Robert Collins wrote:
> 
> On Thu, 2008-02-07 at 16:14 +0300, Dmitry Vasiliev wrote:
> > Very often for my projects I use the following command's sequence:
> > 
> > $ bzr pull URL
> > ... some changes, commits
> > $ bzr push
> > 
> > At the 'bzr push' point I'm almost always forget specify the push URL 
> > which is the same as the parent URL, I guess mainly because there is 
> > some reflections with 'svn ci' for me. The attached patch uses the 
> > parent branch location if no explicit (or saved) location defined.
> 
> bb:comment
> If you have this pattern of working in a single branch, why not use 'bzr
> checkout' ?
> 
> When I make a new branch it always goes somewhere different than the
> place I branched from, so this default would be at best irritating.

I second Robert's concern.

The current behaviour works very well for the workflow of:

    bzr branch base new-branch
    cd new-branch
    ... hack hack hack, commit ...
    bzr push

At this point, I've usually already set a push_location (with
push_location:policy = appendpath) in my locations.conf, so this Just
Works.  I
wouldn't want this behaviour to change.

If I haven't set a location, then bzr will say so (bzr: ERROR: No push
location
known or specified.), so this first time I need to give bzr push a
location, and
then it'll remember it.  For me, I like this behaviour, as I usually do
want to
specify a new location rather than push back to the location I branched
from.
(If I wanted to simply push back to the parent location, and it was
writable, I
probably would have just done my hacking there in the first place...)

Still, if this is a common workflow for you, it'd be good to find a way to
make
it easier for you without compromising the common “make lots of feature
branches” approach.  Perhaps we could teach "bzr push" to recognise some
revisionspecs, so that you could do "bzr push submit:".  Or provide a way
for
you to configure your locations.conf so that like for me, it automatically
does
what you want most of the time.

-Andrew.