[MERGE] Fixes for 'bzr annotate'

[Martin Pool]

On Sat, Oct 18, 2008 at 3:18 AM, John Arbash Meinel
<john at arbash-meinel.com> wrote:
> There are 2 simple bugs in 'bzr annotate' right now.
>
> It seems that Martin didn't actually restore the fast-path when we were
> sorting out the issues with stacking. He accidentally left in a shortcut
> with:
>
> if True or len(self._knit._fallback_vfs) > 0:
>
> This made a significant difference in 'bzr annotate' time for files that
> are modified often. (56s down to 16s on my test file.)

Well, thanks for catching it, and I apologize for putting it in.  I
think it shows: we (or I at least) should be more careful in review;
we possibly rushed the stacking code in a bit too soon; and we should
get more of these cases into usertest and make sure that any changes
will be noticed.

> I'm still trying to figure out why 'annotate' gives a wrong result when
> the parent_ids are swapped. I don't see a particular reason for it in
> 'bzrlib.annotate.reannotate()' it should be order independent.

That would be very good to find out.

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