git-conflict - Man Page

git conflict <commit>
git conflict --continue | --abort | -l

git conflict records a conflict resolution between two branches that git octopus will be able to reuse.

Start the recording process between the current branch and an other branch with

$ git conflict an_other_branch

This will set the index and the working tree in the conflict state. Once you’ve resolved it using git add, run git conflict --continue to record the resolution. It will print a ref that points to the resolution and resets the repository back to HEAD.

$ git conflict --continue
Conflict resolution created: refs/conflicts/75dd2c142264373c96a3341b2f9fb57bec7a7b7d

So git conflict doesn’t write history in your current branch.

git conflict --continue fails if a resolution for the same conflict already exists. To override a resolution you will need to delete the ref with git update-ref -d <ref> before recording the new one and possibly force the push.

If you’re satisfied with the resolution you can push it using a simple git push.

$ git push origin refs/conflicts/75dd2c142264373c96a3341b2f9fb57bec7a7b7d

-l

List all resolutions

--abort

Abort the current conflict resolution process, and try to reconstruct the pre-merge state.

If there were uncommitted worktree changes present when the merge started, git conflict --abort will in some cases be unable to reconstruct these changes. It is therefore recommended to always commit or stash your changes before running git conflict.

--continue

Record the index state as the resolution of the conflict and prints its ref. The index and the working tree are reset back to HEAD.

The first thing to know is that a conflict ref, for instance refs/conflicts/75dd2c142264373c96a3341b2f9fb57bec7a7b7d, is a standard commit. You can inspect it with git show or any logger tool. It has an author and dates etc...

The sha1 in the ref is a hash of the content of the conflict that has been resolved. This allows resolutions to be found easily by other commands given a conflict state by simply hashing the conflict content and looking for a refs/conflict/<sha1> of the same name. A resolution is not related to a particular branch, it can be applied in any circumstances that produced the same conflict. This means that you can continue to commit on your branches, you can rebase them, rename them anyhow. As long as the same conflict is produced, the resolution can be retrieved.

Currently, when reapplying a resolution, it will only look at both modified files. So conflicts with renames, mod changes and deletions are not handled. In addition, If you changed something in a file that wasn’t in conflict state, it won’t be reapply neither.

Managing conflicts with more than a single branch is subtle. Let’s say I’m on a local branch topic_mine and I have two different conflicts with topic_arya and topic_eddard

          A---B---C topic_mine
         /
    D---E---F---G topic_eddard
         \
          H---I topic_arya

I can write two resolutions

$ git conflict topic_arya
...
$ git conflict --continue
Conflict resolution created: refs/conflicts/9a4b40f6ec56260618b820e19c8bd734fbfb744d
$ git conflict topic_eddard
...
git conflict --continue
Conflict resolution created: refs/conflicts/8b3d16d22d0138ddbe6a1cd1a4cb9abf516a4609

If I run git octopus topic_* on my current branch (topic_mine), it will starts by merging topic_arya (branches are sorted alphabeticaly), will find the first resolution and then apply changes from topic_eddard, find the 2nd resolution and end up with a success. Let’s push !

$ git push origin refs/conflicts/9a4b40f6ec56260618b820e19c8bd734fbfb744d refs/conflicts/8b3d16d22d0138ddbe6a1cd1a4cb9abf516a4609

Then Arya fetches and runs git octopus topic_* from her branch topic_arya. It will starts by merging topic_eddard, no conflict happens (assuming), then merges topic_mine. The actual conflict of that merge will combine both conflicts because the current state have changes from both topic_arya and topic_eddard. This means that this conflict has a new signature that doesn’t have any resolution and the octopus will fail. In this case git octopus is not associative.

Part of the git-octopus suit.

git-octopus(1).