Edit report at https://bugs.php.net/bug.php?id=55554&edit=1
ID: 55554 Updated by: g...@php.net Reported by: ryan at zuttonet dot com Summary: Trait methods overriding legacy constructors -Status: Assigned +Status: Open Type: Bug Package: Scripting Engine problem Operating System: Ubuntu PHP Version: 5.4.0alpha3 Assigned To: gron Block user comment: N Private report: N New Comment: This bug has been fixed in SVN. Snapshots of the sources are packaged every three hours; this change will be in the next snapshot. You can grab the snapshot at http://snaps.php.net/. Thank you for the report, and for helping us make PHP better. Previous Comments: ------------------------------------------------------------------------ [2011-10-09 11:13:03] g...@php.net Automatic comment from SVN on behalf of gron Revision: http://svn.php.net/viewvc/?view=revision&revision=317935 Log: Fixed Bug #55554 (Legacy constructors not handled properly) [TRAITS] [DOC] # The handling of legacy constructors defined by traits was corrected. # They are now properly registered and used on instantiation. # The situation for conflicting legacy and __construct constructors is # mostly identical. If they are defined in the class, they override conflicts # and do not collide. However, in case different styles are mixed, between # class and trait definition, we assume a programmer's mistake and report # a collision. # # BTW: +1 for all the fixed tests! `make test` is fun again. ------------------------------------------------------------------------ [2011-09-14 06:17:01] ryan at zuttonet dot com Thanks for the clarification; while that answer isn't very satisfying, it's sort of hard to argue when the problem is so deeply rooted in the language. Is this inconsistency in constructor handling a design decision or a fundamental error in the engine? Are there any plans in place to enforce consistency? If you're going to add those tests, then I guess there's not much left to do here. ------------------------------------------------------------------------ [2011-09-12 19:27:40] g...@php.net Hi Ryan: (I wonder whether I could answer directly to those emails sent be the tracker, this here manually is rather messy ...) My main point is, that PHP does handle classes that define both kinds of constructors in a very specific way. And that is consistent with how it is handled when traits are used. The answer to your question, whether a trait should be able to override the constructor of a class in which it is used, is no. The class definition should always take precedence. However, the problem here is deeper. Since legacy constructor and __construct are not considered equal, we got a problem. The inequality allows you to define both in a class, as my previous example showed. And well, for the same reason it also works with traits, of course with unintended side effects. I will see what pops up when I fix the issue that you can not define legacy constructors with a trait. I will probably add a test that constructors are handled more carefully in traits then. The WTF factor is much higher and not worth keeping the consistency, I think. Best regards Stefan ------------------------------------------------------------------------ [2011-09-12 17:38:08] ryan at zuttonet dot com Thanks, Stefan (the defect directions explicitly say not to include examples over 20 lines; that's why I provided the github link) The real issue is this: should we be allowed to define, in a trait, a method that will override the constructor of the class into which we're importing the trait? I think not, but currently it can be done, although not consistently. Example 1: <?php /** * ============ * Defect: can alias-override class constructor, but only when using legacy constructor naming convention * ============ */ trait SomeTrait { private function constructor() { echo "this is abuse\n"; } } class SomeClass { use SomeTrait { constructor as __construct; } public function SomeClass() { echo "this is a constructor"; } } $c = new SomeClass(); // Fatal error: Call to private SomeClass::__construct() In this example, I can override SomeClass::SomeClass() as the constructor by importing SomeTrait::constructor() as SomeClass::__construct(). However, this works only because I'm using the legacy constructor SomeClass::SomeClass(). If I had defined SomeClass::__construct() in SomeClass' definition, PHP would not allow this trait abuse. Example 2: <?php /** * ============ * Defect: can override class constructor, but only when using legacy constructor naming convention * ============ */ trait SomeTrait { private function __construct() { echo "this is abuse\n"; } } class SomeClass { use SomeTrait; public function SomeClass() { echo "this is a constructor"; } } $c = new SomeClass(); // Fatal error: Call to private SomeClass::__construct() In this case, I can override the legacy constructor SomeClass::SomeClass() with a method name SomeTrait::__construct(). However, if I had a method named SomeClass::__construct(), it would not be overridden by the trait method. Do my concerns make more sense now? It's not about what _should_ be allowed, it's about the inconsistency in what currently is allowed. In order to be consistent, either both types of constructors should be overridable, or neither should (probably the latter). Again, it's about the overriding of an existing constructor, not about the mixing in of one into a class with no constructor. ------------------------------------------------------------------------ [2011-09-12 10:47:22] g...@php.net Hi Ryan: I am sorry, I don't think I understand what inconsistency you are pointing out. Could you elaborate on the problem? (And please include the code you are referring to directly here. Just to make it easier for me to know that we are talking about exactly the same code.) What I understand is you want to provide the constructor with a trait, right? Like this: <?php trait MyTrait { public function constructor() { echo "Foo\n"; } } class MyClass { use MyTrait { constructor as __construct; } } class MyClass2 { use MyTrait { constructor as MyClass2; } } echo "MyClass constructor: "; $o = new MyClass; // echos Foo echo "\nMyClass2 constructor: "; $o = new MyClass2; // doesn't echo Foo echo "\n"; ?> The problem I see here is that for MyClass2 the constructor does not actually get registered as a constructor but just as a normal method. That seems to be an inconsistency that needs to be fixed. Ok, now to the new vs. legacy constructors: class Bar { function Bar() { echo "BarBar new ctor\n"; } function __construct() { echo "Bar new ctor\n"; } } $o = new Bar; ?> Switching the order of the definition of the constructors doesn't influence the result, __construct always wins. Both your examples behaves identical to the situation if the __construct would have been defined directly in the class. So, where is the problem here? It is not an inconsistency with how PHP behaves without traits, from what I can see. Ah, and please try the latest code in the SVN, I am not exactly sure whether I changed anything that could be related to that between alpha3 and now. But I doubt it. Thanks Stefan ------------------------------------------------------------------------ The remainder of the comments for this report are too long. To view the rest of the comments, please view the bug report online at https://bugs.php.net/bug.php?id=55554 -- Edit this bug report at https://bugs.php.net/bug.php?id=55554&edit=1
