's first reaction was the obvious one — don't enums already do this? Seifeddine's answer: enums are a closed set in a new file with an import; a literal union describes existing scalar APIs and open value sets without any of that. And he found time to back a third proposal too — Jorg Sowa's revival of case-sensitive PHP. Busy man.

Remember .phpc from last week — pure-code files with no <?php, the proposal that collapsed so thoroughly even its biggest defender conceded? Hendrik Mennen brought it back. As a full RFC this time, with a working patch — 50 lines of C, ninety-eight hundred thirty-six tests passing, and a sub-vote on the file extension.

The reception? Same as last time. Matteo Beccati: "unnecessary complication for practically no real-world benefit." Kamil Tekiela took the motivation apart piece by piece. But credit to Hendrik — instead of digging in, he conceded the weak arguments and sharpened to one honest sentence: most new PHP files are pure code, the <?php carries zero meaning in them, and PHP is the only modern mainstream language that still makes it mandatory. Then he asked the room to tell him plainly — is the friction worth solving, is the mechanism wrong, or is it just not a real problem? That's how you run a discussion.

Paul Jones opened — and I'm quoting his own subject line — "Function Autoloading, mark 5." The fifth crack at letting PHP autoload functions the way it autoloads classes. Rob Landers, who wrote mark 4, replied with the dry truth: "the consensus was that the SPL autoloader needs to go, and nobody wanted more added to it... maybe the list has changed its mind. Hopefully." And the early replies bring real concerns: Anton Smirnov and Rowan Tommins both flagged the same footgun — lean on function autoloading, forget to fully-qualify a name, and your code might accidentally work, only because something else loaded that function first, then silently fall back to a global on the next refactor. Rowan's fix: a declare() at the top of the file that pins how unqualified names resolve. Function autoloading is PHP's white whale, and attempt 5 is already drawing the hard questions.

Now — voting season. A lot of votes were live this week, and it was brutal at the top end. We covered generics. Two more are going down with it.

Nicolas Grekas put his __exists() method to a vote — a genuinely clever fix: a magic method that finally lets an object tell a property set to null apart from one that doesn't exist at all, the way array_key_exists() does for arrays. Beautifully argued RFC. The result so far: 0 yes, 9 no, 5 abstain. Sometimes the list just doesn't want the extra magic. And Daniel Scherzer's ReflectionAttribute::getCurrent() — which lets an attribute know what it's attached to — is failing 2 to 5, with Larry, Ocramius, and Benjamin Außenhofer all pushing an interface-based alternative instead.

But the smaller, focused stuff is sailing through. Tim Düsterhus's RFC to deprecate returning values from constructors and destructors: 23 to nothing. Weilin Du's Locale display-keyword additions: 6 to nothing. Jakub Zelenka's Polling API already passed, 33 to 1, and it's merged. And a security win — Sjoerd Langkemper's RFC to cap php://filter chains at 16, which shuts down a nasty local-file-inclusion-to-remote-code-execution trick, is cruising at 26 to nothing. The pattern's hard to miss: tight, well-scoped proposals pass easily; the ambitious, magic-heavy ones are getting turned away.

A human follow-up, and a good one. Last week we flagged Steven Wilton, who'd shipped an SNMP module RFC and gotten total silence — "someone go help the man." This week, someone did. Tim Düsterhus picked it up: he leveled with Steven that SNMP is niche enough that reviewers are scarce, said he couldn't do it himself — and then went and recruited one in the Foundation Slack. He left Steven a concrete checklist: rebase, undraft the ready PRs, link the RFC. Steven did all of it within 2 days. The PRs are now waiting on a reviewer. The list came through.

On governance: Ben Ramsey's Working Groups RFC came roaring back this week. The idea — let the project charter small teams with real authority, so not everything needs a full-list vote. Tim Düsterhus still isn't sold; he thinks the RFC process already covers it. But the spicy moment came when Alex Rock floated whether PHP Foundation staff should be barred from voting on RFCs. Larry shut that down hard: "the Foundation employs a majority of the people who really understand the engine — excluding them from voting would be project suicide. The Foundation has no vote. The people who do half the work, who happen to work for the Foundation, have votes. As they should."

The modules saga refuses to die, and it's spinning off children. Alex Rock extended his proposal with Packages — Rust-style crates, with package-level visibility. And Michael Morris opened a whole new thread called "Containers," arguing that "namespaces in PHP are a bit of a hack" and floating containers as cleaner encapsulation. Rowan Tommins and Alex Rock are both in there. No RFC yet — this one's still at the whiteboard. But the appetite for real encapsulation in PHP clearly isn't going away.

Two quieter movers. Nicolas Grekas's serializable-closures RFC — fixing the fact that 8.5's attribute closures silently break serialize()-based caches — got its first real review from Tim Düsterhus, who wants it split in two and thinks the security model is more cautious than it needs to be. And Jordi Kroon announced he'll open voting Friday, June 19, on pulling third-party extension docs — imagick, redis, mongodb — out of the main PHP manual. Watch for that one to go live.

Quick hits. Xavier Leune wants to expose libcurl's socket callbacks so you can do SSRF filtering natively — validate the peer IP before connect — without dropping to FFI. A newcomer pitched a get_favicon() function and got the classic welcome: Kamil Tekiela asking, politely, what problem it actually solves. Larry and Ilija's 5-year-old Pattern Matching RFC picked up sharp new feedback on whether a failed match should be able to fail silently. And Roman Pronskiy revised his social-media policy to add a "discretion" principle — the team can refuse a platform that's a bad fit — a quiet echo of the X.com fight from last week. And fresh overnight: Jorg Sowa's case-sensitive-PHP revival drew immediate fire. Rob Landers, flatly — "I honestly can't think of anything good that this RFC would bring" — while others flagged the casing salad in PHP's own built-ins, where DateTime, PDO, and mysqli already disagree, as the real sticking point.

That's the week — and a heavy one: a generics vote going down, 3 RFCs rejected, 4 passed, and a modules debate that keeps multiplying. Links to every thread are below. We're Artisan Build. See you next week.