Question

Here's a workflow that I commonly deal with at work.

git checkout -b feature_branch
# Do some development
git add .
git commit
git push origin feature_branch

At this point the feature branch is up for review from my colleagues, but I want to keep developing on other features that are dependent on feature_branch. So while feature_branch is in review...

git checkout feature_branch
git checkout -b dependent_branch
# Do some more development
git add .
git commit

Now I make some changes in response to the code review on feature_branch

git checkout feature_branch
# Do review fixes
git add .
git commit
git checkout dependent_branch
git merge feature_branch

Now this is where we get have problems. We have a squash policy on master, which means that feature branches that are merged into master have to be squashed into a single commit.

git checkout feature_branch
git log # Look for hash at beginning of branch
git rebase -i  first_hash_of_branch # Squash feature_branch into a single commit
git merge master

Everything is cool, except with dependent_branch. When I try to rebase dependent branch onto master or try and merge master into it, git is confused by the re-written/squashed history and basically marks every single change in depedendent_branch as a conflict. It's a PITA to go through and basically re-do or de-conflicticize all of the changes in dependent_branch. Is there some solution to this? Sometimes, I'll manually create a patch and apply it off a fresh branch of master, but if there's any real conflicts with that, its even worse to fix.

git checkout dependent_branch
git diff > ~/Desktop/dependent_branch.diff
git checkout master
git checkout -b new_dependent_branch
patch -p1 < ~/Desktop/dependent_branch.diff
# Pray for a clean apply.

Any ideas? I know this happens because of the re-written history during the squash, but that's a requirement that I can't change. What's the best solution / workaround? Is there some magic I can do? Or is there a faster way to do all the steps involved with manually creating the diff?

Was it helpful?

Solution

A little bit about why this happens:

I'll let O be "original master" and FB be "new master", after a feature branch has been merged in:

Say feature_branch looks like:

O - A - B - C 

dependent_feature has a few extra commits on top of that:

O - A - B - C - D - E - F

You merge your original feature branch into master and squash it down, giving you:

O - FB

Now, when you try to rebase the dependent branch, git is going to try to figure out the common ancestor between those branches. While it originally would have been C, if you had not squashed the commits down, git instead finds O as the common ancestor. As a result, git is trying to replay A, B, and C which are already contained in FB, and you're going to get a bunch of conflicts.

For this reason, you can't really rely on a typical rebase command, and you have to be more explicit about it by supplying the --onto parameter:

git rebase --onto master HEAD~3  # instruct git to replay only the last
                                 # 3 commits, D E and F, onto master.

Modify the HEAD~3 parameter as necessary for your branches, and you shouldn't have to deal with any redundant conflict resolution.

Some alternate syntax, if you don't like specifying ranges and you haven't deleted your original feature branch yet:

git rebase --onto master feature_branch dependent_feature

                                 # replay all commits, starting at feature_branch
                                 # exclusive, through dependent_feature inclusive 
                                 # onto master

OTHER TIPS

In this particular case it seems you "know" that only the squashed work of the branch you originally worked on has been put into master.

So you can happily merge by keeping your changes every time when there is a conflict. There is an option for that:

git merge -Xours master

See https://git-scm.com/docs/git-rebase for details more details.

I heartily disagree with the "mash every feature development down into a single commit" policy, but it's their call...

I'd keep the branches as-is, and create a mashed up commit just for publication, in a special branch. Being able to follow the development step by step is valuable, even if management doesn't believe in it. Mark the places of squashes by tags in the "real" branches, you could also add between-squash tags with messages pointing back to the real commits.

In case your dependent branch has multiple non-trivial changes, rebasing like @joshtkling suggested, can become painful. I have often turned to an alternative strategy which might be easier. This doesn't mean using rebase is wrong or anything, I just want to offer an alternative. It uses the "ours" strategy, but in contrast to @nha's answer it also works if there is more work on the master.

Suppose the starting situation

O - A - B - C (feature_branch) - D - E - F (dependent_branch)
  \
   - D - Cs (squash merged feature branch) - G - H (master)

where D, G and H are work done by others, possibly squash merges from other branches not shown here.

Merging dependent_branch into master is likely to produce conflicts, because the changes from A, B and C are duplicated. In order to tell git that it does not need to consider these, we can introduce a new merge base that combines C and Cs, by creating a temporary branch and then merge that into our dependent branch. This way git has much more knowledge for merging.

git checkout -b temp Cs
git merge C --strategy ours
git checkout dependent_branch
git merge temp
git branch -d temp

The second merge might yield some conflicts, but in my experience they are fewer and simpler to solve than when using rebase. Afterwards merging dependent_branch and master should be a lot easier.

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top