Req #64454 [Com]: It is possible to directly call the magic method "__construct".
Edit report at https://bugs.php.net/bug.php?id=64454&edit=1
ID: 64454
Comment by: nikic@php.net
Reported by: wolters dot fl at gmail dot com
Summary: It is possible to directly call the magic method
"__construct".
Status: Wont fix
Type: Feature/Change Request
Package: Class/Object related
Operating System: Microsoft Windows Version 6.2.9
PHP Version: 5.4.13
Block user comment: N
Private report: N
New Comment:
1. parent::__construct() is not a static method call, its a scoped call. It only says: Do
$this->__construct() in the "parent" context.
2. Could you explain why it is a problem? What issues are there with calling __construct() from
outside the class? I don't see any. I don't see much use either. One possible use would be
calling it in a context where the constructor is not called by itself, e.g. one could call
__construct() after an unserialization or something.
Really, I don't get why you want to forbid this. We don't usually prohibit direct calls to
magic methods and I don't see why we should do so in this particular instance. Quite honestly,
I more wonder why the call to __clone() is not allowed.
Previous Comments:
------------------------------------------------------------------------
[2013-03-22 16:19:05] wolters dot fl at gmail dot com
Yes, I do understand that "__construct" is an interceptor method and can be called from a
subclass directly via "parent::__construct", but:
1. "parent::__construct" is a "static" method call which can occur inside of a
class only.
2. The problem is, that it is possible to call "$obj->__construct" from outside a
class. I could understand the behaviour if the call would be allowed from the inside of a class
only.
Imho, calling "__construct" directly should ONLY be possible from the INSIDE of a class.
Plus, such a call MUST be STATIC (otherwise a PHP error is raised).
From a clients PoV, __construct is a member method (similar to __clone):
* __clone is invoked via the "clone" operator.
* __construct is invoked via the "new" operator.
I don't see the reason, why there is a difference in that behaviour. I do not know much about
the internals (C code) of the interceptor methods, but from the clients PoV these distinctions in
behaviour are confusing.
PS: I also updated my example at https://gist.github.com/FlorianWolters/5195734
to emphasize this.
------------------------------------------------------------------------
[2013-03-22 07:03:54] nikic@php.net
Additionally to what laruence said, one should also point out that __construct is quite commonly
called directly in the form of parent::__construct().
------------------------------------------------------------------------
[2013-03-22 04:18:32] laruence@php.net
PHP's __construct, actually, is not creating object, it decorate the object after
it was created.
so, even you directly call to $obj->__construct, then no new instance was create.
so a workaround is you can exam some property you set to see whether that is a
double constructing or not.
thanks
------------------------------------------------------------------------
[2013-03-19 12:45:54] wolters dot fl at gmail dot com
Description:
------------
It is possible to directly call the magic method "__construct". This should NOT be
possible, since one cannot create immutable objects with the current behaviour.
In addition, a direct call to the magic method "__clone" is not possible and raises an
"\E_FATAL". Both magic methods deal with object creation, so the behaviour is inconsistent
between them.
I am using "php-5.4.13-nts-Win32-VC9-x86" on Microsoft Windows Version 6.2.9 x64(Windows 8
64 bit).
Test script:
---------------
https://gist.github.com/FlorianWolters/5195734
Expected result:
----------------
The method call:
$example->__construct();
should raise an "\E_FATAL" with the following message:
"Cannot call __construct() method on objects - use 'new <Class>' instead in
[...] on line 21"
------------------------------------------------------------------------
--
Edit this bug report at https://bugs.php.net/bug.php?id=64454&edit=1
Thread (5 messages)