=head1 NAME

Test::Database::Tutorial - How to use Test::Database

=head1 INTRODUCTION

The goal of the L<Test::Database> module is to provide easy to use test
databases for test scripts that need them.

=head2 The problem

Until now, when a test script needed a database, it either used SQLite
(or some other easy to setup database), or required some environment
variables to be present, or used default credentials, or even set up
the database by itself.

Most of those methods have pros and cons:

=over 4

=item *

using SQLite

No setup needed, but the test script can only use SQLite's dialect of SQL.
So much for portability across database engines.

=item *

using environment variables

The environment variables are different for every module to test, and
usually only the main developers/testers know about them. Since most of
the CPAN testers probably don't bother setting them up, these modules
are most certainly undertested.

=item *

using default credentials

Typically using C<'root'> and C<''> to connect to the C<test> MySQL
database, these test script assume a default installation on the host
system. These credentials often provide full access to the database
engine, which is a security risk in itself (see below).

=item *

setting up the database by itself

This method usually uses the default credentials to access an account
with enough privileges to create a database. The host system data may
be at risk!

=back


=head2 A solution: L<Test::Database>

Many modules use a database to store their data, and often support
several database engines.

Wouldn't it be nice to be able to test on all the supported databases
that are available on the test system? Without breaking (into) anything?

This is the goal of the L<Test::Database> module. It supports:

=over 4

=item *

getting DSN information from a list of pre-configured database and engines

=item *

automatic detection of "file-based" database engines (typically, SQLite).

=back

The rest of this document describes various use cases for L<Test::Database>.


=head1 MODULE AND TEST AUTHOR

L<Test::Database> has a single interface for test authors:

    my @handles = Test::Database->handles( @requests );

C<@request> is a list of "requests" for databases handles. Requests
must declare the DBD they expect, and can optionally add version-based
limitations (only available for drivers supported by L<Test::Database>).

The handles returned are objects of the L<Test::Database::Handle> class.

The data contained in the database is never destroyed or cleaned
up by L<Test::Database>, so it's perfectly fine to have a startup script
that will setup the necessary tables and test data, several tests scripts
that will build and update the data, and a eventually a teardown script
that will drop all created tables.

L<Test::Database> can return two types of databases handles:

=over 4

=item *

either a handle to a newly created database (created especially at the
test script's request)

=item *

or a handle to an already existing database

=back

There is no way for the test script to tell the difference.

In any case, the database is assumed to provide C<DROP TABLE> and
C<CREATE TABLE> rights, and the test script is by definition allowed
to do whatever it pleases with the tables that exist in the database.

Note that L<Test::Database> supports any DSN, not just those for which
it has a driver. If your module supports Oracle, you can add C<'Oracle'>
to your list of requests, and if the host owner configured a C<dsn>
pointing at an Oracle database, then it will be available for your tests.

=head2 Specific database support

It is possible to request specific versions of a database engine.

    use Test::Database;

    # request database handles for all available databases
    my @handles = Test::Database->handles();

    # or for only the databases we support
    my @handles = Test::Database->handles(
        { dbd => 'SQLite' },
        { dbd => 'SQLite2' },
        { dbd => 'mysql', min_version => '4.0' },
    );

See L<Test::Database> documentation for details about how
to write a request.

=head2 Testing on a development box

The first systems on which you are going to test your module are the
ones you own. On these system, it's up to you to configure the databases
you want to make available.

A typical F<~/.test-database> configuration file would look like this:

    dsn      = dbi:mysql:database=test
    username = root

    dsn      = dbi:Pg:database=test
    username = postgres

    dsn      = dbi:Oracle:test

There is no need to add C<dsn> sections for file-based drivers
(at least the ones that have a corresponding L<Test::Database::Driver>),
since the module will automatically detect the available ones and create
databases as needed.

To find out which of the DBD that L<Test::Database> supports are
installed, use the following one-liner:

    $ perl -MTest::Database -le 'print for Test::Database->list_drivers("available")'
    DBM
    SQLite
    mysql

With no parameter, it will return the list of configured ones:

    $ perl -MTest::Database -le 'print for Test::Database->list_drivers()'
    DBM
    SQLite


=head1 CPAN TESTER

The main goal of L<Test::Database> from the point of view of a tester
is: "configure once, test everything".

As a CPAN tester, once you have installed L<Test::Database>, you
should edit the local equivalent of F<~/.test-database> for the
user that will be running the CPAN test suites.

=head2 C<dsn> versus C<driver_dsn>

C<dsn> sections define the information needed to connect to a single
database. Any database listed here can be used by any test script that
requests it.

C<driver_dsn> sections define the information needed to connect to a
database engine (a "driver") with sufficient rights to run a
C<CREATE DATABASE> command. This allows L<Test::Database> to create the
databases on demand, thus ensuring every test suite will get a specific
database.

If you have file-based database engine, there is nothing to setup, as
L<Test::Database> is able to detect available file-based engines and
use them as needed.

Other database engines like C<mysql> and C<Pg> require a little more
configuration. For example, here's the content of my F<~/.test-database>
configuration file:

    driver_dsn  = dbi:mysql:
    username    = root
    
    driver_dsn  = dbi:Pg:
    username    = postgres

For C<Pg>, I had to edit the F<pg_hba.cong> file in F</etc> to make sure
anyone would be able to connect as the C<postgres> user, for example.

=head2 Several test hosts accessing the same database engine

If you have a large scale testing setup, you may want to setup a single
MySQL or Postgres instance for all your test hosts, rather than one per
test host.

Databases created by L<Test::Database::Driver> (using a configured
C<driver_dsn> have a name built after the following template:
C<tdd_I<driver>_I<login>_I<n>>, where I<driver> is the DBD name, I<login>
is the login of the user running L<Test::Database> and I<n> a number that

If the same database server is used by several host running
L<Test::Database> from the same user account, there is a race condition
during with two different host may try to create the a database with
the same name. A simple trick to avoid this is to add a C<key> section
to the F<~/.test-database> configuration file.

If the C<key> entry exists, the template used by L<Test::Database::Driver>
to create new databases is C<tdd_I<driver>_I<login>_I<key>_I<n>>.

=head2 Cleaning the test drivers

When given a C<driver_dsn>, L<Test::Database> will use it to create a
database for each test suite that requests one. Some mapping information
is created to ensure the same test suite always receives a handle to
the same database. (The mapping of test suite to database is based on
the current working directory when L<Test::Database> is loaded).

After a while, your database engine may fill up with unused test databases.

All drivers store their mapping information in the system's temporary
directory, so the mapping information is relatively volatile, which implies
more unused test databases (at least for non file-based drivers, since the
file-based drivers store their database files in the system's temporary
directory too).

The following one-liner will list all the existing databases that were
created by L<Test::Database> in your configured drivers:

    perl -MTest::Database -le 'print join "\n  ", $_->name, $_->databases for Test::Database->drivers'

Example output:

    CSV
      tdd_csv_book_0
      tdd_csv_book_1
    DBM
    SQLite
      tdd_sqlite_book_0
      tdd_sqlite_book_1
    SQLite2
      tdd_sqlite2_book_0
    mysql
      tdd_mysql_book_0
      tdd_mysql_book_1

The following one-liner will drop them all:

    perl -MTest::Database -le 'for$d(Test::Database->drivers){$d->drop_database($_)for$d->databases}'

If a C<key> has been defined in the configuration, only the databases
corresponding to that key will be dropped.

=head1 ADDING SUPPORT FOR A NEW DATABASE ENGINE

L<Test::Database> currently supports the following DBD drivers:
C<CSV>, C<DBM>, C<mysql>, C<Pg>, C<SQLite2>, C<SQLite>.

Adding a new driver requires writing a corresponding
L<Test::Database::Driver> subclass, having the same name as the original
C<DBD> driver.

An example module is provided in F<eg/MyDriver.pm>, and the other
drivers can also be used as an example. See also the I<WRITING A
DRIVER FOR YOUR DATABASE OF CHOICE> section in the documentation for
L<Test::Database::Driver>.


=head1 WHERE DO DSN COME FROM?

The following ASCII-art graph shows where the L<Test::Database::Handle>
objects returned by the C<handles()> method come from:


    ,-------------,      ,-------------,      ,--------------,
    |   DSN from  |      | File-based  |      | Drivers from |
    | config file |      |   drivers   |      |  config file |
    '-------------'      '-------------'      '--------------'
           |                   |                      |
           |                   |    ,-----------,     |
           |                   '--->| Available |<----'
           |                        |  drivers  |
           |                        '-----------'
           |                              |
           |              ,-----------,   |
           '------------->| Available |<--'
                          |    DSN    |
                          '-----------'


Here are a few details about the C<handles()> method works:

=over 4

=item *

L<Test::Database> maintains a list of L<Test::Database::Handle> objects
computed from the DSN listed in the configuration.

The handles matching the request are selected.

=item *

L<Test::Database> also maintains a list of L<Test::Database::Driver>
objects computed from the list of supported file-based drivers that are
locally available and from the list in the configuration file.

The list of matching drivers is computed from the requests.  Each driver
is then requested to provide an existing database (using its existing
mapping information) or to create one if needed, and returns the
corresponding L<Test::Database::Handle> objects.

=item *

Finally, all the collected L<Test::Database::Handle> objects are returned.

=back

So, without any configuration, L<Test::Database> will only be able to
provide file-based databases. It is also recommended to B<not> put DSN
or driver information for the file-based database engines that have
a corresponding L<Test::Database::Driver> class, since it will cause
C<handles()> to return several handles for the same database engine.

=head1 AUTHOR

Philippe Bruhat (BooK), C<< <book@cpan.org> >>

=head1 COPYRIGHT

Copyright 2009-2010 Philippe Bruhat (BooK), all rights reserved.

=head1 LICENSE

You can redistribute this tutorial and/or modify it under the same terms
as Perl itself.

=cut