/ 
Dancer-1.3514_04
River stage three • 211 direct dependents • 215 total dependents
/ Dancer::Development::Integration

NAME

Dancer::Development::Integration - guide for Dancer's core-team members

VERSION

version 1.3514_04

DESCRIPTION

This documentation describes the procedure used for integrators to review and merge contributions sent via pull-requests.

Every core-team member should read and apply the procedures described here. This will allow for a better history and more consistency in our ways of handling the (increasing number!) of pull requests.

TERMS

We will first define the most important terms used in this documentation:

  • PR

    Acronym for Pull Request

  • Contributor

    A GitHub user who had forked and cloned the official Dancer's repo, and who has sent a PR.

  • Integration branch

    This branch is the branch used to merge all contributions. This is a git-flow convention. In Dancer, our integration branch is devel.

    As explained in Dancer::Development, every PR should be based on the integration branch. If not, this is enough to refuse the PR (it makes the life of the integrator much harder if this is not the case).

  • Integrator

    A member of Dancer's core-team who is responsible for reviewing and either rejecting the PR, or merging it into the integration branch.

PROCEDURES

Processing a Pull Request

This procedure describes how an integrator should process a PR.

Let's say the user $user has sent a PR, he has followed the instructions described in Dancer::Development so his work is based on the integration branch (devel).

All the procedure described here is designed to avoid unnecessary recursive-merge, in order to keep a clean and flat history in the integration branch.

Of course, we could just pull from $user into our devel branch, but this would shift the history because of recursive merge, most of the time.

To avoid that, we're going to pull the commits of $user into a temporary branch, and then cherry-pick the commits we want.

In order to have a clean history, like the one we got with git-flow when working on a feature, we're going to do that in a topic branch, named review/$user. Then, this branch will be merged into devel and we will just have to drop it.

First, we make sure we are in sync with origin/devel

git checkout devel
git pull origin devel

Then, from that branch we create a temp sandbox

git checkout -b temp

We pull here from $user

git pull <user repo> <pr/branch>

Here, either the pull was run as a fast-forward or as a recursive merge. If we have a FF, we can forget about the temp branch and do the pull directly in devel. If not, we'll have to cherry-pick the commits by hand.

From devel, we first create the final review branch:

git checkout devel
git checkout -b review/$user

Then we cherry-pick all the commits we want. To know them, we just have to go into temp and inspect the history (with git log).

When we have the list of commits we want:

for commit in C1 C2 C3 ... CN
do
    git cherry-pick $commit
done

(Another option is to use git rebase -i to manually select the list of commits to cherry-pick/rebase.)

Then we can review the code, do whatever we want, maybe add some commits to change something.

When we're happy with the change set, we can merge into devel:

git checkout devel
git merge --no-ff review/$user

Note the --no-ff switch is used to make sure we'll see a nice commit named Merge branch 'review/$user' into devel. This is on purpose and mimic the behaviour of git-flow.

Your local devel branch is now merged, and can be pushed to the remote.

$ git push origin devel

RELEASE CYCLES

We have one main release cycle. This is the release cycle based on the devel branch. We use this branch to build new releases, with new stuff all the new shiny commits we want.

Those release are built with git-flow (with git-flow release) and are then uploaded to CPAN.

Since Dancer 1.2, we also have another parallel release cycle which is what we call the frozen branch. It's a maintenance-only release cycle. That branch is created from the tag of the first release of a stable version (namely a release series with an even minor number).

This branch must be used only for bug-fixing the stable releases. Nothing new should occur in that branch.

Let's take an example with Dancer 1.2003 and Dancer 1.3002.

  • Dancer 1.2003 is the last stable release of Dancer. Its codebase is handled in the frozen branch, that has been created from the tag 1.2000.

  • Dancer 1.3002 is the last release of Dancer. As it belongs to a development series, it can provide new features, code refactoring and deprecations. Its codebase is handled by the integration branch, devel.

  • When a bug is found in 1.2xxx, it's fixed in the frozen branch, and a new release is built from here and then uploaded to CPAN.

  • Whenever the team wants to, they can release new versions of 1.3xxx from the devel branch, using git-flow release start.

  • When the team finds that the current state of devel (namely, the last version of 1.3xxx) is stable and mature enough. They can decide it will be the new stable version.

    Then, a release 1.4000_01 is built from devel, an upload is done to CPAN, and when ready, the 1.40001 can be uploaded the same way.

    From that moment, the master branch is merged into frozen in order to be able to hotfix the frozen branch in the future.

    It's now possible for the team to continue working on new stuff in devel, bumping the version number to 1.5000_01

AUTHOR

This documentation has been written by Alexis Sukrieh <sukria@sukria.net>.

AUTHOR

Dancer Core Developers

COPYRIGHT AND LICENSE

This software is copyright (c) 2010 by Alexis Sukrieh.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.