# Copyright (C) 2004-2008, Parrot Foundation.
# $Id: configuration.pod 37201 2009-03-08 12:07:48Z fperrad $

=pod

=head1 NAME

docs/configuration.pod - Parrot Configuration System

=head1 DESCRIPTION

Parrot configuration is broken up into I<steps>. Each step contains several
related I<prompts>, I<probes>, or I<generations>. Steps should be mostly of a
single type, though some overlap is allowed (for example, allowing a probe to
ask the user what to do in an exceptional situation).

The directory F<config> contains subdirectories for each type of step.  Each
step should consist of I<exactly one> F<.pm> file and any number of supporting
F<.c>, F<.in>, etc. files. Any supporting files should be in a folder whose
name is the same as the basename of the step's F<.pm> file; for example, if
F<foo.pm> uses F<bar_c.in>, F<bar_c.in> should be in a directory called F<foo>;
the full path might be F<config/auto/foo/bar_c.in>.

Generally, when adding a new component to the configuration process you should
add a new step unless that component I<clearly> belongs in a current step. For
example, if you added a new user-configurable type called C<FOOVAL>, you would
add the code used to determine its size in F<config/auto/sizes.pm>.  However,
if you were determining what dynaloading capabilities are available, you would
create a new step.

=head2 Initialization Steps

I<Initialization steps> are run before any other steps. They do tasks
such as preparing the configuration system's data structures and
checking the F<MANIFEST>. New initialization steps will only be added
when the configuration system is getting significant new capabilities.
They're kept in the directory F<config/init>.

=head2 Prompts

Prompts ask the user for some information. These should be used sparingly. A
step containing prompts is an I<interactive step>.  Interactive steps should be
in the F<config/inter> folder.

Interactive steps often include simple probes to determine good guesses of what
the user will answer. See L</Prompt or Probe?> for more information.

Note that, by default, these prompts are turned off. To enable them run
F<Configure.pl> with the C<--ask> option.

=head2 Probes

Probes are automated tests of some feature of the computer. These should be
used wherever a value will not need to be modified often by the user.  A step
containing probes is an I<automatic step>. Automatic steps should be in the
F<config/auto> folder.

=head2 Generations

Generations create files needed after configuration has completed, such as
Makefiles and configuration headers. A step containing generations is a
I<generation step>. Generation steps should be in the F<config/gen> folder.

Templates for files to be generated usually have the extension F<.in>.  There
are variable substitutes and funny macros like C<#IF> and C<#UNLESS>
(conditional lines).

=head2 Prompt or Probe?

It can sometimes be hard to decide whether a given step should be an automatic
or an interactive step. The guiding question is I<Would a user ever want to
change this?>, or conversely, I<Is this something that can be completely
determined without user intervention?>  A step figuring out what the
compiler's command is would probably be an interactive step; conversely, a
step figuring out if that command is connected to a specific compiler (like
F<gcc>) would be an automatic step.

=head2 Configuring Configuration

The configuration system gets its own configuration data from, and is invoked
via, the F<Configure.pl> script.  The system is invoked by instantiating a
L<Parrot::Configure> object, registering one or more steps with that object,
and then calling C<Parrot::Configure::runsteps()>.

=head2 Adding New Steps

New steps should be added in one of the four folders mentioned above.

All steps are really classes; each exists in a unique namespace.  The
namespace used depends on the step's relative path in the source tree sans the
F<config> prefix.  For example, the step F<config/init/defaults.pm> uses the
C<init::defaults> namespace.

Each step inherits its constructor and some other methods from
F<lib/Parrot/Configure/Step.pm>.  Each step needs to define only
two methods:  C<_init()> and C<runstep()>.

The C<_init()> method should follow the following example, defining three
elements within an internal hash:

    sub _init {
        my $self = shift;
        my %data;
        $data{description} = q{This is the step description};
        $data{result}      = q{};
        return \%data;
    }

=over 4

=item C<description>

Of these three elements, C<description> is the most important as it is
used extensively within C<lib/Parrot/Configure.pm>.

Returns a short descriptive message that should be printed before the step
executes.

Some example descriptions:

=over 4

=item F<config/auto/cgoto.pm>

    Does your compiler support computed goto...

=item F<gen/config_h.pm>

    Generate C headers...

=back

Note that on non-interactive steps, the text I<done> will be printed after the
description when the step finishes executing; for example, the user will see:

    Does your compiler support computed goto..............done.

=item C<result>

The C<result> is initialized to an empty string mainly to quiet
'uninitialized variable' warnings.  Most configuration steps override
the C<result> value inside their C<runstep()> method.

=item C<runstep()>

This method is called to actually execute the step.  The return
value should be C<1> if the step accomplishes what it is intended to do.
Otherwise, the step should simply C<return>, I<i.e.>, return an
undefined value.

=back

The configuration system won't execute your step by default unless it's
specifically told to. To do this, edit F<lib/Parrot/Configure/Step/List.pm>.
Steps are run in the order in which that are registered with the
L<Parrot::Configure> object.

Various utility functions for configuration steps are provided by the
L<Parrot::Configure::Utils> module.

A template for a new step might look like this:

    package auto::newstep;

    use strict;
    use warnings;
    our qw($description $result);

    use base qw(Parrot::Configure::Step);

    use Parrot::Configure::Utils;

    sub _init {
        my $self = shift;
        my %data;
        $data{description} = q{This is the step description};
        $data{result}      = q{};
        return \%data;
    }

    sub runstep {
        my ($self, $conf) = @_
        ...
        if ($success) {
            $self->set_result('yes');
            return 1;
        } else {
            $self->set_result('no');
            return;
        }
    }

The step's C<runstep()> method should return C<1> upon success and do a
bare return on failure (thereby returning an undefined value).

=head2 Command-line Arguments

Command-line arguments look like C</--[-\w]+(=.*)?/>; the equals sign separates
the name and the value. If the value is omitted, it's assumed to be C<1>. The
options C<--help> and C<--version> are built in to Configure; any others are
defined by steps.

Command-line arguments are now processed by C<process_options()>, a subroutine
exported by L<Parrot::Configure::Options>.  If you add a new option, don't
forget to add it to this documentation and to appropriate locations.  Most
options should be added to C<@shared_valid_options> in
F<lib/Parrot/Configure/Options/Conf/Shared.pm>.

Arguments passed to F<Configure.pl> are held in a L<Parrot::Configure::Data>
object stored inside the L<Parrot::Configure> object.  The options data object
may be accessed via the C<Parrot::Configure::options()> method.

=head2 Configuration by File

As an alternative to typing a long string of options on the command-line,
Parrot can now be configured from a configuration file.  You put the options
in a configuration file, then call F<Configure.pl> as follows:

    perl Configure.pl --file=/path/to/config/file

That's it!  No other command-line arguments needed (or permitted)!  To learn
how to create such a configuration file, call:

    perldoc Configure.pl

=head2 Building Up Configuration Data

The second step is F<config/init/defaults.pm>, which sets up some defaults in
a L<Parrot::Configure::Data> object contained by the main L<Parrot::Configure>
object.  It can be accessed via the C<Parrot::Configure::data()> method.  You
get and set configuration system's data by interacting with this object.  Some
of its methods are summarized below.

=over 4

=item C<get(keys)>

Returns the values for the given keys.

=item C<set(key, value, [key, value, ...])>

Sets the given keys to the given values.

=item C<add(delim, key, value, [key, value, ...])>

Sets the given keys to the given values or appends values delimited by I<delim>
to existing keys.

=item C<keys()>

Returns a list of all keys.

=item C<dump()>

Returns a string that can be C<eval>ed by Perl to create a hash representing
the configuration system's data.

See the L<Parrot::Configure::Data> documentation for further details.

=back

=head2 Special Configuration Items

Some configuration items, by convention, have a special meaning (mainly
prefixes) and are handled with some magic internally.

=over 4

=item C<i_(\w+)> include files

defines or undefs PARROT_HAS_HEADER_XXX in F<include/parrot/has_header.h>

=item C<HAS_(\w+)> features

defines PARROT_HAS_XXX in F<include/parrot/has_header.h>

=item C<TEMP_(\w+)> temporary settings

These settings are deleted before F<lib/Parrot/Config.pm> is written. These
entries are only used e.g. for Makefile creation.

=back

=head2 Accessing Configuration Information via Perl 5 Variables

Parrot configuration is currently jump-started by extracting
considerable information from variables associated with the instance of
Perl 5 which the user is using to run F<Configure.pl>.  These variables
are largely looked up in the C<%Config> found in the Perl 5
F<Config.pm>, but may also be sought in Perl 5 special variables such as
C<$^O>.  All such lookups should be done in configuration step
C<init::defaults> and B<only> in that step.  Special accessors are
available for working with such variables; see
F<config/init/defaults.pm> and F<lib/Parrot/Configure/Data.pm>.

=head1 HISTORY

The Parrot configuration system was created by Brent Dax and has been heavily
mangled by Joshua Hoblitt.  James Keenan refactored the configuration system.

=head1 SEE ALSO

L<Parrot::Configure>, L<Parrot::Configure::Data>,
L<Parrot::Configure::Utils>, L<Parrot::Configure::Step>

=cut