There is no perfect solution but you can get close, perhaps close enough.
Be sure to start with a clean work tree and index (see require_clean_work_tree
in git-sh-setup
).
For each candidate branch $branch
that might be delete-able:
- Find its merge target (presumably
merge_target=$(git config --get branch.${branch}.merge)
). Check out the merge target.
- Do a merge with
--no-commit
; or in step 1, check out with --detach
so that you will get a commit you can abandon, if the merge succeeds.
- Test whether git thinks the merge succeeded, and if so, whether the current tree matches the previous tree, i.e., brought in no changes. If you can test exact matches, and if you allow the commit to happen (via
--detach
), you can do this last test very simply, without any diff-ing: run both git rev-parse HEAD^{tree}
and git rev-parse HEAD^^{tree}
1 and see if they produce the same hash. If you don't allow the commit, you can still git diff
the current (HEAD
) commit against the proposed merge. If you need to remove some noise from the diff (e.g., config files that should not be, but are anyway, in the commits), this gives you a place to do it.
- Reset (
git merge --abort; git reset --hard HEAD; git clean -f
or similar, depending on how you have decided to implement steps 1-3). This is just meant to make your work tree and index clean again, for the next pass.
- If the merge in step 3 worked and introduced no changes, you may delete the local branch. Otherwise, keep it.
In essence, this is "actually do the merge and see what happens", just fully automated.
1This notation looks a bit bizarre, but it's just HEAD^
—the first parent of HEAD
—followed by ^{tree}
. Alternate spellings might be easier to read: HEAD~1^{tree}
or ${merge_target}^tree
, where ${merge_target}
is the branch you checked out in step 1. Note that this assumes the merge succeeded. The merge result is in the exit status of git merge
: zero means succeeded, nonzero means failed and needs manual assistance, presumably due to a merge conflict.