Boost logo

Boost :

Subject: Re: [boost] [github] default PR destination
From: Bjørn Roald (bjorn_at_[hidden])
Date: 2014-01-08 03:38:50


On 01/08/2014 03:27 AM, Cox, Michael wrote:
> On Sun, Jan 5, 2014 at 8:44 AM, Peter A. Bigot <pab_at_[hidden]> wrote:
>
>> On 01/05/2014 09:34 AM, Peter A. Bigot wrote:
>>
>>> On 01/05/2014 08:50 AM, Peter Dimov wrote:
>>>
>>>> Peter A. Bigot wrote:
>>>>
>>>>> The last one (regex) I did off master, and I expect that most future
>>>>> ones will also be off master.
>>>>>
>>>>
>>>> What is a developer supposed to do with a pull request against master?
>>>>
>>>
>>> Pull the branch to a local repository, merge it into develop, test it,
>>> and reject or approve it. Normal git practice: see
>>> https://help.github.com/articles/merging-a-pull-request under "Merging
>>> locally".
>>>
>>
>> If master and develop do not share commits (as I believe they will not
>> under git-flow):
>
>
> I noticed that there are some submodules where develop is based on master,
> i.e. if you run "git branch --contains master", both master and develop are
> output. Most of them, like you say, show master and develop to be
> distinct. Won't that be a problem when trying to merge from develop to
> master?

I am not sure how much it matters, if any what branches is listed by
"git branch --contains master".

Apply/commit, merge from other remote tracking branches, or pull
contributions such as GitHub PRs to develop as they where applied to
trunk in SVN. It make a lot of sense for all maintainers to maintain
public devlop head as the point they expect every contributor to base
their contribution upon. Every other long running or deferred work not
expected to go into next release should be put into feature branches. I
am not familiar with pulling GitHub PR, but I expect it to be possible
to merge localy and commit, then before pushing to GitHub make the
assessment whether it should be put in a feature branch or stay in
develop, if it need to go in a feature branch you may make the feature
branch at the new commit and reset develop head to previous commit.

git checkout -b feature/PR-id-descriprion
git checkout develop
git reset HEAD^

this is very simple to do, understand, and verify correctness of what
you do using some GUIs, e.g. using "gitk" tool that come with git
distribution. In gitk right click the commits description to select
"create new branch here" and "reset develop to this" commands. You see
the result immediately in the commit tree and know it is what you want
before pushing to GitHub. Note, if you want gitk to show more than
current branch, use gitk --all, or create/edit view in menu. Don't
underestimate gitk based on looks, it is actually very elegant and
powerful to use if you try it a bit.

It is up to the maintainer to merge submodule releases only into master.
Thus it make no sense in general to base contributions on master as
develop may contain a lot of stuff for next release. I would vote
strongly for always using develop head as base for contributions.

> Just for grins, I took the accumulator submodule in which master and
> develop don't share commits and tried to locally rebase develop onto
> master. Unfortunately, the rebase failed early, but the few files I looked
> at have the same change trying to be applied on both sides, i.e. the lines
> between >>>> and ==== were the same as the lines between ==== and <<<<<.
> Any idea why that's the case?

no initial merge in git has been performed yet for accumulators:

https://svn.boost.org/trac/boost/wiki/PostCvtMergePoint

so I don't really know what to expect. And I don't know if we should
care much as it is not fair to git to expect much specific when trying
to blindly merge to disjoint trees.

> It looks like develop might be able to be "reconnected" to master (at least
> in accumulator), but it would take someone who understands the code and
> history of changes.

yes, this needs to be done with some special care as you indicate to
help Git perform sensible automatic of merge later.

-- 
Bjørn

Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk