Re: Recent develop vs master shuffles broke existing pull requests

Subject: Re: Recent develop vs master shuffles broke existing pull requests
From: Mateusz Loskot (mateusz_at_[hidden])
Date: 2018-08-27 20:27:08

On Mon, 27 Aug 2018 at 15:45, Stefan Seefeld <stefan_at_[hidden]> wrote:
> On 2018-08-26 05:58 AM, Mateusz Loskot wrote:
> >
> > I noticed a slightly worrying situation with the pull request
> > reported before release of Boost 1.68, that is before we
> > shuffled the branches so the latest master become the develop.
> >
> > Have a look at https://github.com/boostorg/gil/pull/107
> >
> > GitHub now says the PR brings 250+ commits and changes 1164 files!
> >
> > If I try to rebase the mloskot:ml/remove-boost-type-traits branch against
> > the current origin:develop, I get gazillion conflicts impossible to resolve.
> >
> > For comparison, here is unbroken version of the PR #107
> > which is against the old develop from before Boost 1.68 release
> > https://github.com/mloskot/gil/pull/1
> >
> > Does anyone know how to 'fix' the develop so the existing
> > PRs don't become such a nonsense?
> > I've run out of my git fu.
> >
> > The 'broken' develop affects only two PRs, the #107 and #87
> > - the latter is the clang-format brainstorm, we can ignore it.
> >
> > The #107 is a non-trivial changeset of ~100 files, so I'd be a little bit
> > upset if the broken `develop` sends the effort out of the window.
> > I will try to salvage it though.
> >
> > How can we avoid such problems in future?
>
> I'm sorry to hear about this trouble. I should try to do this rebase of
> the PR. Given that I used the `git rerere` feature during that work, the
> machine may be able to assist in bringing over your PR by remembering
> conflict resolution.

The problem is that rebase-ing the topic branches lead to conflicts
between commits pre-dating Boost switch rom SVN to Git :)

> Of course, we should avoid such problems by never using "history
> rewrites" on public branches.

Yes, I think it's a good policy to follow in future.

> (This is common knowledge, and I simply
> neglected the fact that there were existing PRs against `develop` that
> would have to be migrated if I introduced a "new" `develop` branch.
>
> I'll see what I can do (but unlikely this week, which will be extremely
> busy for me).

Don't worry. I will take care of it when I'm ready to pick that work again.

I just wanted to share the problem as a warning for future.

Best regards,
-- 
Mateusz Loskot, http://mateusz.loskot.net