[darcs-users] unique features + exponential time issue

Petr Rockai me at mornfall.net
Fri Oct 26 10:23:54 UTC 2007

Stefan Monnier <monnier at IRO.UMontreal.CA> writes:
>> Note that the "inexact" version of 3-way is imprecise, in that it
>> depends on few context lines around the patch. It therefore leads to
> AFAIK diff3 has no "inexact version" of 3-way merge.
Yes, sorry, diff3 apparently uses three-way-compare, not exact nor
inexact patching (subversion uses exact patching, or plans to use it,

>> So diff3 produces a conflict, although a clean (and correct) merge is
>> very much achievable.
> Yes, darcs may avoid conflicts.  That's good.  But what I objected to was
> the claim that darcs was safer (i.e. correctly signals conflicts at places
> where diff3 would have incorrectly "resolved" it).
Yes, what i have thought you meant was guessing in general, not to
flag conflicts diff3 wouldn't notice (diff3 notices conflicts way too
often, IMHO). However, if you treat the "huge conflicts" problem as
"not noticing conflicts" (the actual conflict is obfuscated), then
that is a problem as well. (This is basically the same issue as
insisting on "not losing any mail due to counter-spam measures", where
people neglect that when there is too much spam in their inbox, they
will lose precious mail by not noticing it in the flood).

> I've never heard od diff3 using "context".
Yes, that was my confusion, since i first thought diff3 is based on
inexact patching (patch this with diff(base,other)), whereas it is
probably based on three-way compare.

However, for the original ("other scm's") context, git's "rebase" may
be an inexact 3-way, in that it just takes diffs and applies to a
different context semi-blindly (which can cause incorrect placement of
hunks). And rebase is neccessary due to weaknesses of 3-way
merge. Well, thinking of that, rebase looks like poor man's
guess-commutation. Unless it does real commute, but i somewhat doubt
that... Of course i may be wrong, since i do not quite desire studying
git's source code.

Yours, Peter.

Peter Rockai | me()mornfall!net | prockai()redhat!com
 http://blog.mornfall.net | http://web.mornfall.net

"In My Egotistical Opinion, most people's C programs should be
 indented six feet downward and covered with dirt."
     -- Blair P. Houghton on the subject of C program indentation

More information about the darcs-users mailing list