pivotal-github

The pivotal-github gem facilitates a Pivotal Tracker–GitHub workflow inspired by the workflow used by Logical Reality. (Despite its name, pivotal-github also works fine with Bitbucket; see Configuration below.) As per usual, there are several projects (notably git-flow and git-pivotal) that implement similar solutions, but none met my exact needs, so I rolled my own.

Installation

You can install the pivotal-github gem directly as follows:

$ gem install pivotal-github

The full workflow described herein requires some of the Git utilities from git-utils, which is included as a gem dependency.

Usage

The pivotal-github gem adds several additional Git commands to the local environment. The main addition, git story-commit, automatically incorporates the Pivotal Tracker story id(s) into the commit messages, while adding options to mark the story Finished or Delivered.

The git story-commit command makes the assumption that any string of eight or more digits in the branch name is a story id. (As of this writing, Pivotal Tracker ids are eight digits long, so shorter digit strings aren't valid ids.) This means that the branch names 62831853-add-markdown-support, 62831853_add_markdown_support, add-markdown-support-62831853, and rails_4_0_62831853 all correspond to story id 62831853, while add-things-62831853-31415926 corresponds to both 62831853 and 31415926.

The full set of commands is as follows:

git story-commit

git story-commit makes a standard git commit with the story number added to the commit message. This automatically adds a link at Pivotal Tracker between the story and the diff when the branch gets pushed up to GitHub.

For example, when on a branch called add-markdown-support-62831853, the git story-commit command automatically adds [#62831853] to the commit message:

$ git story-commit -am "Add foo bars"
[add-markdown-support-62831853 6f56414] Add foo bars

The commit message is multiline and includes the story id:

Add foo bars

[#62831853]

(Previous versions of pivotal-github put the story id on the same line as the commit summary (per the usage at the Pivotal Tracker API), but placing it in a separate line gives the user direct control over the length of the message. It also looks less cluttered.)

To mark a story as Finished, add the -f flag:

$ git story-commit -f -am "Remove baz quuxes"

This gives the message

Remove baz quuxes

[Finishes #62831853]

To mark a story as Delivered, add the -d flag:

$ git story-commit -d -am "Remove baz quuxes"

The message in this case is

Remove baz quuxes

[Delivers #62831853]

Either the -f flag or the -d flag can be combined with other flags, yielding commands like

$ git story-commit -dam "Remove baz quuxes"

git story commit supports multiple story numbers as well. For example, with a branch called add-things-62831853-31415926, we could deliver both stories as follows:

$ git story-commit -dam "Remove baz quuxes"
[add-things-62831853-31415926 7g56429] Remove baz quuxes

The message here is

Remove baz quuxes

[Delivers #62831853 #31415926]

Options

$ git story-commit -h
    Usage: git story-commit [options]
        -m, --message MESSAGE            add a commit message (including story #)
        -f, --finish                     mark story as finished
        -d, --deliver                    mark story as delivered
        -a, --all                        commit all changed files
        -h, --help                       this usage guide

Additionally, git story-commit accepts any options valid for git commit. (git story-commit supports the -a flag even though that's a valid option to git commit so that the compound flag in git story-commit -am "message" works.)

git story-merge

git story-merge merges the current branch into the target branch (defaults to master). On a branch called add-markdown-support-62831853, git story-merge is equivalent to the following:

$ git checkout master
$ git merge --no-ff --log add-markdown-support-62831853 -m "#[62831853]"

Note that this effectively changes the default merge behavior from fast-forward to no-fast-forward, which makes it possible to use git log to see which of the commit objects together have implemented a story. As noted in A successful Git branching model,

The --no-ff flag causes the merge to always create a new commit object, even if the merge could be performed with a fast-forward. This avoids losing information about the historical existence of a feature branch and groups together all commits that together added the feature… Yes, it will create a few more (empty) commit objects, but the gain is much bigger than that cost.

The --log option puts the commit messages from the individual commits in the merge message, while the -m flag adds the story id to the commit (optionally marking it finished or delivered with the -f or -d flag, respectively). Including the story id arranges for the merge commit itself to appear in the activity log at Pivotal Tracker, which is especially useful for viewing the full diff represented by the story.

Because of the way options are chained, passing -ff or --no-log to git story-merge will override the --no-ff or --log flags (respectively) and thus restore the default behavior of git merge.

Finally, experience shows that it's easy to forget to mark a story finished when making the final commit. As a reminder, the git story-merge command exits with a warning if the most recent commit doesn't contain 'Finishes' or 'Delivers' (or 'Finished', 'Delivered', 'Fixes', or 'Fixed'). This behavior can be overriden with the --override option.

Options

Usage: git story-merge [branch] [options]
    -o, --override                   override unfinished story warning
    -f, --finish                     mark story as finished
    -d, --deliver                    mark story as delivered
    -h, --help                       this usage guide

Additionally, git story-merge accepts any options valid for git merge.

git story-pull-request

git story-pull-request opens the proper remote URL to issue a pull request for the current branch (OS X–only):

$ git story-pull-request

git story-pull-request issues a git push-branch as well (from git-utils), just in case the local branch hasn't yet been pushed up to the remote repository. For reference, it then makes a commit containing a message with links to all the delivered story ids. These ids consist of all the delivered stories that haven't already been delivered by a pull request.

As with git story-merge, by default git story-pull-request exits with a warning if the most recent commit doesn't finish the story.

Options

Usage: git story-pull-request [options]
    -b, --base-branch BRANCH         base branch for delivered ids
    -o, --override                   override unfinished story warning
    -h, --help                       this usage guide

git story-accept

git story-accept examines the repository log and changes every Delivered story to Accepted. This makes it possible to accept a pull request by merging into master and then mark all the associated stories Accepted by running git story-accept. This saves having to manually keep track of the correspondences.

The purpose of git story-accept is to accept stories that have been merged into master, so by default it works only on the master branch. This requirement can be overridden by the --override option.

git story-accept requires the existence of .api_token and .project_id files containing the Pivotal Tracker API token and project id, respectively. The user is prompted to create them if they are not present. (They aren't read from the command line using gets due to an incompatibility with options passing.)

Options

Usage: git story-accept [options]
    -o, --override                   override master branch requirement
    -q, --quiet                      suppress display of accepted story ids
    -h, --help                       this usage guide

story-open

The story-open command (no git) opens the current story in the default browser (OS X–only):

$ story-open

project-open

The project-open command (no git) opens the current project in the default browser (OS X–only):

$ project-open

project-open requires the existence of file containing the project id with filename .project_id in the project's root directory. If it doesn't exist, the user is prompted to create it, and for safety it is automatically added to the .gitignore file.

Configuration

In order to use the pivotal-github gem, you need to configure a post-receive hook for your repository. At GitHub, navigate to Settings > Service Hooks > Pivotal Tracker and paste in your Pivotal Tracker API token. (To find your Pivotal Tracker API token, go to your user profile and scroll to the bottom.) Be sure to check the Active box to activate the post-receive hook. At Bitbucket, click on the gear icon to view the settings, click on Services, select Pivotal Tracker, and paste in your Pivotal Tracker API key. In addition, the git story-accept command requires the existence of .api_token and .project_id files containing the Pivotal Tracker API token and project id, respectively.

The pivotal-github command names follow the Git convention of being verbose (e.g., unlike Subversion, Git doesn't natively support co for checkout), but I recommend setting up aliases as necessary. Here are some suggestions, formatted so that they can be pasted directly into a terminal window:

git config --global alias.sc story-commit
git config --global alias.sm story-merge
git config --global alias.spr story-pull-request
git config --global alias.sa story-accept

I also recommend setting up an alias for git push-branch from git-utils:

git config --global alias.pb push-branch

A single-developer workflow would then look like this:

$ git co -b add-markdown-support-62831853
$ git pb
<work>
$ git sc -am "Added foo"
$ git push
<more work>
$ git sc -am "Added bar"
<complete story>
$ git sc -f -am "Added baz"
$ git push
$ git sync
$ git rebase master
$ git sm
$ git sa

Here git sync is from git-utils.

Workflow with integrated code reivew

The pivotal-github gem is designed to support a workflow involving integrated code review, which has the usual benefits: at least two pairs of eyes see any committed code, and at least two brains know basically what the committed code does. The cost is that having a second developer involved can slow you down. I suggest using your judgment to determine which workflow makes the most sense on a story-by-story basis.

Here's the process in detail:

Developer #1 (Alice)

  1. Start an issue at Pivotal Tracker and copy the story id to your buffer
  2. Create a branch in the local Git repository containing the story id and a brief description: git checkout -b add-markdown-support-62831853
  3. Create a remote branch at GitHub using git push-branch
  4. Use git story-commit to make commits, which includes the story number in the commit message: git story-commit -am "Add syntax highlighting"
  5. Continue pushing up after each commit using git push as usual
  6. When done with the story, add -f to mark the story as Finished using git story-commit -fam "Add paragraph breaks" or as Delivered using git story-commit -dam "Add paragraph breaks"
  7. Rebase against master using git sync followed by git rebase master or git rebase master --interactive (optionally squashing commit messages as described in the article A Git Workflow for Agile Teams)
  8. Push up with git push
  9. At the GitHub page for the repo, select Branches and submit a pull request
  10. (On OS X, replace the previous two steps with git story-pull-request)
  11. Assign the pull request to Bob at GitHub
  12. On the Pivotal Tracker story, change the Owner to Bob
  13. Continue working, taking care to branch off of the current story branch if its changes are required to continue

Rather than immediately submitting a pull request, Alice can also continue by branching off the previous story branch, working on a set of related features, and then issue Bob a pull request for the final branch when she reaches a natural stopping place.

Developer #2 (Bob)

  1. Select Pull Requests at GitHub and review the pull request diffs
  2. If acceptable, merge the pull request into master, run git pull on master to pull in the changes, and run git story-accept to mark the corresponding stories accepted
  3. If not acceptable, manually change the state at Pivotal Tracker to Rejected and leave a note (at GitHub or at Pivotal Tracker) indicating the reason
  4. If the branch can't be automatically merged, mark the story as Rejected

Developer #1 (Alice)

  1. After getting the GitHub notification that the pull request has been merged, mark the Pivotal Tracker story finished (unless assigned to Bob)
  2. If the pull request was rejected, make the necessary changes and follow the previous steps above

Merge conflicts

This section contains some suggestions for resolving merge conflicts. First, set up a visual merge tool by installing diffmerge. Then add the following to the .gitconfig file in your home directory:

[mergetool "diffmerge"]
  cmd = diffmerge --merge --result=$MERGED $LOCAL $BASE $REMOTE
  trustExitCode = false

When the branch can't automatically be merged at GitHub, follow these steps:

Devleloper #1 (Alice)

  1. While on the story branch, run git sync
  2. Rebase against master with git rebase master or merge with master using git merge master
  3. Either handle resulting conflicts by hand or use the visual merge tool: git mergetool
  4. Commit the change: git commit -a
  5. Push up the modified branch: git push
  6. (experimental) Add a Chore to revisit the pull request and assign to Developer #2 (Bob)

Now Bob should be able to merge in the pull request automatically using the nice big green button at GitHub.

Contributing

  1. Fork it
  2. Run the tests with rspec spec/
  3. Create your feature branch (git checkout -b my-new-feature)
  4. Add failing tests, then add the feature
  5. Commit your changes (git commit -am 'Add some feature')
  6. Push to the branch (git push origin my-new-feature)
  7. Create new Pull Request