bazaar crash

Martin Pool mbp at
Mon Jul 30 19:57:50 BST 2007

Hi Bernhard,

Sorry you hit this.

Is this, as the path suggests, a checkout of a svn repository?

On 7/30/07, Bernhard Mulder <bwmulder at> wrote:
> F:\bwm\svn_checkouts\gcc>bzr bundle-revisions > c:/bwm/gcc.dmp
> Using saved location: F:/bwm/svn_checkouts/gcc/

It looks like this is using the same branch as the parent, which will
mean it sends no revisions (if it succeeded).  You should work out
which branch you're trying to merge into, and then do

  bzr bundle --remember THAT_BRANCH > gcc.dump

For Bazaar developers:

This shows one of the downsides of implicitly fetching during commands
like bundle. It's not just an optimization to cache the compared-to
revisions, but a hard requirement.  bundle is conceptually readonly
but i think you couldn't bundle between two physically readonly repos.

It looks like _double_lock is being passed the same object for source
and destination.  I think this traceback shows they're the same
inmemory object, not aliases.  There are a few things we could do

 * allow getting a write lock on an object that's already read-locked,
so that double lock as currently done will just pass
 * Add an inter optimiser for InterSameObject which knows that eg
fetch() doesn't need to do anything.
 * Having cmd_bundle specifically notice when parent==this and so the
bundle is pointless (unless -r is given?), which would help people
understand how to use it.  At this point that would be better added to
cmd_send I guess.


More information about the bazaar mailing list