Re: [HACKERS] pgsql: /home/peter/commit-msg

On Mon, Nov 16, 2009 at 09:05, Heikki Linnakangas
<heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> Magnus Hagander wrote:
>> On Mon, Nov 16, 2009 at 08:29, David Fetter <david(at)fetter(dot)org> wrote:
>>> On Mon, Nov 16, 2009 at 06:56:54AM +0200, Peter Eisentraut wrote:
>>>> Yeah, sorry guys.  I fixed the CVS log message now.
>>> Strangely, the git repo still shows the old message.  For the record,
>>> there's the new one:
>>
>> I don't find that strange at all.
>>
>> In git, the commit message is part of the commit, and thus the SHA1.
>>
>> If it changes, it would be a different commit. Which would change
>> history and break the repositories of anybody pulling from it.
>>
>> So it's not only not strange, I'm very happy it didn't pull those
>> changes and broke my repository :-)
>
> Yeah, I'm glad it didn't do anything funny with the mirror.
>
>> Now, if we want to "correct" that the way to do it is to rebuild the
>> git mirror from scratch and have everybody start over, I think :-)
>> While you're not supposed to change history in any RCS, git makes it a
>> lot harder than cvs to do it...
>
> Nah, you'd only have to back out to the commit before the one that was
> broken. And actually git provides quite simple commands to do that, see
> git-reset for example. I forget how exactly the mirroring software
> works, but I presume it would then see that commit as a new one, as well
> as anything on top of it, and mirror them.

I would assume that yes, but I haven't tested it.

> Rewriting git history like that would mean that anyone who has pulled
> from the mirror since that commit happened would get an error the next
> time they try to pull/fetch again. But you can easily get over that by
> doing "git fetch --force".
>
> So we could rewrite the git history too, and I think it would be quite
> nice to have the right commit message there as well. But I don't care
> enough to volunteer to do the legwork. If we are going to do it, we
> should do it as soon as possible, while we're only a couple of commits
> ahead of that point. It's going to be more painful later on.

Yeah.

Right now, that commit is actually the top, so it would be just one.

I was about to do it right now, then I realized that it hits
backbranches as well so it's not quite so easy. So I won't do it now -
I have to leave for JPUG pretty soon, and I don't want to risk leaving
us with an inconsistent git mirror if things go wrong.

--
Magnus Hagander
Me: http://www.hagander.net/
Work: http://www.redpill-linpro.com/