Hey, > Am 18.06.2020 um 17:18 schrieb Benas IML <benas.molis....@gmail.com>: > > Hey Bob, > > Magic methods are **never** supposed to be called directly (even more if that > method is a constructor or a destructor). If that's not the case, it's just > plain bad code. But by enforcing these rules, we make sure that less of that > (bad code) is written and as a result, we make PHP code less bug-prone and > easier to debug. That's also most likely the reason why
__construct() is invoked directly on parent calls, sometimes to reinitialize an object or after ReflectionClass::newInstanceWithoutConstructor. I invoke __destruct() directly when needing an early freeing of existing resources. > "ensure magic methods' signature" RFC opted in to validate `__clone` method's > signature and ensure that it has `void` return type. > > Just for the sake of making sure that you understand what I mean, here are a > couple of examples that show that no magic method is ever supposed to be > called directly: > ```php > // __toString > (string) $object; I like using ->__toString() in favor of (string) casts when the variable is guaranteed to be an object to highlight that and avoid magic-ness. > // __invoke > $object(); Same here, unless the object is a closure. > // __serialize > serialize($object); > ``` Can't argue much about that one, I never use serialize(). > Moreover, by validating constructors/destructors and allowing an explicit > `void` return type declaration, we are becoming much more consistent > (something that PHP is striving for) with other magic methods (e. g. > `__clone`). Yeah, __clone() is odd. No idea why. > Also, saying that "sometimes you have valid information to pass from the > parent class" is quite an overstatement. After analyzing most of the 95 > Composer packages that had a potential BC break, I found out that either they > wanted to return early (that is still possible to do using `return;`) or they > added a `return something;` for no reason. Thus, no libraries actually > returned something useful and valid from a constructor (as they shouldn't). > > Last but certainly not least, constructors have one and only one > responsibility - to initialize an object. Whether you read Wikipedia's or PHP > manual's definition, a constructor does just that. It initializes. So, the > PHP manual is perfectly correct and documents the correct return type that a > constructor should have. It also is generally a bad idea to have side effects in constructors, but _sometimes_ it is justified. Only because something mostly is a bad idea, it is not always. Also note that other languages completely forbid manual ctor calls. But PHP doesn't (and for good reason, like after using ReflectionClass::newInstanceWithoutConstructor). Bob > Best regards, > Benas > > On Thu, Jun 18, 2020, 4:06 PM Bob Weinand <bobw...@hotmail.com > <mailto:bobw...@hotmail.com>> wrote: > > Am 17.06.2020 um 01:10 schrieb Benas IML <benas.molis....@gmail.com > > <mailto:benas.molis....@gmail.com>>: > > > > Hey internals, > > > > This is a completely refined, follow-up RFC to my original RFC. Based on the > > feedback I have received, this PR implements full validation and implicitly > > enforces `void` rules on constructors/destructors while also allowing to > > declare an **optional** explicit `void` return type. Note, that there is a > > small but justifiable BC break (as stated by the RFC). > > > > RFC: https://wiki.php.net/rfc/make_ctor_ret_void > > <https://wiki.php.net/rfc/make_ctor_ret_void> > > > > Best regards, > > Benas Seliuginas > > Hey Benas, > > I do not see any particular benefit from that RFC. > > Regarding what the manual states - the manual is wrong there and thus should > be fixed in the manual. This is not an argument for changing engine behaviour. > > Sometimes a constructor (esp. of a parent class) or destructor may be called > manually. Sometimes you have valid information to pass from the parent class. > With your RFC an arbitrary restriction is introduced necessitating an extra > method instead. > > In general that RFC feels like "uh, __construct and __destruct are mostly > void, so let's enforce it … because we can"? > > On these grounds and it being an additional (albeit mostly small) unnecessary > BC break, I'm not in favor of that RFC. > > Bob