[Catalyst] setup() called twice

[Matt Lawrence]

Matt S Trout wrote:
> On Tue, Jun 19, 2007 at 06:19:30PM +0100, Matt Lawrence wrote:
>   
>> Matt S Trout wrote:  
>>     
>>> Still completely broken.
>>>
>>> Read half a dozen plugins' setup methods and come back when you have a 
>>> clue.
>>>       
>> Nothing like a bit of random abuse to spark off a bit of development work...
>>     
>
> Shame it didn't spark off you actually going and reading what I suggested you
> did before you wasted your time and the list's.
>   

Well I can't say I didn't ask for that.

> Tell you what, I'll read the setup methods for you and spell it out:
>   
> Now do you understand why your approach isn't going to work?
>
>   

Admittedly my first suggestions were naive to say the least, and looking 
back now they are unnecessarily terse. I didn't intend to seem 
confrontational, it just seemed odd at first glance that there was a 
known set of methods that needed to be called, but because of the MRO 
unwanted side-effects could occur.

For what it's worth, the patch I sent actually does respect the MRO, it 
calls the first defined setup method in the plugin list directly, 
bypassing MyApp. I only realised on the tube home that it would cause an 
infinite loop with real plugins (as opposed to the ones in the test 
suite I now see don't call NEXT::setup). In my haste I didn't clock that 
the plugins necessarily appear before Catalyst in the MRO and I imagined 
that the tests were passing because Catalyst::setup was being bypassed 
too. Once you replace the temporary setup no-op, it is pretty much 
functionally equivalent to redefining MyApp::setup.

It strikes me that there's another case where a setup method could be 
called twice: if the application's base class is a Catalyst subclass 
with an overridden setup. I guess this isn't particularly likely, though.

Anyway, apologies all round for half-baked suggestions. I'll return to 
lurking..... Now.

Matt