scottbez1 / GitChildBranchHelpers

Git helpers for working with child branches

This is a set of utilities to make working on child branches (and grandchild branches and arbitrary trees of branches) easier, without getting messed up due to an incorrect merge-base. I've found this to be particularly helpful when code review support is spotty, since you can just keep working in child branches and can always go back and update your parent and rebase easily if you need to. Working on RC branches is also easier, since there's no more having to remember to land onto the right branch.

The problem being solved

Let's say we have a set of dependent branches, as described by the commit graph below, where each node (eg A2) is a commit and each line represents a branch, with the branch name at the end (eg <= feature_c).

M1 <= master
|
\--A1--A2 <= feature_a
       |
       |--B1 <= feature_b
       |
       \--C1 <= feature_c
          |
          \--D1 <= feature_d

Now let's say we do some more work on our branches and we pull master. Our commit graph now looks like this:

M1--M2 <= master
|
\--A1--A2--A3 <=feature_a
       |
       |--B1--B2 <=feature_b
       |
       \--C1--C2 <=feature_c
          |
          \--D1 <=feature_d

There are two primary operations to support, given a commit graph like the one above. The first is propagating changes to child branches. Let's say commit A3 contains a crucial fix or breaking change that needs to be taken into account in a few of the other feature branches. What we can do is checkout the feature_a branch and propagate our changes. This will recursively cause each child branch to rebase onto the tip of its parent branch. The child rebase command does this currently (though it will also rebase the current branch on top of its parent, which is not done in this example). The result would be as follows:

M1--M2 <= master
|
\--A1--A2--A3 <=feature_a
           |
           |--B1'--B2' <=feature_b
           |
           \--C1'--C2' <=feature_c
                   |
                   \--D1' <=feature_d

Let's go back to the situation before we propagated changes from the feature_a branch. Another workflow we want to support is getting all of the upstream changes in a branch. Say you're working on feature_b and you find a bug preventing you from working, but realize the bug is fixed in master. What we want to do here is pull in upstream changes, or effectively super rebase such that we rebasing not just on our parent, but all the way to the upper-most parent. This causes us to be based on the latest code from each of our parent branches. There are no scripts in this repo that do this behavior yet. Doing this for feature_b would cause the commit graph to look like this:

M1--M2 <= master
|   |
|   \--A1'--A2'--A3' <=feature_a
|                |
|                \--B1'--B2' <=feature_b
|
\--A1--A2(--A3)
       |
       \--C1--C2 <=feature_c
          |
          \--D1 <=feature_d

Note that at this point, feature_c and feature_d are no longer actually based on feature_a. However, these scripts will correctly handle this when you propagate changes from feature_a (or master), or when you rebase onto upstream changes from feature_c or feature_d. Doing so would look like this (with the exception of pulling upstream changes from feature_c, which would leave feature_d not actually based on the current feature_a):

M1--M2 <= master
    |
    \--A1'--A2'--A3' <=feature_a
                 |
                 |--B1'--B2' <=feature_b
                 |
                 \--C1'--C2' <=feature_c
                         |
                         \--D1' <=feature_d

Usage

Add the bin/ folder to your PATH and use child command. Running child -h will give you a list of valid actions and child <action> -h will give you more information about a given action. You can also source the bash_zsh_git_helper_aliases.sh file in your .bashrc (or .zshrc) for a set of shorter aliases for working with child branches.

Make sure to always use these commands when possible, rather than using the raw git alternatives, so that this can keep track of all the branches properly.

In particular:

• Always create branches with the child make-branch command (default alias cmk), even if the branch is off of master.
• Always rebase branches with child rebase (default alias crb)
• Always diff branches with child arc-diff (default alias cad)
• Always land branches with child arc-land (default alias cal)

For a list of all available commands, run child -h. For the list of default aliases, look in the bash_zsh_git_helper_aliases.sh file.

Unfortunately, right now there's no good way to tell it about your existing branches, but they should continue to work fine alongside this so you can try this out with new branches you make.

Implementation details

This solution involves a series of scripts that call through to git while maintaining extra information about the branches you create. Specifically, for each branch it records:

• The branch name
• The parent branch's name
• The commit that the branch is based off of
• Optionally, the commit that the branch is rebasing to

This is stored in the repo's .git folder, at child_branch_helper/branches.csv. By keeping track of the base revision separately from the parent branch name, we're able to allow the parent branch to change and be rebased independently of the child. However, in order to initially obtain this information and keep it up to date, you must use these scripts for certain operations. In particular, these scripts should always be used to:

• Make new branches - so that we have the correct parent branch name and initial base revision
• Rebase a branch - so that we can update the base revision
• Arc diff a branch - so we can ensure you're only diffing against your parent and diff's don't contain your parent's commits.
• Arc land a branch - so we can ensure you're only landing your code, and none of your parent branch's

Future work

I'd like to make it so you don't have to remember to use special commands, but rather it just works seamlessly. Whether this is patching or wrapping git, or something else, it'd be nice to come up with a more seamless integration. Similarly for arc, which might just involve changing the merge-base strategy.