Re: [Gnu-arch-users] Re: [PATCH] Re: bazaar bug #385 (Re: Stabilizing Bazaar?)

[Robert Collins]

Sorry about the two copies Robert, I keep using the from from: on this list :p.

On Wed, 2005-07-06 at 22:48 +0200, Robert Widhopf-Fenk wrote:
> On Monday, July 4, 2005 at 23:24:45, Matthieu Moy wrote:
> > Robert Widhopf-Fenk <address@hidden> writes:
> > 
> > > My problem comes from a call to arch_archive_connect_branch
> > > within arch_build_revision.   It was called for an archive
> > > which I did not register and thus returned NULL.
> > 
> > Then, it was not a real "bug", but a bad error message.
> > 
> > I've just comitted a patch to issue an explicit error message in
> > your case, and "exit(2)".
> > 
> > Is it OK?
> 
> No, IMHI it IS a bug.
> 
> Merging from archive A must NOT require access to archive B. 
> While baz might use B to find a better merge point/strategy
> it must not require access to B, since this effectively prevents 
> merging from A when B is gone.

The error you saw occured when building a revision - nothing to do with
ancestry information. There are three revisions built to do a three way
merge - your revision, the revision you are merging from and the parent.
The merge logic automatically strips out parents that are on unreachable
or unregistered archives (or its meant to). Did it fail before or after
showing  'calculating best merge point' ?

If it was before, then you have a tag with no cacherevs and baz (and
tla) would have no choice but to require archive A because there was
insufficient data in archive B.

Cheers,
Rob

-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.

signature.asc
Description: This is a digitally signed message part