git-pile

Manage a pile of patches on top of a git branch

Requirements

• Python >= 3.6
• git >= 2.19
• Python modules:
  • argcomplete (optional)

Quickstart

Running git-pile

git-pile follows the git naming convention for binaries so we can use it as a subcommand to git. Either of the following forms work if the git-pile is in the search PATH:

$ git pile -h
$ git-pile -h

For autocomplete to work you need Python's argcomplete module installed. In the first case you also need to source the bash autocomplete file extra/git-pile-complete.sh so git knows how to complete the subcommand.

You can also install git-pile as a python package, although that's still not recommended since it's being updated frequently.

$ ./setup.py install

OR

$ ./setup.py install --user

In case a --user option is provided to the command above it will install under $HOME directory and bash completion file will be at $XDG_DATA_HOME/git-pile/bash_completion. There isn't a standard user directory for bash completion we can install to, so the user is expected to source this file.

Repository initialization

Initialize a new empty pile:

$ git pile init

Now you have a (orphan) pile branch to keep track of patch files to be managed on top of the master branch. Branch names may be changed via options to git pile init. Like git init creates a new repository, git pile init will create a new pile.

Alternatively, if the pile branch already exists in a remote repository (or even locally) and you want to setup your work tree to use it you should use the command below:

$ git pile setup internal/pile internal/embargo-branch

In the example above we have a remote called internal which has a branch named pile that contains the patch files and a branch named embargo-branch that will be the branch being generated when we apply the patches.

The second argument is optional and if not given the current branch will be used.

Develop a new commit and send to mailing list

You can use your normal git flow to develop changes in the project and create commits. Here is an example that is by no means restricted to git-pile:

$ git checkout -b newfeature internal/embargo-branch
$ echo "platform 16" >> platforms.txt
$ git add platforms.txt
$ git commit "add new platform"

Continuing from the initialization example, we create a new branch with internal/embargo-branch as the base. The result can be prepared to send to a mailing list in the same vein as how git format-patch works:

$ git pile format-patch -o /tmp/patches
/tmp/patches/0000-cover-letter.patch
/tmp/patches/0001-add-new-platform.patch

The cover letter will contain the diff to the pile branch (with patches added and so on), while the patch files will also be available in isolation: those are diffs to the result branch, and present to ease review.

If the commit you are adding should not be on top, but rather in the middle of the long running patch series, you can just move it down in the tree with git rebase -i. In the example below we will move it 10 commits down:

$ git rebase -i -11

  [ editor opens and you move the commit around. You can also reorder
    commits, edit commit message, etc, etc. Anything you do in a
    interactive rebase you can do here ]

$ git pile format-patch -o /tmp/patches
/tmp/patches/0000-cover-letter.patch
/tmp/patches/0001-old-platform.patch
/tmp/patches/0002-old-platform-2.patch
/tmp/patches/0003-add-new-platform.patch

Any patch that needs to be changed in order to accommodate the patch in the middle of the series will be prepared by git pile format-patch.

Generate the pile from the changed tree

Instead of sending to the mailing list, you can simply transform the changes from the branch to the physical patches maintained in the pile branch.

$ git pile genpatches master..newfeature
$ cd patches
$ git add -A
$ git commit -m "add platform"
$ git pile genbranch

This generates the patches, saving the final state in the patch series, and then recreates embargo-branch locally. Patches may be reordered, added in the middle, reworded etc. By going through the genpatches + genbranch cycle we can always re-generate the branch and keep the history of what was done, i.e. maintain the history of how the patches were changed/added/removed.

Apply a patch series to the pile

The cover-letter in a git-pile generated patch series (i.e. the one generated by git pile format-patch) always contain the diff of the original state of the tree to the current state. It may be used to apply the entire series by targeting the patches directory instead of the normal working directory.

$ # machine 2
$ cd pile
$ git am /tmp/patches/0000-cover-letter.patch
$ git genbranch

This will apply the entire series that was received (even if it was a 10 patches series, only the specially-formatted cover-letter needs to be applied).

Destroying the pile

If anything goes wrong and you'd like to start over, you can call the destroy command. This will remove all configuration saved by git-pile, the pile branch itself and the worktree directory it was using. If you have a backup of the patches (either manual or if the pile branch is in a remote repository), this is pretty safe to do and allow you to redo the configuration.