On Wed, 20 Aug 2003, Craig Latta wrote:

> 	I'd rather not see that idiom proliferate. I'd rather see some way of
> uniquely identifying a method, and looking up further information about
> it. The only things I'd want to attach to the method itself would be the
> bare minimum needed to do the identification.

I agree with you in principal, I think - but from a practical point of
view, this was the lightest-weight and most backwards compatible solution
I could find, because it leverages all the existing tools.  Right now
there are two kinds of information you can easily specify about a
particular class/selector pair:  what category it is in, and what its
source code is.  Anything else and you run into problems with being able
to edit, fileOut, etc, the metadata.

I tend to take an incremental approach to things like this: first, provide
the functionality in the simplest way possible and get people using it,
then bring the toolset up to snuff and get rid of the hack you used to
bootstrap with.  PackageInfo is a perfect example of this: it's a
laughably simple and naive way of specifying package information, but it
got people actually marking their class extensions in a consistent way -
such that if/when we come up with something better, it'll be trivial to
convert existing packages to it.