[PATCH] Use merge3 for text merging

[Martin Pool]

On 27 Jul 2005, Aaron Bentley <aaron.bentley at utoronto.ca> wrote:

> Okay, if you can supply an API for Merge3 to produce rejects, then I'll
> change my mind, and say Merge3 should return a data structure describing
> conflicted and unconflicted regions.  It would be nice if that data
> structure had methods to produce rejects and a THIS+unconflicted changes
> text, as well as a method to produce a diff3-style conflicted file.

OK, we can do that fairly easily.  Do you think we need it now or is it
just nice-to-have?

> Deleting all three is annoying to me, and resolve --all is nice, too.
> Plus I can imagine using it for other types of conflicts, too.

I've been using 

  bzr unknowns
  rm -v $( bzr unknowns)

to get rid of them.

> I'd be more inclined to follow baz's more limited model, where conflicts
> are handled specially.  At least, in the Arch 1.0 model, 'unrecognized'
> is just generally nasty.  Probably not so bad if the layering's
> reasonable (i.e. if you add a forbidden file, it becomes a versioned
> file) and if it only affects commits/merges ('unrecognized' prevents
> *tla undo* from working...).

That would make them better than arch, but you're right, it's probably
still unrecognized complexity.

-- 
Martin