> > except it's not at all trivial to generate.
> 
> Eh?   Just make cvs do it!   As long as you don't do manual surgery on
> the cvs archive, you should be fine.

right, and how do you do it?

"cvs diff -Dnow" vs the stuff that's in the ftp area?

	that doesn't work, because of pathname problems (basically,
	cvs 'diff' only gives you the trailing component).
	Additionally, it's not particularly reliable.

"cvs rdiff diff -D'1 day ago' -Dnow"?

	not reliable.  i suppose one could be very careful, and keep
	exact track of when the last diffs were generated, but i don't
	trust that as very reliable either.

The only way i can see to make it reliable is to tag the tree,
update the tree, rdiff the tree (between the tags), then delete the
tags.  That's non-trivial, and it also involves writing to every
archive file in the tree, a couple of times.

in any case, running a cvs rdiff is decidedly non-trivial, in terms
of the resources it requires (i.e. disk access and CPU time).


basically, _if_ a service like this were to be made available, it
would have to be at least as reliable as SUP, and it's not clear how
we can provide that reliability and make reasonable demands on lamp
and the source tree.  (i.e. i think that doing the 'reliable' scheme
mentioned above every night, or even every week, is unacceptable.)



chris