stacked diffs in git, seamlessly.
previously, if you wanted to adopt a stacked branch workflow, you'd have to do a lot of manual work every time you'd update anything in your stack -- jump thru each branch, re-rebase it on top of the previous one, get rid of duplicate commits, resolve conflicts, try to remember what the next branch was, and repeat...
there must be a better way. and that's exactly how git-stacked-rebase came to be.
a branch is just a reference to some commit (literally, it's a single-line file that contains a 40-character commit SHA -- check your .git/refs/
folder). why not just work on your latest feature branch, rebase comfortably, and then have your tool automatically update the partial branches to make them point to the correct new commits?
from those partial branches, you can create pull requests. with this workflow, you get to comfortably iterate in a single branch; your teammates get the benefits of reviewing smaller PRs (when they're ready). win win. that's it.
git-stacked-rebase got started 6 months before git rebase learned its own --update-refs
.
for the purposes of simply updating branches that are part of your stack,
git's rebase with --update-refs works just fine.
git-stacked-rebase still has unique features (and novel ideas yet to be implemented),
but hasn't been maintained for a little while now. this may change when i get more capacity to work on it.
git-stacked-rebase
is not specific to any host like github or gitlab. it's "specific" to git
itself.
it's not only a CLI either - it's written as a javascript library, with the CLI directly on top. though, we're keeping our options open for a potential rewrite in C (read: we're designing it in a way to make a rewrite in C possible w/o breaking changes to end users, if someone knowledgeable in git's core and C would want to take on this).
in general, the design goal has always been for the experience to feel as similar to git as possible. and this is why i feel that it could eventually become part of core git. (it wasn't a "goal" per se, it just felt like the right approach.)
nonetheless, there are other interesting things for us to explore, e.g.:
- creating host-specific adapters - they could be used to automate some simple tasks, such as creating a pull request, or changing the base branch of a pull request, etc.
- creating a browser extension to improve the experience of exploring stacked PRs.
follow http://kiprasmel.github.io/notes/git-stacked-rebase.html
dependencies:
- git
- a unix-like environment
- node.js
- tested versions: 12 thru 18, except v18 on linux. see details.
- note that after installing node, you can install version managers, e.g.
npm i -g n
, to easily change node's version.
- note that after installing node, you can install version managers, e.g.
- tested versions: 12 thru 18, except v18 on linux. see details.
- yarn (
npm i -g yarn
)
once satisfied, run:
git clone https://github.com/kiprasmel/git-stacked-rebase
# or: git clone git@github.com:kiprasmel/git-stacked-rebase.git
cd git-stacked-rebase
./nightly-setup-and-update.sh
$ git-stacked-rebase --help
git-stacked-rebase <branch>
0. usually <branch> should be a remote one, e.g. 'origin/master'.
1. will perform the interactive stacked rebase from HEAD to <branch>,
2. but will not apply the changes to partial branches until --apply is used.
git-stacked-rebase [-a|--apply]
3. will apply the changes to partial branches,
4. but will not push any partial branches to a remote until --push is used.
git-stacked-rebase [-p|--push -f|--force]
5. will push partial branches with --force (and extra safety),
6. but will not create any pull requests until --pull-request is used.
git-stacked-rebase [--pr|--pull-request]
7. generates a list of URLs that can be used to create stacked PRs.
(experimental, currently github-only.)
git-stacked-rebase --repair
(experimental)
finds branches that have diverged,
checks if they can be automatically re-integrated back into the stack,
and performs the repair if user accepts.
non-positional args:
--autosquash, --no-autosquash
handles "fixup!", "squash!" -prefixed commits
just like --autosquash for a regular rebase does.
can be enabled by default with the 'rebase.autosquash' option.
--git-dir <path/to/git/dir/>
makes git-stacked-rebase begin operating inside the specified directory.
--debug
prints the debug directory where logs are stored.
-V|--version
-h|--help