Re: [RFC] [DISCUSSION] Ensure correct magic methods' signatures when typed

On Tue, Apr 14, 2020 at 7:45 PM Claude Pache <claude.pache@gmail.com> wrote: > > > Le 14 avr. 2020 à 18:53, Nikita Popov <nikita.ppv@gmail.com> a écrit : > > On Tue, Apr 14, 2020 at 6:07 PM Claude Pache <claude.pache@gmail.com> > wrote: > >> >> >> > Le 14 avr. 2020 à 16:54, Nicolas Grekas <nicolas.grekas+php@gmail..com> >> a écrit : >> > >> > I'm just not sold on allowing "void" on __construct, because the very >> concept of a return type on a constructor is ... void, and also because of >> the code style choices this will open (and the CS "wars" I mentioned). >> > >> >> This issue is not specific to magic method like __construct(). It is the >> whole concept of “void” as return type which is, say, “problematic”. >> >> In fact, “void” is not really a return type. It is a way to state that >> the method is not supposed to return anything, which means, as you said >> very well, that “the very concept of return type on [this method] is void”. >> >> That might be a reason to reject the concept of “void” as return type. Or >> to revive https://wiki.php.net/rfc/allow-void-variance < >> https://wiki.php.net/rfc/allow-void-variance> . But >> again, the issue is >> orthogonal to the fact that this particular method is magic, and we should >> not cherry-pick and reject the concept of “void” for __construct() and >> similar magic methods only. > > > Constructors not having a return type is standard behavior across most > (all?) languages. You can't specify a constructor return type in C++. You > can't specify one in C#. You can't specify one in Java. Off-hand, I can't > name a language that both has a first-class constructor concept (Rust's > "new" idiom does not count) and specifies a return type on it. > > It would naturally follow that, yes, you can't specify a constructor > return type in PHP either, just like we enforce right now. Unless we have > some strong reason to deviate from standard behavior that I do not see? > > Regards, > Nikita > > > Do those languages allow to return something from the constructor (as does > PHP currently)? because I’m more interested in the semantics of : > void > than the exact way to have it. > Ah, so that's what this is about! In that case, I'd be happy to simply always enforce that __construct() cannot return a value, in the same way we do for ": void" functions. (If we have backwards compatibility concerns, we can add this as a warning instead of hard error.) Nikita