Scroll to navigation

SGML::Parser::OpenSP(3) User Contributed Perl Documentation SGML::Parser::OpenSP(3)


SGML::Parser::OpenSP - Parse SGML documents using OpenSP


  use SGML::Parser::OpenSP;
  my $p = SGML::Parser::OpenSP->new;
  my $h = ExampleHandler->new;
  $p->warnings(qw(xml valid));


This module provides an interface to the OpenSP SGML parser. OpenSP and this module are event based. As the parser recognizes parts of the document (say the start or end of an element), then any handlers registered for that type of an event are called with suitable parameters.


Returns a new SGML::Parser::OpenSP object. Takes no arguments.
Parses the file passed as an argument. Note that this must be a filename and not a filehandle. See "PROCESSING FILES" below for details.
Parses the data passed as an argument. See "PROCESSING FILES" below for details.
Halts processing before parsing the entire document. Takes no arguments.
Splits OpenSP's error messages into their component parts. See "POST-PROCESSING ERROR MESSAGES" below for details.
See "POSITIONING INFORMATION" below for details.



Report events to the blessed reference $handler.


Describe open entities in error messages. Error messages always include the position of the most recently opened external entity. The default is false.
Show the generic identifiers of open elements in error messages. The default is false.
Show message numbers in error messages.


Generate "comment_decl" events. The default is false.
Generate marked section events ("marked_section_start", "marked_section_end", "ignored_chars"). The default is false.
Generate "general_entity" events. The default is false.


"parse" arguments specify catalog files rather than the document entity. The document entity is specified by the first DOCUMENT entry in the catalog files. The default is false.
Restrict file reading to the specified directories (see the "search_dirs" method and the "SGML_SEARCH_PATH" environment variable). You should turn this option on and configure the search paths accordingly if you intend to process untrusted resources. The default is false.
Map public identifiers and entity names to system identifiers using the specified catalog entry files. Multiple catalogs are allowed. If there is a catalog entry file called "catalog" in the same place as the document entity, it will be searched for immediately after those specified.
Search the specified directories for files specified in system identifiers. Multiple values options are allowed. See the description of the osfile storage manager in the OpenSP documentation for more information about file searching.
Instruct "parse_string" to pass the input data down to the guts of OpenSP using the "OSFD" storage manager (if true) or the "OSFILE" storage manager (if false). This amounts to the difference between passing a file descriptor and a (temporary) file name.

The default is true except on platforms, such as Win32, which are known to not support passing file descriptors around in this manner. On platforms which support it you can call this method with a false parameter to force use of temporary file names instead.

In general, this will do the right thing on its own so it's best to consider this an internal method. If your platform is such that you have to force use of the OSFILE storage manager, please report it as a bug and include the values of $^O, $Config{archname}, and a description of the platform (e.g. "Windows Vista Service Pack 42").


For each name in @include_params pretend that

  <!ENTITY % name "INCLUDE">

occurs at the start of the document type declaration subset in the SGML document entity. Since repeated definitions of an entity are ignored, this definition will take precedence over any other definitions of this entity in the document type declaration. Multiple names are allowed. If the SGML declaration replaces the reserved name INCLUDE then the new reserved name will be the replacement text of the entity. Typically the document type declaration will contain

  <!ENTITY % name "IGNORE">

and will use %name; in the status keyword specification of a marked section declaration. In this case the effect of the option will be to cause the marked section not to be ignored.



Additional warnings can be enabled using


The following values can be used to enable warnings:

Warn about constructs that are not allowed by XML.
Warn about mixed content models that do not allow #pcdata anywhere.
Warn about various dubious constructions in the SGML declaration.
Warn about various recommendations made in ISO 8879 that the document does not comply with. (Recommendations are expressed with ``should'', as distinct from requirements which are usually expressed with ``shall''.)
Warn about defaulted references.
Warn about duplicate entity declarations.
Warn about undefined elements: elements used in the DTD but not defined.
Warn about unclosed start and end-tags.
Warn about empty start and end-tags.
Warn about net-enabling start-tags and null end-tags.
Warn about minimized start and end-tags. Equivalent to combination of unclosed, empty and net warnings.
Warn about unused short reference maps: maps that are declared with a short reference mapping declaration but never used in a short reference use declaration in the DTD.
Warn about parameter entities that are defined but not used in a DTD. Unused internal parameter entities whose text is "INCLUDE" or "IGNORE" won't get the warning.
Warn about notations for which no system identifier could be generated.
Warn about conditions that should usually be avoided (in the opinion of the author). Equivalent to: "mixed", "should", "default", "undefined", "sgmldecl", "unused-map", "unused-param", "empty" and "unclosed".


A warning can be disabled by using its name prefixed with "no-". Thus calling warnings(qw(all no-duplicate)) will enable all warnings except those about duplicate entity declarations.

The following values for warnings() disable errors:

Do not give an error for an ID reference value which no element has as its ID. The effect will be as if each attribute declared as an ID reference value had been declared as a name.
Do not give an error when a character that is not a significant character in the reference concrete syntax occurs in a literal in the SGML declaration. This may be useful in conjunction with certain buggy test suites.
Do not require the document to be type-valid. This has the effect of changing the SGML declaration to specify "VALIDITY NOASSERT" and "IMPLYDEF ATTLIST YES ELEMENT YES". An option of "valid" has the effect of changing the SGML declaration to specify "VALIDITY TYPE" and "IMPLYDEF ATTLIST NO ELEMENT NO". If neither "valid" nor "no-valid" are specified, then the "VALIDITY" and "IMPLYDEF" specified in the SGML declaration will be used.


The following warnings are turned on for the "xml" warning described above:

Warn about inclusions in element type declarations.
Warn about exclusions in element type declarations.
Warn about RCDATA declared content in element type declarations.
Warn about CDATA declared content in element type declarations.
Warn about comments in parameter separators.
Warn about name groups in attribute declarations.
Warn about name groups in element type declarations.
Warn about PI entities.
Warn about internal SDATA entities.
Warn about internal CDATA entities.
Warn about external SDATA entities.
Warn about external CDATA entities.
Warn about bracketed text entities.
Warn about attribute definition list declarations for notations.
Warn about external identifiers without system identifiers.
Warn about content reference attributes.
Warn about current attributes.
Warn about attributes with a declared value of NUTOKEN or NUTOKENS.
Warn about attributes with a declared value of NUMBER or NUMBERS.
Warn about attributes with a declared value of NAME or NAMES.
Warn about named character references.
Warn about ommitted refc delimiters.
Warn about TEMP marked sections.
Warn about RCDATA marked sections.
Warn about INCLUDE marked sections in the document instance.
Warn about IGNORE marked sections in the document instance.
Warn about AND connectors in model groups.
Warn about ranked elements.
Warn about empty comment declarations.
Warn about attribute values which are not literals.
Warn about ommitted attribute names in start tags.
Warn about spaces before the MDC in comment declarations.
Warn about comment declarations containing multiple comments.
Warn about marked sections without a status keyword.
Warn about marked sections with multiple status keywords.
Warn about parameter entities in the document instance.
Warn about minimization parameters in element type declarations.
Warn about cases of mixed content which are not allowed in XML.
Warn about name groups with a connector different from OR.
Warn about processing instructions which don't start with a name.
Warn about spaces between DSO and status keyword in marked sections.
Warn about references to external data entities in the content.
Warn about references to external data entities in attribute values.
Warn about occurances of `<' and `&' as data.
Warn about an explicit SGML declaration.
Warn about marked sections in the internal subset.
Warn about a default entity declaration.
Warn about numeric character references to non-SGML characters.
Warn about parameter entity references in parameter separators in the internal subset.
Warn about parameter entity references in token separators in the internal subset.
Warn about parameter entity references in parameter literals in the internal subset.


In order to start processing of a document and recieve events, the "parse" method must be called. It takes one argument specifying the path to a file (not a file handle). You must set an event handler using the "handler" method prior to using this method. The return value of "parse" is currently undefined.


In order to receive data from the parser you need to write an event handler. For example,

  package ExampleHandler;
  sub new { bless {}, shift }
  sub start_element
      my ($self, $elem) = @_;
      printf "  * %s\n", $elem->{Name};

This handler would print all the element names as they are found in the document, for a typical XHTML document this might result in something like

  * html
  * head
  * title
  * body
  * p
  * ...

The events closely match those in the generic interface to OpenSP, see <> for more information.

The event names have been changed to lowercase and underscores to separate words and properties are capitalized. Arrays are represented as Perl array references. "Position" information is not passed to the handler but made available through the "get_location" method which can be called from event handlers. Some redundant information has also been stripped and the generic identifier of an element is stored in the "Name" hash entry.

For example, for an EndElementEvent the "end_element" handler gets called with a hash reference

    Name => 'gi'

The following events are defined:

  * appinfo
  * processing_instruction
  * start_element
  * end_element
  * data
  * sdata
  * external_data_entity_ref
  * subdoc_entity_ref
  * start_dtd
  * end_dtd
  * end_prolog
  * general_entity       # set $p->output_general_entities(1)
  * comment_decl         # set $p->output_comment_decls(1)
  * marked_section_start # set $p->output_marked_sections(1)
  * marked_section_end   # set $p->output_marked_sections(1)
  * ignored_chars        # set $p->output_marked_sections(1)
  * error
  * open_entity_change

If the documentation of the generic interface to OpenSP states that certain data is not valid, it will not be available through this interface (i.e., the respective key does not exist in the hash ref).


Event handlers can call the "get_location" method on the parser object to retrieve positioning information, the get_location method will return a hash reference with the following properties:

  LineNumber   => ..., # line number
  ColumnNumber => ..., # column number
  ByteOffset   => ..., # number of preceding bytes
  EntityOffset => ..., # number of preceding bit combinations
  EntityName   => ..., # name of the external entity
  FileName     => ..., # name of the file

These can be "undef" or an empty string.


OpenSP returns error messages in form of a string rather than individual components of the message like line numbers or message text. The "split_message" method on the parser object can be used to post-process these error message strings as reliable as possible. It can be used e.g. from an error event handler if the parser object is accessible like

  sub error
    my $self = shift;
    my $erro = shift;
    my $mess = $self->{parser}->split_message($erro);

See the documentation of "split_message" in the SGML::Parser::OpenSP::Tools documentation.


All strings returned from event handlers and helper routines are UTF-8 encoded with the UTF-8 flag turned on, helper functions like "split_message" expect (but don't check) that string arguments are UTF-8 encoded and have the UTF-8 flag turned on. Behavior of helper functions is undefined when you pass unexpected input and should be avoided.

"parse" has limited support for binary input, but the binary input must be compatible with OpenSP's generic interface requirements and you must specify the encoding through means available to OpenSP to enable it to properly decode the binary input. Any encoding meta data about such binary input specific to Perl (such as encoding disciplines for file handles when you pass a file descriptor) will be ignored. For more specific information refer to the OpenSP manual.


OpenSP supports a number of environment variables to control specific processing aspects such as "SGML_SEARCH_PATH" or "SP_CHARSET_FIXED". Portable applications need to ensure that these are set prior to loading the OpenSP library into memory which happens when the XS code is loaded. This means you need to wrap the code into a "BEGIN" block:

  use SGML::Parser::OpenSP;
  # ...

Otherwise changes to the environment might not propagate to OpenSP. This applies specifically to Win32 systems.

See <>.
The "User-Agent" header for HTTP requests.
The "Accept" header for HTTP requests.
Enable run time selection of message format, Value is one of "XML", "NONE", "TRADITIONAL". Whether this will have an effect depends on a compile time setting which might not be enabled in your OpenSP build. This module assumes that no such support was compiled in.
See <>.
See <>.

Note that you can use the "search_dirs" method instead of using "SGML_SEARCH_PATH" and the "catalogs" method instead of using "SGML_CATALOG_FILES" and attributes on storage object specifications for "SP_BCTF" and "SP_ENCODING" respectively. For example, if "SP_CHARSET_FIXED" is set to 1 you can use

  $p->parse("<OSFILE encoding='UTF-8'>example.xhtml");

to process "example.xhtml" using the "UTF-8" character encoding.


OpenSP must be compiled with "SP_MULTI_BYTE" defined and with "SP_WIDE_SYSTEM" undefined, this module will otherwise break at runtime or not compile.


Please report bugs in this module via <>

Please report bugs in OpenSP via <>

Please send comments and questions to the spo-devel mailing list, see <> for details.



  Terje Bless <> wrote version 0.01.
  Bjoern Hoehrmann <> wrote version 0.02+.


  Copyright (c) 2006-2008 Bjoern Hoehrmann <>.
  This module is licensed under the same terms as Perl itself.
2023-08-09 perl v5.38.2