Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:20254
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Date: Tue, 22 Nov 2005 23:54:36 +0100
Reply-To: Marcus Boerger <helly@php.net>
Message-ID: <797358253.20051122235436@marcus-boerger.de>
To: Matthias Pigulla <mp@webfactory.de>
Cc: Rasmus Lerdorf <rasmus@lerdorf.com>,
	internals <internals@lists.php.net>
In-Reply-To: <00A2E2156BEE8446A81C8881AE117F192C1C4F@companyweb>
References: <00A2E2156BEE8446A81C8881AE117F192C1C4F@companyweb>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Subject: Re: AW: [PHP-DEV] PDM Meeting Notes
From: helly@php.net (Marcus Boerger)

Hello Matthias,

Tuesday, November 22, 2005, 10:29:09 PM, you wrote:

>> I don't see why you can't specify that a class definition must have a 
>> constructor.  Obviously the constructor is not for the 
>> interface itself.

> Ok, that is, having __construct in the interface asserts that everything
> you get passed (as an implementation of the interface) has been
> constructed by a constructor that has a certain signature? :--/ Of
> course it also doesn't make sense to call __construct on the
> implementation passed along.

> Luckily this is nothing I'm forced to write in my code ;).

> I only wanted to utter I have a Bad Feeling(TM) if you invest time and
> effort to make stuff like this work if the result is something that
> seems to make no sense [to me only?]. And once you allow it, be sure ppl
> out there use & abuse it and will complain should you ever have to
> remove it again.

The patch is ready already and actually it is a single line change that was
obvious from the first place. What really causes work here is explaning this
over and over again.

> I'm just trying to make up a scenario where the above might make sense -
> it could be where you have something like a factory method. It has a
> type hint on it's argument to make sure what gets passed in implements a
> certain interface.

> That interface defines the signature of __construct, so the method can
> construct instances of the "thing" - only knowing that the instances
> constructed will implement the interface (?), but not knowing their
> implementation. 

> But how should it construct instances? new INameOfTheInterface()? And
> besides that, problem is again that the "thing" passed in would have to
> be something representing a class, and not an instance of the class
> itself. You could always (gut feeling) re-design this to work "the usual
> way" by passing in something representing a class and providing
> behaviour to construct instances.

The typical way out is to completley re-design the software once reached
this need or change to reflection to at least prevent from more severe
errors.

Best regards,
 Marcus