[RFC] I want to disable submit_branch on my computer for all branches. How can I do that?

[vila]

>>>>> Alexander Belchenko <bialix at ukr.net> writes:

    > For me *personally* submit_branch location stored in the
    > branch.conf in 99% of cases is the wrong thing or simply useless,

For me it's right in 90% of the cases, still the remaining 10% have been
painful on occasion.

    > and because it affects what bzr-explorer tries to show me in
    > submit pane, I want it disabled at all.

Well, you want it to be *right* :)

<snip/>

    > submit_branch could be useful only for feature branches, but in
    > the most cases submit_branch is the same as
    > parent_branch.

I think the truth is that it *defaults* to parent_location if it's not
set.

    > Therefore it's should be enough to use just parent_branch.

    > The problem here is that submit_branch only set by merge,

That's indeed the source of the problems.

    > but used by send command (and in the explorer).

And, more importantly for me, used to implement '-rsubmit:' ! 'bzr diff
-rsubmit:' always shows me what I will propose (submit) to the branch
I'm targeting.

This is really different from the parent branch (while identical in many
cases) and is also a useful default source for merge (in many
workflows).

    > And this is confusing a bit. If merge remember the last location
    > where from it merged first(last) time why such location is not
    > named as merge_branch or merge_from_branch?

Indeed. I think it should be named this way but in this case some use
cases of submit_branch will break (so we probably want to add
merge_branch and fix the fallouts).

    > I don't use send command for my work, just because LP merge
    > proposals are awesome. So submit_branch for me is not needed as a
    > way to provide a hint for send.

Same here, but I use 'bzr diff -rsubmit:' before creating the merge
proposal to do my own review.

    > 1) For trunk branches of my projects like qbzr and explorer I
    > often merge changes from other branches I reviewed. Of course for
    > the sake of review I get the copy of other branches on my local
    > computer and then merge them into trunk. Therefore submit_branch
    > was set after first merge, and then I delete/move merged branch
    > and submit_branch always points to non-existent
    > location. Therefore it's just waste.

I use the following trick for my "trunks" to fix merge aggressive
setting:

  bzr config submit_branch=lp:<project>

And since I have only one trunk by project, the issue is addressed. I
agree that this is not ideal but... wfm.

<snip/>

    > 2) I often merge into feature branches changes from other
    > branches, but it does not mean I will submit my current branch
    > into the branch where from I've merged. Therefore setting
    > submit_branch on that branch is incorrect thing.

I use the following^W *same* trick here.

    > 3) Furthermore, if I merge unrelated branch as `merge -r0..-1`
    > then I definitely never will submit my current branch into that
    > one. Therefore submit_branch is simply useless.


But there is:

4) I merge trunk in my branch to ensure I can still land it. In this
case, submit_branch is right.

I think that's the use case that was considered when making merge
remember submit_branch.

In this case the merge source *is* the submit target and often the
parent branch too...

    > So, as a matter of backward compatibility I don't think
    > submit_branch can be removed at all. But I can't prevent merge to
    > store the location to branch.conf even if I use --no-remember
    > command-line option,

Should be fixed.

Alternatively you can use:

  bzr config --remove submit_branch

or just set it correctly:

  bzr config submit_branch=<here-and-nowhere-else-thank-you>

    > although bzr never complains about that option. It just ignores it
    > and does what it does.

That's another bug, I think --remember works but -=no-remember doesn't
(that's the bug).

    > I can just rework bzr-explorer to be able to configure it to
    > avoid using submit_branch at all, but anyway submit_branch will
    > be remembered and will appear in the output of `bzr info`. And I
    > prefer to not see it there.

    > So, what could be better way to workaround that? Teach merge to
    > properly handle --no-remember

Yes.

    > or have global option in bazaar.conf to do the same but from
    > inside bzrlib?

There are certainly several ways to fix the whole issue including
accepting a default value for submit_branch defaulting to trunk (local
or remote), but let's start by finding what the broken workflows are.

You started with 3 of them, I added one, others may have encountered the
same problem.

    > Or just give up/take it easy and only concentrate on what I want
    > from explorer?

I think it's worth fixing in bzr in a way that can accommodate the
various workflows and that includes working from bzr-explorer or from
the command-line, so let's try to fix it in core ;)

    Vincent