Re: Master and origin/master diverged

From: Neal Kreitzinger <nkreitzinger@gmail.com>
Date: Jun 27, 2012, 05:54 UTC
Message-ID: <4FEAA027.7050307@gmail.com>
In-Reply-To: <CAE1pOi24EVq__XzxtBoAryzQ+F=sAy8-hY23M-P4YoQFXdpSSw@mail.gmail.com>

On 6/25/2012 9:49 PM, Hilco Wijbenga wrote:

Show 53 quoted lines

> On 22 June 2012 16:47, Neal Kreitzinger <nkreitzinger@gmail.com> wrote:
>> On 6/22/2012 3:18 PM, Hilco Wijbenga wrote:
>>>
>>> On 22 June 2012 12:01, Neal Kreitzinger <nkreitzinger@gmail.com> wrote:
>>>> On 6/22/2012 12:53 PM, Hilco Wijbenga wrote:
>>>>>
>>>>> One of my developers managed to push something that somehow "diverged"
>>>>> origin/master from everyone else's local master.
>>>>>
>>>>> A --> B --> C --> D (everybody's local master)
>>>>> |
>>>>> \--> B' --> C' --> D' --> E (origin/master)
>>>>>
>>>>> (i.e., A is the commit where things diverged; everyone's local master
>>>>> points to D but the new commit (E) that was pushed to origin/master
>>>>> uses different SHA1s for B, C, and D)...
>>>>>
>>>>>
>>>>> Now running git pull creates a merge commit joining D and E.
>>>>>
>>>>> ...Does anyone have any idea as to what might have happened? Perhaps if
>>>>> I
>>>>>
>>>>> understand how this happened I might be able to prevent it from
>>>>> happening again.
>>>>>
>>>> Some ways you can prevent it from happening again:
>>>
>>>> (2) have your devs do git pull --ff-only
>>>
>>> Is this something that can be set in git config? I looked but didn't
>>> see anything obvious.
>>
>> OTTOMH, you could change the git fetch config for master and take away the
>> leading '+' sign which would not allow non-fastforward fetches of master.
>>   That in turn would prevent merging such a non-ff remote tracking branch of
>> master into your branch master.
>>
>>
>> Actually, I guess what I really want is
>>> something for git push, right?
>>>
>> Some ways to do it:
>> (1) I think you could have rebase and commit hooks locally that prevent
>> someone from rewriting history on master.  That in turn would prevent
>> someone from pushing a rewritten history.
> Yes, I have been thinking about that.
>
> How does one create "portable" hooks? I have to deal with GNU/Linux,
> OS X, and MS Windows. We all have Java installed so I first thought of
> using JGit but I am not clear on how well JGit supports using it in a
> hooks. Should I make Ruby a required part of the dev environment and
> use Ruby hooks?

I don't know about java, ruby, or JGit (yet). I make hook updates easy with this alias:

get-hooks = !rm -f .git/hooks/pre-commit && git init --template=/opt/mydir/git-templates/dev/templates

I update the master copy of the pre-commit hook (in this case) in the template and then have the users run git get-hooks. All my users are on the linux server. Maybe this idea is helpful to you in some way.

Show 7 quoted lines

>
>> (2) When merging topic branches to master use git merge --ff-only.  Then
>> when you push it to remote master you know it's a fastforward and not a
>> history rewrite.
> Given how hard it is to teach devs to only push fast-forward merges, I
> am not sure how well this would work.
>

You could create an alias 'git merger' and have them run that and it will do the --ff-only option. Maybe post-merge hook and/or pre-commit hook. I don't think --ff-only was part of your requirements so you can probably ignore this since I got off track by suggesting it. I think your problem is history rewrites and not merge commits. I think (1) --ff-only denies merge commits altogether, and (2) denyNonFastForwards allows merge commits but denies history rewrites so they (1 and 2) are not really the same though they both use the term 'fastforward' they have different definitions of what that means. Someone please correct me if I'm wrong because this does seem a bit confusing now that I'm saying it out loud.

v/r, neal