About KCP and automatic initialize

ducasse ducasse at iam.unibe.ch
Mon Sep 15 19:33:32 UTC 2003


On Lundi, sep 15, 2003, at 20:52 Europe/Zurich, Colin Putney wrote:

> Hi all,
>
> After reading all the posts on this topic with interest, here what's 
> beginning to become clear to me. Although there would be some benefit 
> in having a standard pattern for initialization, it wouldn't be great 
> enough to justify the chaos of migrating to it.

What chaos?
Why do you call that a chaos?  I really do not get it. I have a similar 
changes in my image since two years now and
this is not a problem.

> Making this change would create an *enormous* number of bugs. A lot of 
> them would be easy to find and fix, sure. Others will be more subtle. 
> And this is pretty much an "all or nothing" change. Unlike, say, the 
> network rewrite, you can't have both mechanisms in place at the same 
> time and gradually move things over.

But if a class does not call initialize then it is not call after. Are 
there classes that define initialize but call it
in other ways than via new in the image?

> During the transition period, we'd have a lot of really ugly code 
> written by people who want to have their package work in both the old 
> system and the new system. Remember all the little work-arounds to 
> avoid deprecation warnings? This will be worse:
>
> new
> 	^ self systemSendsInitialize
> 			ifTrue: [super new]
> 			ifFalse: [super new initialize]
>
> <shudder>

how many packages really have to do that: 2000 or 3000 or simply may be 
3 or 4?

> Let's not go there.

I should say that I admire your dramatic way of presenting the 
situation.

>
> Colin
>
>



More information about the Squeak-dev mailing list