Another problem with the current fileout format

Stephane Ducasse ducasse at iam.unibe.ch
Sun Jun 29 18:33:28 UTC 2003


Hi avi,

> I don't store changesets in CVS, in any format.  I only store
> representations of complete packages.  So you don't explicitly mention
> removals, you implicitly mention them by not having that class 
> declaration
> exist anymore (and the tools for loading these representations figure 
> out
> that what in fact needs to be done is a removal, if that class is 
> still in
> the image but not in the fileout you're loading).
>
> The representations themselves work exactly as you're describing, ie, 
> what
> I store looks like
>
> MCClassDeclaration
>   name: #Foo
>   superclass: #Bar
>   ...
>
> so that you get an object representing the declaration, *not* actually
> create a class when you evaluate it.

:)

I can easily understand that for historical reasons, the chunk format 
was a
good answer for storage of smalltalk code. Now more than 30 years later
we can fix the problems of these stupid strings that prevent us to build
better tools.

I like what you do in Monticello. This is the way to go this way you can
manipulate any code entity.


More recently I also wrote a VW5PackageExporter, which uses the newer 
XML
format, puts in namespace and package information etc.  This version
didn't make it into the image, but if you need it I can point you to it.

No thanks I generate chunk format and trap error, analyse the stack to 
create superclass
on the fly to avoid to have to linearize inheritance. My point was more 
general.



More information about the Squeak-dev mailing list