Working with Branches

Basics

Enumerate your local branches:

  cd src
  git branch

Switching from one branch to another: Example: Switching from branch 'branch1' to branch 'branch2'.

  cd src
  git checkout branch2

Note that - can be used to refer to the previous branch, which is useful when switching between two branches.

git checkout some_branch
git checkout -  # back to previous branch!

Suggested branching workflow

We normally do our feature work in branches to keep changes isolated from each other. The recommended workflow is to make local branches off the server master, referred to as the origin/master branch.  The git-new-branch command (in depot_tools) will do this:

      git new-branch branch_name
Note that this is equivalent to the following:
      git checkout -b new_branch origin/master

Branch off a branch

If you have dependent changes, a very productive workflow is to have a branch off a branch. Notably, this means that your patch sets (in Rietveld) will show the separate changes, and be easy to review, rather than showing the merged changes (including irrelevant changes), and means you can commit downstream CLs without having to rebase them after the upstream has landed. Do this as follows:
git checkout master
git checkout branch1
# some edits, commit
git checkout -b branch2
git branch --set-upstream-to=branch1
# some edits, commit

Now when you update the first branch, you can simply git pull on the second branch as normal to pull in your changes:
git checkout branch1
# some edits, commit
git checkout branch2
git pull

(Note: Below is the trick which seems to have been required only with Rietveld, but is not necessary with Gerrit anymore.)

Note that these incremental patch sets are easy to review, but cannot be applied to HEAD (because they are based off another branch), so you cannot test them on try bots. To test these, upload the diff from HEAD to a different (test-only) issue and test there:
ISSUE=$(git cl issue | sed -r 's/[^0-9]*([0-9]*).*/\1/')  # Save current issue via git cl issue
git cl issue 0  # Clear issue
git cl upload -t "[TEST ONLY] Run tests for $ISSUE" -f origin/master  # Upload to new issue, no prompts
TEST_ISSUE=$(git cl issue | sed -r 's/[^0-9]*([0-9]*).*/\1/')  # Save test issue
git cl try
git cl issue $ISSUE  # Restore original issue

To open the test issue in Rietveld:
git cl issue $TEST_ISSUE
git cl web
git cl issue $ISSUE

To try again:
git cl issue $TEST_ISSUE
git cl upload -t 'Try again' -f origin/master
git cl try
git cl issue $ISSUE

...and then when finished, delete the CL to clean up (via Rietveld):
git cl issue $TEST_ISSUE
# git cl delete  # This command does not exist, go through Rietveld
git cl web
git cl issue $ISSUE

Lastly, when an upstream branch has landed, and you've rebased it so it agrees with master, you can set the downstream branch's upstream branch to be origin/master, which means you can now forget about the earlier branch:
git checkout branch2
git pull  # assuming branch1 is up-to-date with origin/master
git branch --set-upstream-to=origin/master

Splitting up a CL

A common variant of "branch off a branch" is splitting up a large CL into pieces. Given a local branch big, you'd like to split it up into branch1 and branch2.

One way to do this is to split off branch1, have that reviewed and committed, update local repo (gclient sync), and then rebase big to master. This is ok, but adds latency, and you can get the same result locally, so the second part of the CL can be uploaded and reviewed even before the first part is committed.

This can be done manually, particularly if the files don't overlap, by making a new branch branch1, manually copying the changed files in the first part, then branching branch2 off of that, and manually copying the files in the second part.

However, this can be done more cleanly via git. The main points are git-add -i (Interactive Staging: to interactively choose files or hunks of a patch set) and git-rebase (to set the second part to be dependent on the first part, removing the common changes); git-cherry-pick is a technicality:
# first split off branch1
git checkout origin/master
git new-branch branch1
git cherry-pick -n ..big  # apply and stage all ancestors of big that are not ancestors of HEAD, do not commit
git add -i  # interactively choose which files or hunks to stage
git commit  # commit staged changes
git checkout -- . # discard unstaged changes

# next set big to be a branch off of branch1
git checkout big
git branch --set-upstream-to=branch1
git rebase branch1  # may need to resolve conflicts
git branch -m branch2  # done!

Concretely, git-cherry-pick -n applies and stages all changes, but does not commit them. In git-add -i, you can unstage files via 3: revert, and then stage files (or hunks of files, or edit the diff manually) via 5: patch, with manually editing via by the e - manually edit the current hunk option when staging a hunk. While tedious, this is often easier than resolving conflicts during rebasing.

Deleting an obsolete branch

You generally want to delete local branches after the changes have been committed to master. It is safest to check that your work has, in fact, been committed before deleting it.
Remember that you can always apply a CL that has been posted via:
  git checkout -b myworkfrompatch
  git cl patch 12345  # Using CL issue number, not bug number, applies latest patch set
  git cl patch https://codereview.chromium.org/download/issue12345_1.diff  # Use URL of raw patch for older patch sets

...so as long as your CL has been posted, you can easily recover your work. Otherwise, you'll need to dig through the git repository, so be careful.

Simply, if master (remote or local) is up-to-date and your branch has been rebased, git branch -d will delete the branch. If not, it will refuse; to force deletion, use git branch -D. So make sure master is up-to-date, rebase branch, and then try deleting (optionally check manually before).

Beware that if your local branch has many revisions (instead of always amending a single revision), rebasing to master may fail, since it will try to apply the patches incrementally. You can avoid this by squashing your local revisions into a single revision (see 6.4 Git Tools - Rewriting History, or more simply squashing commits with rebase). This may be more trouble than it's worth, but it's the safest way.

To check that the branch has been committed upstream:

  git checkout mywork
  git rebase origin/master
  git diff --stat origin/master  # optional check

If there are no differences, delete the branch:

  git checkout origin/master
  git branch -d mywork  # will only work if has been merged

NOTE: If you haven't waited long enough after your commit, it is possible that 'git fetch' did not get your commit and git will continue to believe that you have local changes, which will prevent "git branch -d" from succeeding. It's best to redo the steps later, (The repo is updated every 3 minutes) but you can also instruct git to force-delete your branch.

Note that when you delete your branch, it will give you the SHA-1 hash for its tip:

    Deleted branch mywork (was 123abc0).

You can then recover the branch via:

    git checkout -b mywork 123abc0

If you forget the hash, you can find it via git reflog. (Reference: Can I recover branch after the deletion in git?)

Prevent commits to master

If you commit to master, updating will be messy.  (This is a good reason to simply delete master entirely, as noted near the top of this document.  You only need to read the remainder of this section if you don't do this.)

You can prevent this by adding a pre-commit hook that checks if you're in master and stops you from doing this. Create a file named chromium/src/.git/hooks/pre-commit and add the below to it, then mark executable. (Blink developers: add in blink directory as well.)

#!/bin/bash

# Prevent commits against the 'master' branch
if [[ `git rev-parse --abbrev-ref HEAD` == 'master' ]]
then
    echo 'You cannot commit to the master branch!'
    echo 'Stash your changes and apply them to another branch, using:'
    echo 'git stash'
    echo 'git checkout <branch>'
    echo 'git stash apply'
    exit 1
fi

If you've accidentally uploaded a change list from master, you can clear the association of an issue number with master via:

  git cl issue 0

If you've accidentally committed to master, then, after copying your work to a new branch (e.g., make patch and then apply to new branch), you can clean up master by deleting your accidental commit as per this answer to How to undo the last Git commit? -- see more details there.

  git reset --hard HEAD~1
Comments