On Mon, 28 Jul 2003, Andreas Raab wrote:

> If you're going for some fileout-like format (I interpret Ned's message that
> way) then I'd strongly recommend defining and using these stubs. In fact,
> I'd go as far as to say don't make them "MCClassDefinition" but rather some
> app-registry kind of thing. It tells people what interfaces they can rely
> on, it documents (for your own knowledge) what interfaces have been released
> and are used out there and other package/versioning systems could hook into
> the same places if they know how to interpret them. The "default stub" would
> then be the "empty package/versioning system".

Andreas,

Would this be better, in your opinion, than simply using chunk format
directly?  Or the VW XML fileOut format, which is a little more flexible?
Or Rosetta, for which there is XSLT to convert to chunk format?

What makes me a little uneasy about this suggestion is that it seems to
make it harder to have multiple packaging systems loaded at the same time.

The one thing that I do like about it is that it allows us to extend what
we think is a valid model for packages without being tied to the
limitations of other formats.  For example, many formats do not have a
good way of representing the extension of an existing class with a new
instance or class variable - Tim was complaining about that just the other
day when you did this recently and the changeset included the entire class
definition...

If we got our tools to the point that we could properly mark an instance
variable as belonging to a separate package from the rest of the class, it
would certainly be nice to be able to easily introduce an
InstanceVariableDefinition interface (of course, how "easily" this can be
done depends on how many packaging systems need to be updated to support
it...).