Re: [Fwd: RFC GDL2: move +load implementations to +initialize]

[Adam Fedor]

On Friday, January 17, 2003, at 08:31 AM, Mirko Viviani wrote:

> David Ayers <d.ayers@inode.at> ha scritto:
>
> > > Why the EODatabaseChannel should have received the +initialize 
> > > before the
> > > notification of the EOObjectStoreCoordinator ?
> > > You should invoke it manually, it has no sense.
> > No, no, both +load and +initialize are guaranteed to be sent by the
> > runtime, not by any explicit code!
>
> I know...
>
> > In the very unlikely event that no messages get sent to
> > EODatabaseContext before the the EOCooperatingObjectStoreNeeded
> > Notification gets posted, we should just make sure to call some
> > EODatabaseContext method (like [EODatabaseContext class]) real early 
> > to
> > force the runtime to send +initialize to EODatabaseContext.
>
> The problem is really this... how to be sure that a message will be 
> sent to
> these classes ?

(1)
+initialize will always be sent to a class before it is used. If you 
are really paranoid, you could take one class's +initialize method that 
you know is being used first, and add a bunch of

[ThisClass class]
[AnotherClass class]
[Etc class]

and all those classes will be initialized also.

(2) Objective-C code should never be used in a +load method. A lot of 
basic Objective-C/Foundation code is not setup when the +load messages 
are sent (like NSString). Trying to send messages in +load can 
potentially cause a nightmare of recursive initialization.

(3) My personal preference is to not put any 'executable' code 
(Objective-C or otherwise) in +load OR +initialize methods. If at all 
possible, classes should be setup in a lazy manner, which avoids a lot 
of startup costs. Although I realize this is not possible in many cases.