=head1 NAME

Extending.pod - How to extend Wiki::Toolkit with your own plugins.

=head1 LIMITATIONS

The extension mechanism is currently only defined for database-backed
setups, but since nobody has written any other kind of backend I think
we're fine for now.

=head1 THE SIMPLEST WAY

You can extend L<Wiki::Toolkit> in a fairly simplified way without the use
of plugins, by supplying a hash of metadata when you write a node. For
example:

  $wiki->write_node( $node, $content, $checksum,
                     { postcode => $postcode }   );

and on node retrieval you'll get it back again:

  my %node = $wiki->retrieve_node( $node );
  my $postcode = $node{metadata}{postcode}[0];

You can supply more than one value for each type of metadata:

  $wiki->write_node( $node, $content, $checksum,
      { postcode => "W6 9PL",
        category => [ "Thai Food", "Restaurant", "Hammersmith" ] } );

And get back a list of nodes which have a given value for a given
metadata type:

  my @nodes = $wiki->list_nodes_by_metadata(
      metadata_type  => "category",
      metadata_value => "Hammersmith" );

For anything more complicated you will need to write a plugin.

=head1 PLUGIN BASE CLASS

Plugins should inherit from C<Wiki::Toolkit::Plugin>. This base class
provides the following methods to give access to a C<Wiki::Toolkit>
object's backends:

=over 4

=item * C<datastore> - returns the store object

=item * C<indexer> - returns the search object

=item * C<formatter> - returns the formatter object

=back

If you want these methods to return anything useful then call

  $wiki->register_plugin( plugin => $plugin);

before calling say

  my %node_data = $plugin->datastore->retrieve_node( "Foo" );

=head1 CALLING API

  my $plugin = Wiki::Toolkit::Plugin::Foo->new( ...any required args... );
  $wiki->register_plugin( plugin => $plugin );

  $wiki->write_node( "Test Node" ,"Test", $checksum,
                     { foo_data => { a => "apple",
                                     b => "banana" }
                     } );

  my $bee = $plugin->get_word( node => "Test Node", letter => "b" );

or

  my $plugin = OpenGuides::London::Underground->new;
  $wiki->register_plugin( plugin => $plugin );
  $wiki->write_node( "Hammersmith Station", "a station", $checksum,
                     { tube_data => [
                         { line => "Piccadilly",
                           direction => "Eastbound",
                           next_station => "Baron's Court Station"
                         },
                         { line => "Piccadilly",
                           direction => "Westbound",
                           next_station => "Acton Town Station"
                         }
                                    ]
                      }
                    );

  # Put more data in, then

  my @route = $plugin->find_route( from => "Holborn Station",
                                   to   => "Acton Town Station" );


=head1 STORE ACCESS

A plugin named Wiki::Toolkit::Plugin::Foo may access the backend
database directly like so:

=over 4

=item * Read-only access to any table

=item * Read-write access to any table whose name begins with
C<"p_" . $Wiki::Toolkit::Plugin::Foo::plugin_key . "_">

C<$Wiki::Toolkit::Plugin::Foo::plugin_key> should be different from the
keys of all other plugins. No, I haven't set anything up to ensure this.

=back

=head1 REQUIREMENTS FOR PLUGIN AUTHORS

Either be database-agnostic, or state clearly in your docs which
databases you support, and handle errors nicely.  Be aware that
non-database backends may exist in the future.

Be aware of whether you need to check for locks explicitly in
different databases (see code of Wiki::Toolkit::Store::* to find out).

=head1 REQUIRED METHODS

=over 4

=item B<on_register>

Check that any tables you require are set up, and set them up if not.

=back

=head1 OPTIONAL METHODS

=over 4

=item B<post_write>

This will be called every time a node is written, with the arguments like so:

  $plugin->post_write( node     => $node_name,
                       version  => $version_number,
                       content  => $content,
                       metadata => \%user_defined_metadata );

This will happen after the node data is all written, but before any
lock is released.

We could probably reimplement the searches as plugins like this if we
want to, but this will require writing extra backends for
Search::InvertedIndex so it can work within the same database.

The user-defined metadata will already have been stored in the backend
but it is available here for you to do what you will with it.

Its return value should be true on success and false on error.

=item B<post_read>

B<THIS IS NOT YET IMPLEMENTED.>

This will be called every time a node is read, with the arguments like so:

  $plugin->post_read( node     => $node_name,
                      version  => $version_number,
                      content  => $content,
                      metadata => \%user_defined_metadata );

It cannot affect the data returned to the caller. It should be used for its
side-effects, for example tracking the number of times a given node is
accessed.

Its return value should be true on success and false on error.

=back

=head1 PLUGIN CONFLICTS

What if we have more than one plugin registered? What if we change the
mechanism to allow the plugins to change the data stored in the
database/returned to the caller?