Preparing patches and pull requests for email submission is tedious and takes multiple commands. Revised patches must be labeled with increasing version numbers like v2, v3, and so on. Patch series start with a cover letter that contains a changelog describing the differences between revisions. All these details are repetitive and time-consuming to manage manually.
git-publish prepares patches consistently and stores them as git tags for future reference. It works with individual patches as well as patch series.
No constraints are placed on git workflow, both vanilla git commands and custom workflow scripts are compatible with git-publish.
Email sending and pull requests are fully integrated so that publishing patches can be done in a single command.
Hook scripts called during patch preparation can be used to add custom format checking or automated test runs.
First, put the git-publish script somewhere and make sure it has execute permissions:
$ mv ~/Downloads/git-publish ~/bin/ $ chmod u+x ~/bin/git-publish
Then run git-publish in setup mode to configure the git alias:
$ ~/bin/git-publish --setup
You can now use 'git publish' like a built-in git command.
Create a "topic branch" on which to do your work (implement a new feature or fix a bug):
$ git checkout -b add-funny-jokes ... $ git commit ... $ git commit
Send a patch series via email:
$ git publish --to patches@example.org --cc maintainer@example.org
Address code review comments and send a new revision:
$ git rebase -i master ... $ git publish --to patches@example.org --cc maintainer@example.org
Refer back to older revisions:
$ git show add-funny-jokes-v1
To store the first revision of a patch series:
$ git checkout my-feature $ git publish
This creates the my-feature-v1 git tag. Running git-publish again at a later point will create tags with incrementing version numbers:
my-feature-v1 my-feature-v2 my-feature-v3 ...
To refer back to a previous version, simply check out that git tag. This way a record is kept of each patch revision that has been published.
The version number can be set manually. This is handy when starting out with git-publish on branches that were previously manually versioned:
$ git checkout my-existing-feature $ git publish --number 7
This creates the my-existing-feature-v7 tag.
By default git-publish refuses to create a revision for the 'master' branch. Usually one works with so-called topic branches, one branch for each feature under development. Using the 'master' branch may indicate that one has forgotten to switch onto the intended topic branch. It is possible to override the topic name and even publish on 'master':
$ git checkout branch-a $ git publish --topic branch-b
This creates branch-b-v1 instead of branch-a-v1 and can be used to skip the check for 'master'.
Tag messages have a summary (or subject line) and a description (or blurb). When send email integration is used the summary is put into the cover letter Subject: line while the description is put into the body.
When prompting for tag messages on v2, v3, or other incremental revisions, the previous revision's tag message is used as the starting point. This is handy for updating the existing description and keeping a changelog of the difference between revisions.
The git-config(1) format.coverLetter value is honored. The default 'auto' value adds a cover letter if there is more than 1 patch. The cover letter can also be forced with 'true' or 'false'.
To insist on creating a tag message:
$ git publish --message
To refrain from creating a tag message:
$ git publish --no-message
For convenience these options are also available as --cover-letter and --no-cover-letter just like in git-format-patch(1).
Sometimes it is useful to edit the tag message before publishing. This can be used to note down changelog entries as you prepare the next version of a patch series.
To edit the tag message without publishing:
$ git publish --edit
This does not tag a new version. Instead a -staging tag will be created and the tag message will be picked up when you publish next time. For example, if you on branch my-feature and have already published v1 and v2, editing the tag message will create the tag my-feature-staging. When you publish next time the my-feature-v3 tag will be created and use the tag message you staged earlier.
git-publish detects whether the branch contains a single commit or multiple commits by comparing against a base branch ('master' by default). You can specify the base branch like this:
$ git publish --base my-parent
Most of the time 'master' works fine.
It is also possible to persist which base branch to use. This is useful if you find yourself often specifying a base branch manually. It can be done globally for all branches in a reposity or just for a specific branch:
$ git config git-publish.base origin/master # for all branches $ git config branch.foo.gitpublishbase origin/master # for one branch
git-publish can call git-send-email(1) after creating a git tag. If there is a tag message it will be used as the cover letter. Email can be sent like this:
$ git publish --to patches@example.org \
--cc alex@example.org --cc bob@example.org
After the git tag has been created as usual, commits on top of the base branch are sent as the patch series. The base branch defaults to 'master' and can be set manually with --base.
The git-send-email(1) aliasesfile feature works since the email addresses are passed through without interpretation by git-publish.
Patch emails can be manually edited before being sent, these changes only affect outgoing emails and are not stored permanently:
$ git publish --to patches@example.org --annotate
git-publish can background itself so patch emails can be inspected from the shell:
$ git publish --to patches@example.org --inspect-emails
Signed-off-by: <self> lines can be applied to patch emails, only outgoing emails are affected and not the local git commits:
$ git publish --to patches@example.org --signoff
Sending [RFC] series instead of regular [PATCH] series can be done by customizing the Subject: line:
$ git publish --to patches@example.org --subject-prefix RFC
git-publish remembers the list of addresses CC'd on previous revisions of a patchset by default. To clear that internal list:
$ git publish --to patches@example.org --forget-cc --cc new@example.org
In the above example, new@example.org will be saved to the internal list for next time.
CC addresses accumulate and cascade. Following the previous example, if we want to send a new version to both new@example.org and old@example.org:
$ git-publish --cc old@example.org
To temporarily ignore any CCs in the profile or saved list, and send only to the addresses specified on the CLI:
$ git-publish --override-cc --cc onetime@example.org --to patches@example.org
CCs specified alongside --override-cc are not remembered for future revisions.
Instead of providing command-line options each time a patch series is published, the options can be stored in git-config(1) files:
$ cat >>.git/config [gitpublishprofile "example"] prefix = PATCH for-example to = patches@example.org cc = maintainer1@example.org cc = maintainer2@example.org ^D $ git checkout first-feature $ git publish --profile example $ git checkout second-feature $ git publish --profile example
The "example" profile is equivalent to the following command-line:
$ git publish --subject-prefix 'PATCH for-example' --to patches@example.org --cc maintainer1@example.org --cc maintainer2@example.org
If command-line options are given together with a profile, then the command-line options take precedence.
The following profile options are available:
[gitpublishprofile "example"] base = v2.1.0 # same as --base remote = origin # used if branch.<branch-name>.remote not set prefix = PATCH # same as --patch to = patches@example.org # same as --to cc = maintainer@example.org # same as --cc suppresscc = all # same as --suppress-cc message = true # same as --message signoff = true # same as --signoff inspect-emails = true # same as --inspect-emails
The special "default" profile name is active when no --profile command-line option was given. The default profile does not set any options but can be extended in git-config(1) files:
$ cat >>.git/config [gitpublishprofile "default"] suppresscc = all # do not auto-cc people
If a file named .gitpublish exists in the repository top-level directory, it is automatically searched in addition to the git-config(1) .git/config and ~/.gitconfig files. Since the .gitpublish file can be committed into git, this can be used to provide a default profile for branches that you expect to repeatedly use as a base for new work.
git-publish can send signed pull requests. Signed tags are pushed to a remote git repository that must be readable by the person who will merge the pull request.
Ensure that the branch has a default remote repository saved:
$ git config branch.foo.remote my-public-repo
The remote must be accessible to the person receiving the pull request. Normally the remote URI should be git:// or https://. If the remote is configured for ssh:// then git-config(1) can be supplemented with a public url and private pushurl. This ensures that pull requests always use the public URI:
[remote "<name>"] url = https://myhost.com/repo.git pushurl = me@myhost.com:repo.git
Send a pull request:
$ git publish --pull-request --to patches@example.org --annotate
git-publish supports the githooks(5) mechanism for running user scripts at important points during the workflow. The script can influence the outcome of the operation, for example, by rejecting a patch series that is about to be sent out.
Available hooks include:
- pre-publish-send-email is invoked before git-send-email(1). It takes the path to the patches directory as an argument. If the exit code is non-zero, the series will not be sent.
- pre-publish-tag is invoked before doing the stage tag on current branch. It gets one argument which refers to the base commit or branch. If the exit code is non-zero, git-publish will abort.
Please report bugs to Stefan Hajnoczi <stefanha@gmail.com>.