Re: [gentoo-catalyst] rfc: merging catalyst git branches

[Sebastian Pipping]

On 12/08/2011 10:42 PM, William Hubbs wrote:
>> The cleanest way to do this this in Git may be:
>>
>>   # git checkout master
>>   # git merge -s theirs catalyst_2
>>
>> Haven't tested it though.

Just noticed that I mis-read the git-merge man page: "theirs" is an
option of the recursive merge strategy.  Sorry.


> I checked it out with the folks on #git, and they are recommending
> that I rename catalyst_2 to master. I was given a series of commands to
> do this.
>
> The down side is that this will cause a forced update, so everyone will
> have to re-clone the repository.

Actually there is a way to do this *without* the downside that you describe.

The trick is to create a fake merge commit using git commit-tree to sort
of emulate merge strategy "theirs".  The commit to make needs to:

 - point to catalyst_2^{tree} as its content.

 - have current catalyst_2 as its *first* parent in order to
   - indicate where the data actually came from
   - make commands like "git show HEAD^" descend into
     the old catalyst_2 branch later as that's the content
     that matters

 - have current master as the second parent
   (so people can keep working without trouble)


This time I tested it myself.  This is what to do:

  1)  Make sure your local master and catalyst are *both* up to date.

  2)  # git checkout master

  3)  Create and merge a fake merge commit as defined above:

        # git merge $(git commit-tree catalyst_2^{tree} \
          -p catalyst_2 -p master \
          <<<"Replace content on master with content from catalyst_2")

      Again, the *order* of parents matters.

  4)  Confirm it went fine, e.g. this diff should now be empty:

      # git diff master catalyst_2


> The target time I am considering for this is Monday, Dec 11, 0:00 utc.
> What that means is, everyone needs to have their changes pushed by then,
> then I'll make the change and send out an email here once I'm done.
> Then, you will have to re-clone your repositories.
>
> Any comments?

Please consider the alternative explained above.

What do you think?


Best,



Sebastian