Master to 4.2.8

[usit-rt.git] / docs / hacking.pod

=head1 Development of RT

RT's source code is stored in a C<git> repository.  If you are not
familiar with git, see L</git quickstart>, below, for a short tutorial
which will give you enough information to get started submitting patches
to RT.

The rest of this document details conventions and tips surrounding the
organization of RT's version control, source code conventions, and how
to submit patches.



=head1 Organization of rt.git

The RT source repository is available via git from GitHub; you can
browse it at L<http://github.com/bestpractical/rt/> or obtain a local
copy via:

    git clone git://github.com/bestpractical/rt.git

The bleeding-edge development happens in the C<master> branch.  When a
major release is anticipated, a "trunk" branch will be branched from
this -- for example, C<4.0-trunk>.  This will allow the trunk to
stabilize while feature development continues on C<master>.
Additionally, as a release is impending for a particular series, a
release engineering branch will be created, named, for example
C<4.0.0-releng>.

New feature development should always be based off of the C<master>
branch.  Branches to fix bugs should be based off of whichever trunk the
bug was first found in.  If you found the bug in your RT 4.0.0 install,
you'd branch from 4.0-trunk.

Branches should be named based on the trunk they are branched
from -- which is to say, the earliest branch they might be merged into.
For example, a bugfix branched from C<4.0-trunk> might be named
C<4.0/fail-taint-mode-early>.  A feature branched from C<master> when
there exists a C<4.0-trunk> but no C<4.2-trunk> might be named
C<4.2/rename-LogToScreen>.  For consistency, branches should use dashes,
not underscores, to separate words.  Branches which are destined for
4.2, but which are branched from 4.0 (to provide for easy extraction as
a 4.0 extension) should be named 4.2-on-4.0/branch-name.

Branches should be reviewed by another developer before being merged.
Reviewers should make sure that the branch accomplishes what it claims
to, and does not introduce any unwanted behavior in doing so.  Commit
messages explain the B<why> as much as the B<what> of each commit, and
not include extranous changes.


=head1 Code conventions

The RT codebase is more than ten years old; as such, there are sections
which do not (yet) conform to the guidelines below.  Please attempt to
follow the guidelines, even if the code surrounding your changes does
not yet.

RT also includes a F<.perltidyrc> in its top-level which encodes many of
the conventions.

=over

=item Indentation

Each level of indentation should be four spaces; tabs should never be
used for indentation.

=back

=head1 Internationalization

RT has been translated into several dozen languages. We use Launchpad
( https://translations.launchpad.net/rt ) to crowdsource our
translations into C<po> files. RT uses L<Locale::Maketext> to
localize its user interface.

Your first stop on this magical journey of internationalization
is L<Locale::Maketext::TPJ13>, which explains the whys of
L<Locale::Maketext>. RT uses most of the features developed in that
article.

Strings that are displayed to users should be passed through the
C<loc("...")> function or the C<< <&|/l&>...</&> >> Mason template.
C<loc> and C</l> both take parameters, which are used in place of
string interpolation (much like C<sprintf>). It's acceptable to use
HTML in C</l> calls, especially for bold and emphasis. However, you
should limit the amount of HTML that translators must keep exactly
correct, which means avoid including tags that wrap the entire
translatable string, especially C<< <p> >>.

    <p><&|/l, $button &>Do <em>not</em> click [_1]</&></p> # ok

    <&|/l, $button &><p>Do <em>not</em> click [_1]</p></&> # not ok

In a few places in RT we also pass HTML as parameters to C<loc()>
so that translators do not have to reproduce it exactly, and we can
also change it more freely. For example:

    <&|/l,
        '<a href="http://www.gnu.org/licenses/gpl-2.0.html">',
        '</a>',
    &>Distributed under [_1]version 2 of the GNU GPL[_2].</&>

F<devel/tools/extract-message-catalog> looks for C<loc("...")> and
C<< <&|/l&>...</&> >> in our source code to pick out translatable
strings, clean them up, and put them into F<share/po> files. We use
our C<.po> files not only to populate L<Locale::Maketext>'s lexicons,
but also to sync new translatable strings and translations with
Launchpad. This Launchpad sync is typically done early during the
freeze of RC releases to give our volunteer translators time to
translate all the new strings which, because of the RC freeze, won't
continue changing.

Because C<loc()> and C</l> are used to generate strings for human
eyes, they generally must be used "close to the browser". These are
directly in Mason templates, or in functions that return text that
will be passed through Mason. However, in many places in RT we have
hardcoded strings which need translations. For example, the C<$RIGHTS>
hash in F<lib/RT/Queue.pm> maps rights' names (which must be
translatable) to their descriptions (which also must be translatable).
However, when we're declaring such structures, we do not want to
translate them straight away. RT uses English internally, including
in its web forms, so we do not want to localize rights' names except
for display, otherwise things might break weirdly when you check
if a user has the "Superusuario" right. Furthermore, when we're
declaring such data structures at compile time, there is no current
user to select which language to use for localization. Thus, we
cannot call C<loc()> when declaring C<$RIGHTS> and other similar
places.

For this reason, F<devel/tools/extract-message-catalog> lets you
denote translatable strings with comments. That's what the C<#loc_pair>
comments in the C<$RIGHTS> hash in F<lib/RT/Queue.pm> indicate.
Since we have those comments, our toolchain will put the rights'
names and descriptions into F<share/po> files, which enables
translation by our lovely volunteers. Later on, when RT displays
information about rights in the web UI, we'll pass the right's name
through C<loc>, and L<Locale::Maketext> will then be able to find
our "Superusuario". So although we never used a literal
C<loc("SuperUser")>, we still get its effects thanks to the
C<#loc_pair> comments and using C<loc($RightName)>.

C<#loc_pair> is used for declaring that the both the key and value
of a particular C<< key => value >> pair are translatable. There
are other markers that you can use.

C<#loc> is used for declaring that a particular string is translatable.
Its parsing is pretty strict so you can use it to declare that only
the value of a particular C<< key => value >> pair is translatable.

C<#loc_left_pair> is used for declaring that the I<key> of a
particular C<< key => value >> pair is translatable. This is of
very limited usefulness.

C<#loc_right_pair> does NOT exist. C<#loc> works in such cases since its
parser does not extend beyond the string at the end of a line.  However,
if the string is I<not> at the end of the line, C<#loc{word}> declares
that the value associated with the key I<word> (earlier on the same
line) is to be loc'd.  This is useful for inline hashes:

    # Note the string "baz" is to be loc'd
    foo => { bar => "baz", troz => "zort" },  # loc{bar}

=head1 Development tips

=head2 Setting up a development environment

=head2 Test suite

RT also comes with a fairly complete test suite.  To run it, you will
need to set environment variables to a database user and password which
can create and drop databases:

    export RT_DBA_USER=root
    export RT_DBA_PASSWORD=

You'll need to configure RT and make sure you have all the dependencies
before running tests.  To do this in place without installing:

    ./configure.ac --with-my-user-group --enable-layout=inplace --enable-developer
    make testdeps
    make fixdeps

Adjust the relevant database options as necessary if you want to test on
Postgres, Oracle, or SQLite.  The default is MySQL.

To run the test suite:

    make test

If you have multiple processors, you can run the test suite in parallel,
which will be significantly faster:

    make test-parallel

The C<*-trunk> and C<master> branches are expected to always be passing
all tests.  While it is acceptable to break tests in an intermediate
commit, a branch which does not pass tests will not be merged.  Ideally,
commits which fix a bug should also include a testcase which fails
before the fix and succeeds after.



=head1 git quickstart

=over

=item 1.

You will first need to obtain a copy of git; this is accomplished via
C<sudo yum install git> in RedHat and derivatives, or C<sudo apt-get
install git> for Debian or Ubuntu.

=item 2.

Next, obtain a copy of the RT source from git:

    git clone git://github.com/bestpractical/rt.git
    cd rt

=item 3.

Configure git to know your name and email address; git uses these when
it makes commits.

    git config user.email your.email@example.com
    git config user.name Examp L. Name

=item 4.

Switch to the appropriate point to base your work on; this is generally
C<origin/> followed by the major version, followed by C<-trunk>.  For
example, if your bug was observed in version 3.8.9, you would choose
C<origin/3.8-trunk>; if it was in 4.0.0, you would choose
C<origin/4.0-trunk>.  New features should be based on C<origin/master>.

    git checkout --track origin/4.0-trunk

=item 5.

Give your branch a name based on what you are attempting to accomplish.
We suggest that branch names be lower-case and separate words with
dashes, but this branch name is purely for your own reference.

    git branch -m gnupg-encryption

=item 6.

Edit the source tree to make your changes.  A few commands you may find
useful in doing so are listed below.

To see what files you have changed:

    git status

To see a line-by-line list of changes:

    git diff

To revert a file to the original version:

    git checkout path/to/file

To revert only individual parts of a file:

    git checkout -p path/to/file

See L</Development tips> for more tips for working with the RT codebase.

=item 7.

Check that you have no extraneous changes using C<git diff>, then commit
your changes:

    git commit -a

You will be prompted to type your commit message.  The first line should
be a short (E<lt> 80 character) summary of the changes, followed by a
blank line, followed by a longer description, if necessary.  The commit
message should not simply restate the diff of which lines were added and
subtracted, but should rather explain B<what> those changes accomplish,
and B<why> they are desired.

If your changes are easily split into multiple components, you may wish
to split your changes into more than one commit; simply return to step 6
and repeat the with the next related change.  If your changes are B<not>
related to each other, you should submit them separately; finish step 9,
then start over from step 4.

=item 8.

Save your commits to patch files:

    git format-patch @{u}

This will print out the names of the files as it creates them.

=item 9.

Attach these files to an email using your standard email client, and
send it to C<rt-devel@bestpractical.com>.

=back

If you have another bug or feature to implement, simply restart the
process at step 4.

=cut