On Wed, 2004-05-26 at 06:51, Branko Èibej wrote:
> We don't use sliding windows, we use sequential adjacent windows. This
> error is probably historical: Once upon a time, we used sequential
> _overlapping_ windows instead, as that tended to reduce the size of the
> delta (often by some 10 percent). But overlapping windows would have
> complicated the delta combiner enormously
You're only considering deltas in the FS here.
> , so we moved to adjacent ones
> and put in a check to make sure that they really are. Well. A window
> "slides back" if it starts before the end of the previous window; at
> least, that's what this message is trying to say.
You're only considering deltas in the FS here. We could still use
overlapping source windows for deltas sent over the net. (We don't, but
we could.)
You're also factually inaccurate. The code only checks, as it always
has, that S2start >= S1start and S2end >= S1end, if S2 follows S1.
> I'm almost of the opinion that we can ditch this error message
> completely, because it signals a bug in the code rather than an error
> condition. We should be aborting or assering instead.
svndiff data often comes from an untrusted source. We can't abort on
invalid data from the other side of the network. (We also can't abort
on invalid data from a trusted source, like a repository file, although
it's less of a big deal if we do.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Wed May 26 17:57:59 2004