XML::SAX::Machine - Manage a collection of SAX processors

    ## Note: See XML::SAX::Pipeline and XML::SAX::Machines first,
    ## this is the gory, detailed interface.

    use My::SAX::Machines qw( Machine );
    use My::SAX::Filter2;
    use My::SAX::Filter3;

    my $filter3 = My::SAX::Filter3->new;

    ## A simple pipeline.  My::SAX::Filter1 will be autoloaded.
    my $m = Machine(
        #
        # Name   => Class/object            => handler(s)
        #
        [ Intake => "My::SAX::Filter1"      => "B"        ],
        [ B      => My::SAX::Filter2->new() => "C"        ],
        [ C      => $filter3                => "D"        ],
        [ D      => \*STDOUT                              ],
    );

    ## A parser will be created unless My::SAX::Filter1 can parse_file
    $m->parse_file( "foo.revml" );

    my $m = Machine(
        [ Intake   => "My::SAX::Filter1"  => qw( Tee     ) ],
        [ Tee      => "XML::Filter::SAXT" => qw( Foo Bar ) ],
        [ Foo      => "My::SAX::Filter2"  => qw( Out1    ) ],
        [ Out1     => \$log                                ],
        [ Bar      => "My::SAX::Filter3"  => qw( Exhaust ) ],
    );

WARNING: This API is alpha!!! It will be changing.

A generic SAX machine (an instance of XML::SAX::Machine) is a container of SAX processors (referred to as parts) connected in arbitrary ways.

Each parameter to CWMachine() (or CWXML::SAX::Machine-new()>) represents one top level part of the machine. Each part has a name, a processor, and one or more handlers (usually specified by name, as shown in the SYNOPSIS).

Since SAX machines may be passed in as single top level parts, you can also create nested, complex machines ($filter3 in the SYNOPSIS could be a Pipeline, for example).

A SAX machines can act as a normal SAX processors by connecting them to other SAX processors:

    my $w = My::Writer->new();
    my $m = Machine( ...., { Handler => $w } );
    my $g = My::Parser->new( Handler => $w );

Although it's not required, each part in a machine can be named. This is useful for retrieving and manipulating the parts (see part, for instance), and for debugging, since debugging output (see trace_parts and trace_all_parts) includes the names.

Part names must be valid Perl subroutine names, beginning with an uppercase character. This is to allow convenience part accessors methods like

    $c = $m->NameOfAFilter;

to work without ever colliding with the name of a method (all method names are completely lower case). Only filters named like this can be accessed using the magical accessor functions.

The names c<Intake> and CWExhaust are reserved. CWIntake refers to the first part in the processing chain. This is not necessarily the first part in the constructor list, just the first part to receive external events.

CWExhaust refers to the output of the machine; no part may be named CWExhaust, and any parts with a handler named CWExhaust will deliver their output to the machine's handler. Normally, only one part should deliver it's output to the Exhaust port.

Calling CW$m->set_handler() alters the Exhaust port, assuming any processors pointing to the CWExhaust provide a CWset_handler() method like XML::SAX::Base's.

CWIntake and CWExhaust are usually assigned automatically by single-purpose machines like XML::SAX::Pipeline and XML::SAX::Manifold.

The XML::SAX::Machine class is very agnostic about what SAX processors it supports; about the only constraint is that it must be a blessed reference (of any type) that does not happen to be a Perl IO::Handle (which are assumed to be input or output filehandles).

The major constraint placed on SAX processors is that they must provide either a CWset_handler or CWset_handlers method (depending on how many handlers a processor can feed) to allow the SAX::Machine to disconnect and reconnect them. Luckily, this is true of almost any processor derived from XML::SAX::Base. Unfortunately, many SAX older (SAX1) processors do not meet this requirement; they assume that SAX processors will only ever be connected together using their constructors.

SAX machines allow you to connect the parts however you like; each part is given a name and a list of named handlers to feed. The number of handlers a part is allowed depends on the part; most filters only allow once downstream handler, but filters like XML::Filter::SAXT and XML::Filter::Distributor are meant to feed multiple handlers.

Parts may not be connected in loops (cycles in graph theory terms). The machines specified by:

    [ A => "Foo" => "A" ],  ## Illegal!

and

    [ A => "Foo" => "B" ],  ## Illegal!
    [ B => "Foo" => "A" ],

. Configuring a machine this way would cause events to flow in an infinite loop, and/or cause the first processor in the cycle to start receiving events from the end of the cycle before the input document was complete. Besides, it's not a very useful topology :).

SAX machines detect loops at construction time.

These methods are meant to be used by users of SAX machines.

new()

    my $m = $self->new( @machine_spec, \%options );

Creates CW$self using CW%options, and compiles the machine spec. This is the longhand form of CWMachines( ... ).

find_part

Gets a part contained by this machine by name, number or object reference:

    $c = $m->find_part( $name );
    $c = $m->find_part( $number );
    $c = $m->find_part( $obj );    ## useful only to see if $obj is in $m

If a machine contains other machines, parts of the contained machines may be accessed by name using unix directory syntax:

    $c = $m->find_part( "/Intake/Foo/Bar" );

(all paths must be absolute). Parts may also be accessed by number using array indexing:

    $c = $m->find_part(0);  ## Returns first part or undef if none
    $c = $m->find_part(-1); ## Returns last part or undef if none
    $c = $m->find_part( "Foo/0/1/-1" );

There is no way to guarantee that a part's position number means anything, since parts can be reconnected after their position numbers are assigned, so using a part name is recommended. Throws an exception if the part is not found, so doing things like

   $m->find_part( "Foo" )->bar()

garner informative messages when Foo is not found. If you want to test a result code, do something like

    my $p = eval { $m->find_part };
    unless ( $p ) {
        ...handle lookup failure...
    }

parts

    for ( $m->parts ) { ... }

Gets an arbitrarily ordered list of top level parts in this machine. This is all of the parts directly contained by this machine and none of the parts that may be inside them. So if a machine contains an XML::SAX::Pipeline as one of it's parts, the pipeline will be returned but not the parts inside the pipeline.

all_parts

    for ( $m->all_parts ) { ... }

Gets all parts in this machine, not just top level ones. This includes any machines contained by this machine and their parts.

set_handler

    $m->set_handler( $handler );
    $m->set_handler( DTDHandler => $handler );

Sets the machine's handler and sets the handlers for all parts that have CWExhaust specified as their handlers. Requires that any such parts provide a CWset_handler or (if the part has multiple handlers) a CWset_handlers method. NOTE: handler types other than Handler are only supported if they are supported by whatever parts point at the CWExhaust. If the handler type is CWHandler, then the appropriate method is called as:

    $part->set_handler( $handler );
    $part->set_handlers( $handler0, $handler1, ... );

If the type is some other handler type, these are called as:

    $part->set_handler( $type => $handler );
    $part->set_handlers( { $type0 => $handler0 }, ... );

trace_parts

    $m->trace_parts;          ## trace all top-level parts
    $m->trace_parts( @ids );  ## trace the indicated parts

Uses Devel::TraceSAX to enable tracing of all events received by the parts of this machine. Does not enable tracing of parts contained in machines in this machine; for that, see trace_all_parts.

trace_all_parts

    $m->trace_all_parts;      ## trace all parts

Uses Devel::TraceSAX to trace all events received by the parts of this machine.

untracify_parts

    $m->untracify_parts( @ids );

Converts the indicated parts to SAX processors with tracing enabled. This may not work with processors that use AUTOLOAD.

XML::SAX::Machine provides all SAX1 and SAX2 events and delgates them to the processor indicated by CW$m->find_part( Intake ). This adds some overhead, so if you are concerned about overhead, you might want to direct SAX events directly to the Intake instead of to the machine.

It also provides parse...() routines so it can whip up a parser if need be. This means: parse(), parse_uri(), parse_string(), and parse_file() (see XML::SAX::EventMethodMaker for details). There is no way to pass methods directly to the parser unless you know that the Intake is a parser and call it directly. This is not so important for parsing, because the overhead it takes to delegate is minor compared to the effort needed to parse an XML document.

These methods are meant to be used/overridden by subclasses.

_compile_specs

    my @comp = $self->_compile_specs( @_ );

Runs through a list of module names, output specifiers, etc., and builds the machine.

    $scalar     --> "$scalar"->new
    $ARRAY_ref  --> pipeline @$ARRAY_ref
    $SCALAR_ref --> XML::SAX::Writer->new( Output => $SCALAR_ref )
    $GLOB_ref   --> XML::SAX::Writer->new( Output => $GLOB_ref )

generate_description

    $m->generate_description( $h );
    $m->generate_description( Handler => $h );
    $m->generate_description( Pipeline ... );

Generates a series of SAX events to the handler of your choice. See XML::Handler::Machine2GraphViz on CPAN for a way of visualizing machine innards.

•

Separate initialization from construction time; there should be somthing like a CW$m->connect( ....machine_spec... ) that new() calls to allow you to delay parts speficication and reconfigure existing machines.

•

Allow an XML doc to be passed in as a machine spec.

    Barrie Slaymaker <barries@slaysys.com>

Artistic or GPL, any version.