__END__

=pod

=head1 NAME

CHI::Driver::Development - Manual for developing new CHI drivers

=head1 VERSION

version 0.61

=head1 SYNOPSIS

    package CHI::Driver::MyDriver;
    use Moo;
    use strict;
    use warnings;

    extends 'CHI::Driver';

    has ...;

    sub fetch {
        my ( $self, $key ) = @_;

    }

    sub store {
        my ( $self, $key, $data[, $expires_in] ) = @_;

    }

    sub remove {
        my ( $self, $key ) = @_;

    }

    sub clear {
        my ($self) = @_;

    }

    sub get_keys {
        my ($self) = @_;

    }

    sub get_namespaces {
        my ($self) = @_;

    }

=head1 DESCRIPTION

This document describes how to implement a new CHI driver.

The easiest way to start is to look at existing drivers, such as
L<CHI::Driver::Memory|CHI::Driver::Memory> and
L<CHI::Driver::FastMmap|CHI::Driver::FastMmap>.

=head1 NAMING

If you are going to publicly release your driver, call it
'CHI::Driver::I<something>' so that users can create it with

    CHI->new(driver => 'I<something>');

If it's an internal driver, you can call it whatever you like and create it
like

    CHI->new(driver => '+My::Internal::CHI::Driver');

=head1 MOO / MOOSE

CHI driver classes must be L<Moo|Moo> or L<Moose|Moose> based to be fully
functional, since we use Moose roles to implement various features. For
backward compatibility, non-Moo/Moose drivers will still work at a basic level,
but you will see an error if using any feature requiring a role.

All drivers must directly or indirectly extend L<CHI::Driver|CHI::Driver>.

=head1 NAMESPACE

All cache handles have an assigned namespace that you can access with
C<$self-E<gt>namespace>. You should use the namespace to partition your data
store. That is, two cache objects with different namespaces should be able to
access the same key without any collision.

Examples:

=over

=item *

The Memory driver uses a separate sub-hash inside its main memory hash for each
namespace.

=item *

The File driver uses a separate top-level directory for each namespace.

=item *

The FastMmap driver uses a separate Cache::FastMmap file for each namespace.

=back

=head1 METHODS

=head2 Required methods

The following methods have no default implementation, and MUST be defined by
your subclass:

=over

=item store ( $self, $key, $data[, $expires_in] )

Associate I<$data> with I<$key> in the namespace, overwriting any existing
entry.  Called by L</set>. I<$data> will contain any necessary metadata,
including expiration options, so you can just store it as a single block.

I<$expires_in> is optionally the number of seconds from now when the entry will
expire. This will only be passed if L<CHI/expires_on_backend> is set. If your
driver does not support expiration, or if you'd rather just let CHI manage
expiration, you can ignore this.

=item fetch ( $self, $key )

Returns the data associated with I<$key> in the namespace. Called by L</get>.
The main CHI::Driver superclass will take care of extracting out metadata like
expiration options and determining if the value has expired.

=item remove ( $self, $key )

Remove the data associated with the I<$key> in the namespace.

=item clear ( $self )

Remove all data associated with the namespace. (Technically not required, but
the default implementation, which iterates over all keys and calls L</remove>
for each, is very inefficient).

=back

=head2 Overridable methods

The following methods have a default implementation, but MAY be overridden by
your subclass:

=over

=item BUILD ( $self, $options )

Define the BUILD method if you want to process any options specific to your
driver.  This is a standard Moo/Moose feature.

=item fetch_multi_hashref ( $keys )

Override this if you want to efficiently process multiple fetches. Return a
hash reference from keys to fetched data. If a key is not available, it may be
left out of the hash or paired with undef. The default method will iterate over
I<$keys> and call fetch for each.

This method is called by L<get_multi_arrayref> and L<get_multi_hashref>.

=item store_multi ( $key_data, $options )

Override this if you want to efficiently process multiple stores. I<$key_data>
is a hash of keys and data that should be stored. The default will iterate over
I<$key_data> and call store for each pair.

This method is called by L<set_multi>.

=back

=head2 Optional methods

The following methods have no default implementation, and MAY be defined by
your subclass, but are not required for basic cache operations.

=over

=item get_keys ( $self )

Return all keys in the namespace. It is acceptable to either include or omit
expired keys.

=item get_namespaces ( $self )

Return namespaces associated with the cache. It is acceptable to either include
or omit namespaces with no valid keys.

=back

=head1 DISCARD POLICIES

You can create new discard policies for L<size aware|CHI/SIZE AWARENESS>
caches, to choose items to discard when the cache gets full. For example, the
Memory driver implements an LRU policy.

To implement a discard policy I<foo>, define a subroutine
I<discard_policy_foo>, which takes a driver object and returns a closure that
returns one key each time it is called. The closure should maintain state so
that each key is only returned once.

For example, here's the Memory driver's LRU implementation. It utilizes a hash
containing the last used time for each key.

   sub discard_policy_lru {
       my ($self) = @_;
   
       my $last_used_time = $self->{metadata_for_namespace}->{last_used_time};
       my @keys_in_lru_order =
         sort { $last_used_time->{$a} <=> $last_used_time->{$b} } $self->get_keys;
       return sub {
           shift(@keys_in_lru_order);
       };
   }

You can set the default discard policy for your driver by overriding
default_discard_policy; otherwise the default is 'arbitrary'.

   sub default_discard_policy { 'lru' }

=head1 TESTING

CHI has a standard set of unit tests that should be used to ensure your driver
is fully implementing the CHI API.

To use CHI's tests (replacing I<MyDriver> with the name of your driver):

=over

=item *

Install L<Test::Class> and add it to the build dependencies for your
distribution.

=item *

Add a module called I<CHI::Driver::MyDriver::t::CHIDriverTests> to your
distribution containing:

    package CHI::Driver::MyDriver::t::CHIDriverTests;
    use strict;
    use warnings;
    use CHI::Test;
    use base qw(CHI::t::Driver);

    sub testing_driver_class { 'CHI::Driver::MyDriver' }

    sub new_cache_options {
        my $self = shift;

        return (
            $self->SUPER::new_cache_options(),

            # Any necessary CHI->new parameters for your test driver
        );
    }

    1;

=item *

Add a test script called I<t/CHI-driver-tests.t> to your distribution
containing:

    #!perl -w
    use strict;
    use warnings;
    use CHI::Driver::MyDriver::t::CHIDriverTests;
    CHI::Driver::MyDriver::t::CHIDriverTests->runtests;

=item *

You may need to override other methods in
I<CHI::Driver::MyDriver::t::CHIDriverTests>, e.g. to skip tests that do not
apply to your driver. See CHI::t::Driver::Memory and CHI::t::Driver::File in
this distribution for examples.

=back

=head2 Test cleanup

You are responsible for cleaning up your datastore after tests are done. The
easiest way to do this is to place your datastore wholly inside a temporary
directory, or use a L<guard|Guard> to remove it at process end.

For example, the L<File|CHI::Driver::File>, L<FastMmap|CHI::Driver::FastMmap>,
and L<DBI|CHI::Driver::DBI> tests place all data inside a tempdir that is
automatically cleaned up at process end.

=head1 SEE ALSO

L<CHI|CHI>

=head1 AUTHOR

Jonathan Swartz <swartz@pobox.com>

=head1 COPYRIGHT AND LICENSE

This software is copyright (c) 2021 by Jonathan Swartz.

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

=cut