Re: [RFC] __exists(), a magic method for distinguishing "missing" from "set to null"

Hi Larry, Le sam. 2 mai 2026 à 17:47, Larry Garfield <larry@garfieldtech.com> a écrit : > On Fri, May 1, 2026, at 4:37 AM, Nicolas Grekas wrote: > > Dear internals, > > > > It's my pleasure to submit this new RFC to yours. > > Please have a look and let me know: > > https://wiki.php.net/rfc/exists-magic-method > > > > TL;DR: I'm proposing a new opt-in magic method: > > > > public function __exists(string $name): bool; > > > > It'd let userland tell "set to null" apart from "missing" on > > objects, > > it'd restore > > isset() <=> ?? equivalence on magic properties, and it'd fixe GH-12695 > > as a > > corollary. It's forward-compatible as a regular method on PHP <= 8.5 > > (probeable > > via method_exists()), and would be magic on 8.6+. > > > > Cheers, > > Nicolas > > Thanks for the very detailed RFC! > > > If it returns false, the fetch produces “uninitialized” (treated as null > by ??, so y is evaluated). __get() is not called. > > What is y? Looks like a reference to a previous iteration of the text. > > The Sequencing section also doesn't describe the behavior when __exists() > isn't defined. Does it just skip to step 4? (Please specify in the text..) > > Is the intent to eventually deprecate __isset()? The RFC it doesn't now, > but do you feel that is the eventual end-game, in PHP 10 or whatever? > > The big one: Should any parallel changes or additions be made to property > hooks? An exists() operation in addition to get/set, for instance? (I'm > not sure at the moment, but it should be thought through and discussed, and > if not, documented in the RFC.) Thanks for having a look! I made some changes to clarify, see: https://wiki.php.net/rfc/exists-magic-method?do=diff&rev2%5B0%5D=1777627745&rev2%5B1%5D=1777752909&difftype=sidebyside y => $y I added a description of the current sequencing of magic methods. About deprecating __isset, I think it's way too early to consider doing it for real. Still my answer is yes, but not on any timescale I can anticipate. I added words about the doc, which should now encourage __exists IMHO. About hooks, that deserves a separate RFC to me. I've no intuition on this as I don't use hooks often enough to feel the need for any missing "exists" hook. If you still want my noob take on this: a hooked property exists - because it has hooks. Not super helpful :D I added some words about this in the RFC. Cheers, Nicolas