Revised BestPractices

[gitmo/Moose.git] / lib / Moose / Manual / Contributing.pod

=pod

=head1 NAME

Moose::Manual::Contributing - How to get involved in Moose

=head1 GETTING INVOLVED

Moose is an open project, and we are always willing to accept bug fixes,
more tests, and documentation patches. Commit bits are given out freely, and
the L</STANDARD WORKFLOW> is very simple. The general gist is: clone the Git
repository, create a new topic branch, hack away, then find a committer to
review your changes.

Note that this document applies to both Moose and L<Class::MOP> development.

=head1 NEW FEATURES

Moose already has a fairly large feature set, and we are currently
B<not> looking to add any major new features to it. If you have an
idea for a new feature in Moose, you are invited instead to create a
MooseX module first.

At this stage, no new features will even be considered for addition
into the core without first being vetted as a MooseX module, unless
it is absolutely 100% impossible to implement the feature outside the
core.

If you think it is 100% impossible, please come discuss it with us on IRC or
via e-mail. However, your feature may need a small hook in the core, or a
refactoring of some core modules, and we are definitely open to that.

Moose was built from the ground up with the idea of being highly
extensible, and quite often the feature requests we see can be
implemented through a couple of small and well placed extensions. Try
it, it is much easier than you might think.

=head1 PEOPLE

As Moose has matured, some structure has emerged in the process.

=over

=item Contributors - people creating a topic or branch

You.

If you have commit access, you can create a topic on the main Moose.git,
otherwise either give us your SSH key or create your own clone of the
L<git://git.moose.perl.org/Moose.git> repository or fork of the GitHub mirror.

The relevant repository URIs are:

=over

=item Read-Only

L<git://git.moose.perl.org/Moose.git>

=item Read+Write

L<gitmo@git.moose.perl.org:Moose.git>

=back

=item Core Committers - people reviewing and merging a branch

These people have worked with the Moose codebase for a while.

They've been responsible for large features or branches and can help review
your changes and apply them to the master branch using the basic
L</APPROVAL WORKFLOW>.

They are also fairly well versed in Git, in order to merge the branches with
no mistakes (especially when the merge fails), and to provide advice to
contributors.

=item Cabal - people who can release moose

These people are the ones who have co-maint on Moose itself and can create a
release. They're listed under L<Moose/CABAL> in the Moose documentation. They
merge from Master to Stable.

=back

=head1 BRANCH LAYOUT

The repository is divided into several branches to make maintenance easier for
everyone involved. The branches below are ordered by level of stability.

=over

=item Stable (refs/heads/stable)

The branch from which releases are cut. When making a new release, the
release manager merges from master to stable. The stable branch is only
updated by someone from the Cabal during a release.

=item Master (refs/heads/master)

The branch for new development. This branch is merged into and branched from.

=item Branches (refs/heads/*)

Large community branches for big development "projects".

=item Topics (refs/heads/topic/*)

Small personal branches that have been published for review, but can get
freely rebased. Targeted features that may span a handful of commits.

Any change or bugfix should be created in a topic branch.

=back

=head1 STANDARD WORKFLOW

    # update your copy of master
    git checkout master
    git pull --rebase

    # create a new topic branch
    git checkout -b topic/my-feature origin/master

    # hack, commit, feel free to break fast forward
    git commit --amend                       # allowed
    git rebase --interactive                 # allowed
    git push --force origin topic/my_feature # allowed

Then ask for a review/approval (see L</APPROVAL WORKFLOW>), and merge
to master. If it merges cleanly and nobody has any objections, then it
can be pushed to master.

If it doesn't merge as a fast forward, the author of the branch needs to run

    git remote update
    git rebase origin/master # or merge

and bring the branch up to date, so that it can be merged as a fast forward
into master.

No actual merging (as in a human resolving conflicts) should be done when
merging into master, only from master into other branches.

=head2 Preparing a topic branch

Before a merge, a topic branch can be cleaned up by the author.

This can be done using interactive rebase to combine commits, etc, or even
C<git merge --squash> to make the whole topic into a single commit.

Structuring changes like this makes it easier to apply git revert at a later
date, and encourages a clean and descriptive history that documents what the
author was trying to do, without the various hangups that happened while they
were trying to do it (commits like "oops forgot that file" are not only
unnecessary noise, they also make running things like git bisect or git revert
harder).

However, by far the biggest benefit is that the number of commits that go into
master is eventually reduced, and they are simple and coherent, making it much
easier for people maintaining branches to stay up to date.

All large changes should be documented in L<Moose::Manual::Delta>.

=head1 APPROVAL WORKFLOW

Moose is an open project but it is also an increasingly important one. Many
modules depend on Moose being stable. Therefore, we have a basic set of
criteria for reviewing and merging branches. What follows is a set of rough
guidelines that ensures all new code is properly vetted before it is merged to
the master branch.

It should be noted that if you want your specific branch to be approved, it is
B<your> responsibility to follow this process and advocate for your branch.
The preferred way is to send a request to the mailing list for review/approval,
this allows us to better keep track of the branches awaiting approval and those
which have been approved.

=over 4

=item Small bug fixes, doc patches and additional passing tests.

These items don't really require approval beyond one of the core contributors
just doing a simple review.

=item Larger bug fixes, doc additions and TODO or failing tests.

Larger bug fixes should be reviewed by at least one cabal member and should be
tested using the F<xt/author/test-my-dependents.t> test.

New documentation is always welcome, but should also be reviewed by a cabal
member for accuracy.

TODO tests are basically feature requests, see our L</NEW FEATURES> section
for more information on that. If your feature needs core support, create a
topic/ branch using the L</STANDARD WORKFLOW> and start hacking away.

Failing tests are basically bug reports. You should find a core contributor
and/or cabal member to see if it is a real bug, then submit the bug and your
test to the RT queue. Source control is not a bug reporting tool.

=item New user-facing features.

Anything that creates a new user-visible feature needs to be approved by
B<more than one> cabal member.

Make sure you have reviewed L</NEW FEATURES> to be sure that you are following
the guidelines. Do not be surprised if a new feature is rejected for the core.

=item New internals features.

New features for Moose internals are less restrictive than user facing
features, but still require approval by B<at least one> cabal member.

Ideally you will have run the F<test-my-dependents.t> script to be sure you
are not breaking any MooseX module or causing any other unforeseen havoc. If
you do this (rather than make us do it), it will only help to hasten your
branch's approval.

=item Backwards incompatible changes.

Anything that breaks backwards compatibility must be discussed by the cabal
and agreed to by a majority of the members.

We have a policy for what we see as sane L</BACKWARDS COMPATIBILITY> for
Moose. If your changes break back-compat, you must be ready to discuss and
defend your change.

=back

=head1 RELEASE WORKFLOW

    git checkout master
    # edit for final version bumping, changelogging, etc
    # prepare release (test suite etc)
    perl-reversion -bump
    make manifest
    git commit
    git checkout stable
    git merge master # must be a fast forward
    git push both
    shipit # does not ship the tarball, but does everything else
    cpan-upload ~/shipit-dist/Moose-X.YZ.tar.gz

Development releases are made without merging into the stable branch.

=head2 Release How-To

Moose (and L<Class::MOP>) releases fall into two categories, each with their
own level of release preparation. A minor release is one which does not
include any API changes, deprecations, and so on. In that case, it is
sufficient to simply test the release candidate against a few different
different Perls. Testing should be done against at least two recent major
version of Perl (5.8.8 and 5.10.1, for example). If you have more versions
available, you are encouraged to test them all. However, we do not put a lot
of effort into supporting older 5.8.x releases.

For major releases which include an API change or deprecation, you should run
the F<xt/author/test-my-dependents.t> test. This tests a long list of MooseX
and other Moose-using modules from CPAN. In order to run this script, you must
arrange to have the new version of Moose and/or Class::MOP in Perl's include
path. You can use C<prove -b> and C<prove -I>, install the module, or fiddle
with the C<PERL5LIB> environment variable, whatever makes you happy.

This test downloads each module from CPAN, runs its tests, and logs failures
and warnings to a set of files named F<test-mydeps-$$-*.log>. If there are
failures or warnings, please work with the authors of the modules in question
to fix them. If the module author simply isn't available or does not want to
fix the bug, it is okay to make a release.

Regardless of whether or not a new module is available, any breakages should
be noted in the conflicts list in the distribution's F<Makefile.PL>.

Both Class::MOP and Moose have a F<.shipit> file you can use to make sure the
release goes smoothly. You are strongly encouraged to use this instead of
doing the final release steps by hand.

=head1 EMERGENCY BUG WORKFLOW (for immediate release)

Anyone can create the necessary fix by branching off of the stable branch:

    git remote update
    git checkout -b topic/my-emergency-fix origin/stable
    # hack
    git commit

Then a cabal member merges into stable:

    git checkout stable
    git merge topic/my-emergency-fix
    git push
    # release
    git checkout master
    git merge stable

=head1 PROJECT WORKFLOW

For longer lasting branches, we use a subversion style branch layout, where
master is routinely merged into the branch. Rebasing is allowed as long as all
the branch contributors are using C<git pull --rebase> properly.

C<commit --amend>, C<rebase --interactive>, etc. are not allowed, and should
only be done in topic branches. Committing to master is still done with the
same review process as a topic branch, and the branch must merge as a fast
forward.

This is pretty much the way we're doing branches for large-ish things right
now.

Obviously there is no technical limitation on the number of branches. You can
freely create topic branches off of project branches, or sub projects inside
larger projects freely. Such branches should incorporate the name of the branch
they were made off so that people don't accidentally assume they should be
merged into master:

    git checkout -b my-project--topic/foo my-project

(unfortunately Git will not allow C<my-project/foo> as a branch name if
C<my-project> is a valid ref).

=head1 THE "PU" BRANCH

To make things easier for longer lived branches (whether topics or projects),
the 'pu' branch is basically what happens if you merge all of the branches and
topics together with master.

We can update this as necessary (e.g. on a weekly basis if there is merit),
notifying the authors of the respective branches if their branches did not merge
(and why).

To update 'pu':

    git checkout pu
    git remote update
    git reset --hard origin/master
    git merge @all_the_branches

If the merge is clean, 'pu' is updated with C<push --force>.

If the merge is not clean, the offending branch is removed from
C<@all_the_branches>, with a small note of the conflict, and we try again.

The authors of the failed branches should be told to try to merge their branch
into 'pu', to see how their branch interacts with other branches.

'pu' is probably broken most of the time, but lets us know how the different
branches interact.

=head1 BRANCH ARCHIVAL

Merged branches should be deleted.

Failed branches may be kept, but consider moving to refs/attic/ (e.g.
http://danns.co.uk/node/295) to keep git branch -l current.

Any branch that could still realistically be merged in the future, even if it
hasn't had work recently, should not be archived.

=head1 TESTS, TESTS, TESTS

If you write I<any> code for Moose or Class::MOP, you B<must> add
tests for that code. If you do not write tests then we cannot
guarantee your change will not be removed or altered at a later date,
as there is nothing to confirm this is desired behavior.

If your code change/addition is deep within the bowels of
Moose/Class::MOP and your test exercises this feature in a non-obvious
way, please add some comments either near the code in question or in
the test so that others know.

We also greatly appreciate documentation to go with your changes, and
an entry in the Changes file. Make sure to give yourself credit!

=head1 BACKWARDS COMPATIBILITY

Change is inevitable, and Moose is not immune to this. We do our best
to maintain backwards compatibility, but we do not want the code base
to become overburdened by this. This is not to say that we will be
frivolous with our changes, quite the opposite, just that we are not
afraid of change and will do our best to keep it as painless as
possible for the end user.

The rule is that if you do something that is not backwards compatible, you
B<must> do I<at least> one deprecation cycle (more if it is larger change).
For really larger or radical changes dev releases may be needed as well (the
Cabal will decide on this on a case-per-case basis).

Our policy with deprecation is that each deprecation should go through several
stages. First, we simply add a deprecation notice the documentation in
F<Changes> and L<Moose::Manual::Delta>. In a future release, we then make the
deprecated feature warn loudly and often so that users will have time to fix
their usages. Finally, the feature is removed in a later release.

All backwards incompatible changes B<must> be documented in
L<Moose::Manual::Delta>. Make sure to document any useful tips or workarounds
for the change in that document.

=head1 AUTHOR

Stevan Little E<lt>stevan@iinteractive.comE<gt>

Chris (perigrin) Prather

Yuval (nothingmuch) Kogman

=head1 COPYRIGHT AND LICENSE

Copyright 2009 by Infinity Interactive, Inc.

L<http://www.iinteractive.com>

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

=cut