—

              
HideShow 221 lines of Pod
=pod
=for vim
   vim: tw=72 ts=3 sts=3 sw=3 et ai :
=encoding utf8
=head1 NAME
Data::Annotation::Expression
=head1 SYNOPSIS
   use Data::Annotation::Expression qw< evaluator_factory >;
   my $evaluator = evaluator_factory(\%definition, \%parse_context);
   my $retval = $evaluator->($data);
=head1 DESCRIPTION
This module exports L</evaluator_factory>, a factory function that
produces sub references based on an expression definition and additional
context.  These sub references can later be used to evaluate the
expression based on additional data that is provided as input.
=head2 Example
To make an example, suppose that we have the following I<target>
function (from a behavioural point of view, at least):
   my $expression = sub ($data) {
      return ($data->{foo} eq 'bar')
         && ($data->{baz}  ne $data->{galook});
   }
It's a boolean expression that makes use of some parts taken from
C<$data> (keys C<foo>, C<baz>, and C<baz>) as well as a constant (string
C<bar>), also leveraging boolean operators like C<&&> and comparison
operators like C<eq> and C<ne>.
The first step is to represent the expression as data, e.g. in
YAML it would be:
   and:
      - eq: [ '.foo', '=bar' ]
      - ne: [ '.baz', '.galook' ]
In this module's conventions, strings starting with a dot represent
access to the input data (like C<< .foo >>), while strings starting with
an equal sign represent verbatim data (like C<< =baz >>). When we have
this expression parsed as a Perl data structure, we can generate our
equivalent function:
   my $equivalent_expression = evaluator_factory($definition);
=head2 Definition format(s)
This section will use YAML format to represent data structure, although
this module does only accept Perl data structures (i.e. parsing of YAML
or any other representation format is supposed to be performed
elsewhere).
A definition is a hierarchical data structure composed of I<nodes>, that
are hash references, with some specific fields that indicate what the
node does. Some of these nodes represent functions/operators that take
parameters, in the form of array references.
In its most verbose form, the representation is very I<explicit>, like
this that represents C<< $input->{foo} eq 'bar' >>:
   type: sub
   name: eq
   args:
      - type: context
        path: run.foo
      - type: data
        value: bar
There are three node C<type>s:
=over
=item *
C<sub>: indicate a sub or an operator. This will be looked around based
on its C<name> and invoked with the provided C<args>, after they have
been evaluated. It is also possible to set a C<package> where the
C<name> function will be searched, otherwise the L</Parse context> will
be used.
=item *
C<data>: indicate verbatim data, provided as C<value>. This can itself
be some complicated data structure, but it will be used as-is.
=item *
C<context>: indicate data that has to be taken from the available
context, using a I<path> that is evaluated using
L<Data::Annotation::Traverse> functions. The top-most key in the path
can be C<run>, meaning whatever is passed to the evaluation function at
runtime, as well as C<parse>, meaning the L</Parse context>, as well as
C<definition>, which means the expression definition itself. Most of the
times you will probably want to use C<run>, possibly C<parse> (to keep
some less-dynamic data like configurations), almost never C<definition>
(it's there mostly for debugging reasons).
=back
The fully explicit form above is the I<normal> form. As it is quite
verbose, it's possible to use shortcuts as we already saw in the
example. First of all, plain strings can serve as expressions, like
this:
   &sub_name     -->    { type: sub, name: sub_name, args: [] }
   =foo_bar      -->    { type: data, value: foo_bar }
   run.place     -->    { type: context, path: run.place }
   .place        -->    { type: context, path: run.place }
Then, we can have hash-references that contain a single key/value pair
(like in the first examples). Keys C<data>, C<sub>, and C<context> do
I<the right thing> much like the string versions above; other keys
always lead to C<sub> nodes, where it's possible to also specify the
C<package> by setting the value to a I<fully qualified> Perl function
(i.e. including the package with the usual C<::> notation).
=head2 Parse context
When calling L</evalutor_function> it's possible to pass an optional
second parameter with a hash reference of options. These are used to set
the behaviour of parsing, as well as some runtime behaviour (e.g. where
to find functions or retrieving values from the "context").
The normalization process described in the previous section can be
completely overridden by passing option C<definition-normalizer>, which
is a sub reference with the following signature:
   sub normalizer ($parse_ctx, $definition)  --> $normalized_definition
You can find the default one in the L<Data::Annotation::Expression> code
(function C<default_definition_normalizer>), which does what explained
before.
The parse context can be used at runtime to retrieve values using a
C<context> node type. In this case, the path must start with string
C<parse>.
The parse context is also used in C<sub> node types. In particular, key
C<locator-relative-prefixes> is (if present) an array reference holding
a list of package prefixes that will be tried to find each function. As
an example, suppose that it is set like this:
   locator-relative-prefixes:
      - Foo::Bar
      - Baz::Galook
The following applies:
   { type: sub, name: blorg }
      candidates: package Foo::Bar           function blorg
                  package Baz::Galook        function blorg
   { type: sub, name: blorg, package: Barf }
      candidates: package Foo::Bar::Barf     function blorg
                  package Baz::Galook::Barf  function blorg
   { type: sub, name: blorg, package: '/Barf' }
      candidates: package Barf               function blorg
=head2 Builtin functions
This module comes with a set of wrappers around Perl most common
operators which are used as I<default> built-ins available in
expressions. These are contained in
L<Data::Annotation::Expression::Builtin> and this module is injected by
default inside L</Parse context> option C<locator-relative-prefixes>.
It is possible to avoid the injection of these built-ins by passing a
true value to L</Parse context> option C<no-builtin> when calling the
factory function.
=head1 INTERFACE
=head2 B<< evaluator_factory >>
   my $sub1 = evaluator_factory(\%definition); # OR
   my $sub2 = evaluator_factory(\%definition, \%parse_context);
Generate an evaluator function based on a C<%definition> and optional
C<%parse_context>. See L</DESCRIPTION> for everything.
=begin private
=over
=item default_definition_normalizer
=item generate_function
=item generate_function_context
=item generate_function_data
=item generate_function_sub
=item require_module
=item resolve_function
=back
=end private
=head1 ANYTHING ELSE (INCLUDING AUTHOR, COPYRIGHT AND LICENSE)
See documentation for Data::Annotation.
=cut