update -r does a lookup in whole-branch history

[John A Meinel]

The current WorkingTree._update_tree code has this clause:

        if revision is None:
            revision = self.branch.last_revision()
        else:
            if revision not in self.branch.revision_history():
                raise errors.NoSuchRevision(self.branch, revision)

Which has a couple issues:

1) I don't really know why we refuse to allow the revision to not be in the mainline of the branch. I don't see a specific problem doing "bzr update -r 1.2.1" for example.
2) Perhaps we meant to check if it was in the ancestry at all? I'm not sure that doing "bzr update -r revid:XXX" (for XXX not in the ancestry) needs to be forbidden, though.
3) Regardless, using 'revision_history()' for this requires reading the entire lefthand ancestry, and for something like emacs, this can be 100k revisions. If we want to keep the left-hand check, we could just do:
  revno = self.branch.revision_id_to_revno(revision)

Which will raise NoSuchRevision for us if the revision is not in the lefthand ancestry. I don't think we have a quick shortcut for the non-mainline case, though revision_id_to_dotted_revno() will check if it is in the ancestry, otherwise raising a NoSuchRevision exception, and it will probably be the fastest method to use. (Alternative is to use graph.iter_ancestry() and iterate until the revision is found, but that may not be super efficient anyway, and doesn't use the fast-path for loading index ancestry.)

[Martin Pool (mbp)]

On 5 February 2010 21:40, John A Meinel <email address hidden> wrote:
> 1) I don't really know why we refuse to allow the revision to not be in the mainline of the branch. I don't see a specific problem doing "bzr update -r 1.2.1" for example.
> 2) Perhaps we meant to check if it was in the ancestry at all? I'm not sure that doing "bzr update -r revid:XXX" (for XXX not in the ancestry) needs to be forbidden, though.

I don't see any good reason to forbid it either.

--
Martin <http://launchpad.net/~mbp/>