Yet another diff extension

[Matt Mackall]

On Wed, Jun 22, 2005 at 01:53:47PM -0400, Chris Mason wrote:
> On Tuesday 21 June 2005 19:09, Matt Mackall wrote:
> >
> > This was causing a segfault about 80% of the way through my kernel
> > test suite, on generating the _reverse_ of the delta that's actually
> > checked in. With this fix, the entire kernel repo is passing again.
> > Please try it against your test case.
> >
> > Here's a current version:
> >
> Just FYI, the speeds of the bdiff in tip are still better than difflib, but 
> 20% slower than the buggy version (before the hashing changes).
> 
> I would suggest calling it good enough for now, but you might want
> to revisit things after the next release.

Argh. Ok, I've tweaked it a bit more. 

Also the tip now uses bdiff for normal delta generation and annotate.
It's still using difflib's SequenceMatcher for generating unidiffs
which I'll deal with eventually.

-- 
Mathematics is the supreme nostalgia of our time.