How to manage merges from one line while excluding its merges from another

I have a tree like this:

      G---H---I---J---K   devel
     /       /       /
A---B---C---D---E---F    main
         \
          L---M---N---O   my

The 'my' branch forked off the 'main' line and for reasons out of my control cannot merge the 'main' changes (D,E,F) back.

The 'devel' line is a another line that I need to track and merge from, but I only want the changes local to that branch that are not on from the mainline. i.e., I only want commits G,H,J.

I want to continue to merge from the devel line as additional commits are made that are not on the main branch, but I do not want any of the future merge commits.

Is there an automated way to do this or do I have to cherry pick everything by hand?

Thanks, Lane

Hi Lane,

Lane Brooks wrote:

Nice diagram.

What is missing is a picture of what you want the result to be.

[...]

(I) I am guessing that the 'main' line is not part of the published history, in which case you what you want might look like this:

         ... devel
      /
 A---B---C- ... main
          \
           L---M---N---O---H'---J' my

The J commit itself could not be included in the history of the 'my' branch because its ancestor D is not meant to be published.

(II) In an alternative scenario, the 'main' changes (D, E, F) are forbidden because they introduce bugs. In this case, a solution might look something like this:

       ...                  K devel
     /               /       \
A---B---C---D---E---F main    \
         \                     \
          L---M---N---O---------P---Q---R---S

where Q, R, and S are commits (created with ‘git revert’) that undo the effect of F, E, and D. From then on, you can just merge from devel as usual.

In git 1.7.2, you will be able to automate (I) as follows:

 git cherry-pick --no-merges ^my ^main devel

Until then, you might want to try experimenting with something like this:

 git rev-list --no-merges ^my ^main devel |
 while read rev
 do
	git cherry-pick $rev ||
	{
		echo >&2 cherry-pick failed
		break
	}
 done

Hope that helps, Jonathan

On 06/25/2010 06:03 PM, Jonathan Nieder wrote:

I guess (I) is what I want, but your enumerating those two options makes me wonder how well things will go when 'devel' and 'main' merge and then 'my' needs to merge in from 'main'. Seems like a disaster waiting to happen.

I'll take a look at your suggested cherry-pick script. Can it be rerun multiple times or is it run one-time only. In others, as 'devel' continues to grow can I run it again and will it cherry-pick intelligently or will it try to reapply commits already cherry-picked?

Thanks, Lane

Lane Brooks wrote:

Good point.

  git cherry my devel main |
  grep -v ^- |
  while read mark rev
  do
	git cherry-pick $rev ||
	{
		echo >&2 cherry-pick failed
		break
	}
 done

I hadn’t known about the third argument to ‘git cherry’ before. Thanks for the example.