On Mar 4, 2004, at 1:03 PM, ducasse wrote:

> Sorry it does. I know what you mean but to be correct any class 
> reference to another class creates a dependency.
> Now you are right we have tool to manage them and I forgot to mention 
> that.

To be "correct" by whose definitions?  In any reasonable definition of 
"dependency" for Smalltalk, class extensions should not add extra 
dependencies to the class itself.  I remember looking at Roel's prolog 
code (in VW) for detecting dependencies between packages and I'm pretty 
sure it agreed with that (in fact I based 
PackageInfo>>browseExternalReferences on his rules).

> I checked here is the answer: So this was before PackageInfo was 
> introduced I guess.
>
> 'From Squeak3.4 of 1 March 2003 [latest update: #5170] on 28 March 
> 2003 at 4:31:22 pm'!
> "Change Set:		KCP-0019-rmBrowseInBehavior
> Date:			28 March 2003
> Author:			stephane ducasse, alexandre bergel, and nathanael schaerli

Yes.  PI went into the base in 5250.  But why it happened isn't 
important, what's important is that we recognize the error and fix it.

> So may be I want to go too fast but why can we replace class 
> categories by first class packages?
>
> This is fun to see that in the realm of everything is an object we are 
> still relying on idiot string and dummy folder
> without any behavior.

We'll get there, we'll get there.