[darcs-users] What if darcs is generating too inclusive of hunks?

Yuval Kogman nothingmuch at woobling.org
Sun Nov 12 17:06:54 UTC 2006


On Sun, Nov 12, 2006 at 08:31:52 -0800, David Roundy wrote:

> In your example you left out a + on the space between the two subroutines.
> But it does look like in this example darcs doesn't generate a true LCS, as
> even with the added line, darcs is generating a non-minimal diff.  This
> would be a result of a performance improvement, which was added because a
> true LCS takes O(N^2) time (which is problematic for people with
> multi-megabyte text files).  Perhaps we should have a heuristic that allows
> us to use a true LCS on smaller files.  (Note that if there was a
> whitespace change in the _generate_accessor_method line, then darcs did
> make an optimal choice).  The algorithm darcs now uses, by the way, is the
> same one used by GNU diff.  (But not original diff, which used a true LCS,
> but scaled worse than GNU diff.)

I would really like the possibility of forcing real LCS or fake LCS,
as well as a middle ground specified using file size for my cases.

There was no whitespace change, for the record.

> There's also often an ambguity in deciding which lines were added where,
> and there's no way to determine what the user means.  We've tried to use a
> heuristic to (among equivalently minimal descriptions of a change) select a
> reasonable change, but it's tricky, as the "best" choice is often very
> different, depending on 

Perhaps the interactivity of darcs can be used here? (Allow the user
to have an option to try to cycle through the computed changes on a
per file or even  per hunk basis).

-- 
  Yuval Kogman <nothingmuch at woobling.org>
http://nothingmuch.woobling.org  0xEBD27418

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.osuosl.org/pipermail/darcs-users/attachments/20061112/5d815678/attachment.pgp 


More information about the darcs-users mailing list