Bug #73987 [Opn]: Method compatibility check looks to original definition and not parent

Edit report at https://bugs.php.net/bug.php?id=73987&edit=1 ID: 73987 Updated by: ab@php.net Reported by: requinix@php.net Summary: Method compatibility check looks to original definition and not parent Status: Open Type: Bug Package: Class/Object related PHP Version: 7.1.1 Block user comment: N Private report: N New Comment: It's clear, that there's no method overloading of the methods with the same name in PHP, but the suggested way will in fact disallow even what works today. Shouldn't the interface be defined explicitly in this case? As semantically the examples look not controversial, in other programming languages with stronger OOP, even more advanced examples are possible. As "B extends A" doesn't imply "B implements I" indeed. Thanks. Previous Comments: ------------------------------------------------------------------------ [2017-01-27 09:13:13] mail at pmmaga dot net I had a go at fixing this issue. Please review the PR. Thanks ------------------------------------------------------------------------ [2017-01-24 14:29:36] requinix@php.net Related To: Bug #73985 ------------------------------------------------------------------------ [2017-01-24 14:24:53] requinix@php.net Description: ------------ Originally spotted as bug #73985. When PHP validates method signatures for compatibility, if a method is defined in an interface then compatibility is measured against the interface and any parent method's signature is ignored. This leads to inconsistencies... Example #1: method is defined in an interface (valid) - https://3v4l.org/pblqI ---------- interface I { public function example($a, $b, $c); } class A implements I { public function example($a, $b = null, $c = null) { } // compatible with I::example } class B extends A { public function example($a, $b, $c = null) { } // compatible with I::example } Example #2: method is not defined in an interface (invalid) - https://3v4l.org/bJJ1h ---------- //interface I { // public function example($a, $b, $c); //} class A { public function example($a, $b = null, $c = null) { } } class B extends A { public function example($a, $b, $c = null) { } // not compatible with A::example } The same signature is used in A and B, however only the second has a problem. The first is easy to explain on its own ("example" was defined in I so methods must be compatible with I::example) and the second is easy to explain on its own ("example" was defined in A so methods must be compatible with A::example) however the two together are inconsistent. The problem appears when calling a function using an I or A parameter type - https://3v4l.org/OneV3 --- function accepts_i(I $i) { $i->example(1, 2, 3); } accepts_i(new B); // no problem function accepts_a(A $a) { $a->example(1); } accepts_a(new B); // problem --- PHP <7.1: missing argument 2 for B::example PHP >=7.1: ArgumentCountError: Too few arguments to function B::example, 1 passed A::example() only has one required argument, therefore it should be safe for accepts_a to call ->example(1). But it isn't. Like with method parameters, this problem also exists for return types. Example #3: interface does not have return type (valid) - https://3v4l.org/Jik7I ---------- interface I { public function example(); } class A implements I { public function example(): int { } // compatible with I::example } class B extends A { public function example(): string { } // compatible with I::example } Example #4: class has a return type (invalid) - https://3v4l.org/n1q0G ---------- <?php //interface I { // public function example(); //} class A { public function example(): int { } } class B extends A { public function example(): string { } // not compatible with A::example } Like with method parameters, this can result in unexpected behavior. Unlike with method parameters, there's no warning about it - https://3v4l.org/TUbfJ --- function accepts_i_expects_any(I $i) { var_dump($i->example()); } accepts_i_expects_any(new B); // receives string, no problem function accepts_a_expects_int(A $a) { var_dump($a->example()); } accepts_a_expects_int(new B); // receives string, problem --- Proposed solution: methods in a subclass are validated against methods in the nearest ancestor who (re)defines the method - be that normally, as abstract, or using a trait. The result is that since A implemented example(), B::example gets validated against that rather than the original definition of I::example. BC: Yes, but code reliant on current behavior is susceptible to the "inconsistencies" noted earlier so it's already flawed. Test script: --------------- <?php // https://3v4l.org/gl1Gt interface I { public function example($a, $b, $c); } class A implements I { public function example($a, $b = null, $c = null): int { } } class B extends A { public function example($a, $b, $c = null): string { } } ?> Expected result: ---------------- Error that B::example is not compatible with A::example, due to the return type (a fatal error by itself) and the second required argument (a warning by itself). Actual result: -------------- No error(s). ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=73987&edit=1