bzr.dev rsync copy is broken?

[Martin Pool]

On 28 Mar 2006, Matthieu Moy <Matthieu.Moy at imag.fr> wrote:
> "Erik Bågfors" <zindar at gmail.com> writes:
> 
> >> Indeed, a simpler semantics would be:
> >>
> >> If there is a working tree, update it.
> >> Otherwise, don't update it
> >
> > I think the problem is that it's very expensive to update the working
> > tree on a none local location.
> 
> Yes, but if it is too costly to update the tree, then why would you
> want to have one anyway? And with my proposal, if you don't have a
> tree, you don't pay the penality.
> 
> Maybe it can be interesting to have a --dont-update-tree flag for
> people who explicitely want to run "bzr update" on the server.

Yes, now that you can have branches with no trees it would make sense
to update the tree if it is there.

> > Each file has to be read to verify that it hasn't been updated in
> > the tree,
> 
> This is not true. You can compare the inventories, and then take into
> account only the files which have been modified by the push (which
> you'll have to read and write back).

A good first step towards this would be to change the workingtree
methods to always go through a Transport object, rather than direct to
disk.  The merge code in ttransform could also be changed to do this, if
it doesn't already do so.

-- 
Martin