[RFC] Patience diff & merge?

[Aaron Bentley]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Martin Pool wrote:
| I would like to do #3, though I'm not sure if it's a totally safe
| change to do before 0.8 release.

Given that our deadlines are fast-approaching, I can keep it on ice
until 0.9 opens up.  Good to have a sense of direction, though.

| Perhaps we can get more confidence  by
| doing some stress tests on many inputs to make sure that the diffs  are
| at least correct, if not optimal.

Hmm.  Like by running the revision history of bzr through patch, and
seeing if it gets anything wrong?

| One of the previous diffs used difflib to try to reduce the  mismatched
| areas produced by patience diff.  Do you still do it?

I haven't changed that.

| Are  you sure it's a good idea?

I think it is.  It plasters over a weakness in patience diffs that
causes extra mis-match when there are no unique lines.

Theoretically, if one line can be a unit of meaning, then surely two or
more lines can be a single unit of meaning.  So it would be interesting
to have patience try pairs of lines, then triplets of lines, etc, until
it matched everything that could be matched.

But practically, the results of using standard difflib in combination
with patience should be nearly as good, and certainly better than using
either in isolation.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFEOvPu0F+nu1YWqI0RAgIpAJ9kod1kn2pAsSJjeog3CItvGGwwYACfXnI5
gb4uQ7TrnD54+BXCvngvvAc=
=jgHf
-----END PGP SIGNATURE-----