++ed by:

1 PAUSE user

Barrie Slaymaker


XML::Filter::Merger - Assemble multiple SAX streams in to one document


    ## See XML::SAX::Manifold and XML::SAX::Pipeline for easy ways
    ## to use this processor.  XML::SAX::Manifold uses this
    ## processor to implement multipass document processing, for
    ## instance.

    my $w = XML::SAX::Writer->new(           Output => \*STDOUT );
    my $h = XML::Filter::Merger->new(           Handler => $w );
    my $p = XML::SAX::ParserFactory->parser( Handler => $h );

    $h->start_manifold_document( {} );
    $p->parse_file( $_ ) for @ARGV;
    $h->end_manifold_document( {} );


Combines several documents in to one "manifold" document. This is done by defining two non-SAX events--start_manifold_document and end_manifold_document--that are called before the first document to be combined and after the last one, respectively.

The first full document to be started after the start_manifold_document is the master document and is emitted as-is except that it will contain the contents of all of the other documents.

Unlike a normal SAX filter, however, documents may be inserted by issuing a start_document ... end_document event sequence inside the root element of the master document. The bodies of such documents are inserted inline in the master document without the root element or events before/after the root element. The root element may be inserted by calling set_include_all_roots with a true value. This is how the ByRecord SAX machine works, and is as though an XInclude directive had been placed in the master document at the point where the secondary document's events were received.

Additionally, any documents received after the master document's end_document and before the end_manifold_document are inserted just before the master document's root end_element. To accomplish this, the master document's root end_element and all remaining events are buffered and only forwarded when the end_manifold_document is received.


In case the above was a bit vague, here are the rules this filter lives by.

For the master document:

  • Events before the root end_element are forwarded as received. Because of the rules for secondary documents, any secondary documents sent to the filter in the midst of a master document will be inserted inline as their events are received.

  • All remaining events, from the root end_element are buffered until the end_manifold_document() received, and are then forwarded on.

For secondary documents:

  • All events before the root start_element are discarded. There is no way to recover these (though we can add an option for most non-DTD events, I believe).

  • The root start_element is discarded by default, or forwarded if set_include_all_roots( $v ) has been used to set a true value.

  • All events up to, but not including, the root end_element are forwarded as received.

  • The root end_element is discarded or forwarded if the matching start_element was.

  • All remaining events until and including the end_document are forwarded and processing.

  • Secondary documents may contain other secondary documents.

  • Secondary documents need not be well formed. The must, however, be well balanced.

This requires very little buffering and is "most natural" with the limitations:

  • All of each secondary document's events must all be received between two consecutive events of it's master document. This is because most master document events are not buffered and this filter cannot tell from which upstream source a document came.

  • If the master document should happen to have some egregiously large amount of whitespace, commentary, or illegal events after the root element, buffer memory could be huge. This should be exceedingly rare, even non-existent in the real world.

  • If any documents are not well balanced, the result won't be.


The events before and after a secondary document's root element events are discarded. It is conceivable that characters, PIs and commentary outside the root element might need to be kept. This may be added as an option.

The DocumentLocators are not properly managed: they should be saved and restored around each each secondary document.

If either of these bite you, contact me.


    my $d = XML::Filter::Merger->new( \%options );

This must be called before the first document's start_document arrives.

It is passed an empty ({}) data structure, which is passed on to the handler's start_document.


This must be called after the last document's end_document is called. It is passed an empty ({}) data structure which is passed on to the next processor's end_document() call. This call also causes the end_element() for the root element to be passed on.

    $h->set_include_all_roots( 1 );

Setting this option causes the merger to include all root element nodes, not just the first document's. This means that later documents are treated as subdocuments of the output document, rather than as envelopes carrying subdocuments.

Given two documents received are:

 Doc1:   <root1><foo></root1>

 Doc1:   <root2><bar></root2>

 Doc3:   <root3><baz></root3>

then with this option cleared (the default), the result looks like:


. This is useful when processing document oriented XML and each upstream filter channel gets a complete copy of the document. This is the case with the machine XML::SAX::Manifold and the splitting filter XML::Filter::Distributor.

With this option set, the result looks like:


This is useful when processing record oriented XML, where the first document only contains the preamble and postamble for the records and not all of the records. This is the case with the machine XML::SAX::ByRecord and the splitting filter XML::Filter::DocSplitter.

The two splitter filters mentioned set this feature appropriately.


Does not yet buffer all events after the first document's root end_element event.


    Barrie Slaymaker <barries@slaysys.com>


    Copyright 2002, Barrie Slaymaker, All Rights Reserved.

You may use this module under the terms of the Artistic, GNU Public, or BSD licenses, you choice.