ADC Home > Reference Library > Reference > Mac OS X > Mac OS X Man Pages

 

This document is a Mac OS X manual page. Manual pages are a command-line technology for providing documentation. You can view these manual pages locally using the man(1) command. These manual pages come from many different sources, and thus, have a variety of writing styles.

For more information about the manual page format, see the manual page for manpages(5).



Pod::Simple::Subclassing(3)          User Contributed Perl Documentation         Pod::Simple::Subclassing(3)



NAME
       Pod::Simple::Subclassing -- write a formatter as a Pod::Simple subclass

SYNOPSIS
         package Pod::SomeFormatter;
         use Pod::Simple;
         @ISA = qw(Pod::Simple);
         $VERSION = '1.01';
         use strict;

         sub _handle_element_start {
           my($parser, $element_name, $attr_hash_r) = @_;
           ...
         }

         sub _handle_element_end {
           my($parser, $element_name) = @_;
           ...
         }

         sub _handle_text {
           my($parser, $text) = @_;
           ...
         }
         1;

DESCRIPTION
       This document is about using Pod::Simple to write a Pod processor, generally a Pod formatter. If you
       just want to know about using an existing Pod formatter, instead see its documentation and see also
       the docs in Pod::Simple.

       The zeroeth step in writing a Pod formatter is to make sure that there isn't already a decent one in
       CPAN. See <http://search.cpan.org/, and run a search on the name of the format you want to render
       to. Also consider joining the Pod People list <http://lists.perl.org/showlist.cgi?name=pod-people
       and asking whether anyone has a formatter for that format -- maybe someone cobbled one together but
       just hasn't released it.

       The first step in writing a Pod processor is to read perlpodspec, which contains notes information on
       writing a Pod parser (which has been largely taken care of by Pod::Simple), but also a lot of
       requirements and recommendations for writing a formatter.

       The second step is to actually learn the format you're planning to format to -- or at least as much
       as you need to know to represent Pod, which probably isn't much.

       The third step is to pick which of Pod::Simple's interfaces you want to use -- the basic interface
       via Pod::Simple or Pod::Simple::Methody is event-based, sort of like HTML::Parser's interface, or
       sort of like XML::Parser's "Handlers" interface), but Pod::Simple::PullParser provides a token-stream
       interface, sort of like HTML::TokeParser's interface; Pod::Simple::SimpleTree provides a simple tree
       interface, rather like XML::Parser's "Tree" interface. Users familiar with XML-handling will find one
       of these styles relatively familiar; but if you would be even more at home with XML, there are
       classes that produce an XML representation of the Pod stream, notably Pod::Simple::XMLOutStream; you
       can feed the output of such a class to whatever XML parsing system you are most at home with.

       The last step is to write your code based on how the events (or tokens, or tree-nodes, or the XML, or
       however you're parsing) will map to constructs in the output format. Also sure to consider how to
       escape text nodes containing arbitrary text, and also what to do with text nodes that represent
       preformatted text (from verbatim sections).

Events
       TODO intro... mention that events are supplied for implicits, like for missing >'s

       In the following section, we use XML to represent the event structure associated with a particular
       construct.  That is, TODO

       "$parser->_handle_element_start( element_name, attr_hashref )"
       "$parser->_handle_element_end( element_name  )"
       "$parser->_handle_text(  text_string  )"

       TODO describe

       events with an element_name of Document
           Parsing a document produces this event structure:

             <Document start_line="543">
               ...all events...
             </Document>

           The value of the start_line attribute will be the line number of the first Pod directive in the
           document.

           If there is no Pod in the given document, then the event structure will be this:

             <Document contentless="1" start_line="543">
             </Document>

           In that case, the value of the start_line attribute will not be meaningful; under current
           implementations, it will probably be the line number of the last line in the file.

       events with an element_name of Para
           Parsing a plain (non-verbatim, non-directive, non-data) paragraph in a Pod document produces this
           event structure:

               <Para start_line="543">
                 ...all events in this paragraph...
               </Para>

           The value of the start_line attribute will be the line number of the start of the paragraph.

           For example, parsing this paragraph of Pod:

             The value of the I<start_line> attribute will be the
             line number of the start of the paragraph.

           produces this event structure:

               <Para start_line="129">
                 The value of the
                 <I>
                   start_line
                 </I>
                  attribute will be the line number of the first Pod directive
                 in the document.
               </Para>

       events with an element_name of B, C, F, or I.
           Parsing a B<...> formatting code (or of course any of its semantically identical syntactic
           variants B<< ... >>, or B<<<< ... >>>>, etc.)  produces this event structure:

                 <B>
                   ...stuff...
                 </B>

           Currently, there are no attributes conveyed.

           Parsing C, F, or I codes produce the same structure, with only a different element name.

           If your parser object has been set to accept other formatting codes, then they will be presented
           like these B/C/F/I codes -- i.e., without any attributes.

       events with an element_name of S
           Normally, parsing an S<...> sequence produces this event structure, just as if it were a B/C/F/I
           code:

                 <S>
                   ...stuff...
                 </S>

           However, Pod::Simple (and presumably all derived parsers) offers the "nbsp_for_S" option which,
           if enabled, will suppress all S events, and instead change all spaces in the content to non-breaking nonbreaking
           breaking spaces. This is intended for formatters that output to a format that has no code that
           means the same as S<...>, but which has a code/character that means non-breaking space.

       events with an element_name of X
           Normally, parsing an X<...> sequence produces this event structure, just as if it were a B/C/F/I
           code:

                 <X>
                   ...stuff...
                 </X>

           However, Pod::Simple (and presumably all derived parsers) offers the "nix_X_codes" option which,
           if enabled, will suppress all X events and ignore their content.  For formatters/processors that
           don't use X events, this is presumably quite useful.

       events with an element_name of L
           Because the L<...> is the most complex construct in the language, it should not surprise you that
           the events it generates are the most complex in the language. Most of complexity is hidden away
           in the attribute values, so for those of you writing a Pod formatter that produces a non-hypertextual nonhypertextual
           hypertextual format, you can just ignore the attributes and treat an L event structure like a
           formatting element that (presumably) doesn't actually produce a change in formatting.  That is,
           the content of the L event structure (as opposed to its attributes) is always what text should be
           displayed.

           There are, at first glance, three kinds of L links: URL, man, and pod.

           When a L<some_url> code is parsed, it produces this event structure:

             <L content-implicit="yes" to="that_url" type="url">
               that_url
             </L>

           The "type="url"" attribute is always specified for this type of L code.

           For example, this Pod source:

             L<http://www.perl.com/CPAN/authors/

           produces this event structure:

             <L content-implicit="yes" to="http://www.perl.com/CPAN/authors/" type="url">
               http://www.perl.com/CPAN/authors/
             </L>

           When a L<manpage(section)> code is parsed (and these are fairly rare and not terribly useful), it
           produces this event structure:

             <L content-implicit="yes" to="manpage(section)" type="man">
               manpage(section)
             </L>

           The "type="man"" attribute is always specified for this type of L code.

           For example, this Pod source:

             L<crontab(5)>

           produces this event structure:

             <L content-implicit="yes" to="crontab(5)" type="man">
               crontab(5)
             </L>

           In the rare cases where a man page link has a specified, that text appears in a section
           attribute. For example, this Pod source:

             L<crontab(5)/"ENVIRONMENT">

           will produce this event structure:

             <L content-implicit="yes" section="ENVIRONMENT" to="crontab(5)" type="man">
               "ENVIRONMENT" in crontab(5)
             </L>

           In the rare case where the Pod document has code like L<sometext|manpage(section)>, then the
           sometext will appear as the content of the element, the manpage(section) text will appear only as
           the value of the to attribute, and there will be no "content-implicit="yes"" attribute (whose
           presence means that the Pod parser had to infer what text should appear as the link text -- as
           opposed to cases where that attribute is absent, which means that the Pod parser did not have to
           infer the link text, because that L code explicitly specified some link text.)

           For example, this Pod source:

             L<hell itself!|crontab(5)>

           will produce this event structure:

             <L to="crontab(5)" type="man">
               hell itself!
             </L>

           The last type of L structure is for links to/within Pod documents. It is the most complex because
           it can have a to attribute, or a section attribute, or both. The "type="pod"" attribute is always
           specified for this type of L code.

           In the most common case, the simple case of a L<podpage> code produces this event structure:

             <L content-implicit="yes" to="Net::Ping" type="pod">
               podpage
             </L>

           For example, this Pod source:

             L<Net::Ping>

           produces this event structure:

             <L content-implicit="yes" to="Net::Ping" type="pod">
               Net::Ping
             </L>

           In cases where there is link-text explicitly specified, it is to be found in the content of the
           element (and not the attributes), just as with the L<sometext|manpage(section)> case discussed
           above.  For example, this Pod source:

             L<Perl Error Messages|perldiag>

           produces this event structure:

             <L to="perldiag" type="pod">
               Perl Error Messages
             </L>

           In cases of links to a section in the current Pod document, there is a section attribute instead
           of a to attribute.  For example, this Pod source:

             L</"Member Data">

           produces this event structure:

             <L content-implicit="yes" section="Member Data" type="pod">
               "Member Data"
             </L>

           As another example, this Pod source:

             L<the various attributes|/"Member Data">

           produces this event structure:

             <L section="Member Data" type="pod">
               the various attributes
             </L>

           In cases of links to a section in a different Pod document, there are both a section attribute
           and a to attribute.  For example, this Pod source:

             L<perlsyn/"Basic BLOCKs and Switch Statements">

           produces this event structure:

             <L content-implicit="yes" section="Basic BLOCKs and Switch Statements" to="perlsyn" type="pod">
               "Basic BLOCKs and Switch Statements" in perlsyn
             </L>

           As another example, this Pod source:

             L<SWITCH statements|perlsyn/"Basic BLOCKs and Switch Statements">

           produces this event structure:

             <L section="Basic BLOCKs and Switch Statements" to="perlsyn" type="pod">
               SWITCH statements
             </L>

           Incidentally, note that we do not distinguish between these syntaxes:

             L</"Member Data">
             L<"Member Data">
             L</Member Data>
             L<Member Data>    [deprecated syntax]

           That is, they all produce the same event structure, namely:

             <L content-implicit="yes" section="Member Data" type="pod">
               &#34;Member Data&#34;
             </L>

       events with an element_name of E or Z
           While there are Pod codes E<...> and Z<>, these do not produce any E or Z events -- that is,
           there are no such events as E or Z.

       events with an element_name of Verbatim
           When a Pod verbatim paragraph (AKA "codeblock") is parsed, it produces this event structure:

             <Verbatim start_line="543" xml:space="preserve">
               ...text...
             </Verbatim>

           The value of the start_line attribute will be the line number of the first line of this verbatim
           block.  The xml:space attribute is always present, and always has the value "preserve".

           The text content will have tabs already expanded.

       events with an element_name of head1 .. head4
           When a "=head1 ..." directive is parsed, it produces this event structure:

             <head1>
               ...stuff...
             </head1>

           For example, a directive consisting of this:

             =head1 Options to C<new> et al.

           will produce this event structure:

             <head1 start_line="543">
               Options to
               <C>
                 new
               </C>
                et al.
             </head1>

           "=head2" thru "=head4" directives are the same, except for the element names in the event
           structure.

       events with an element_name of over-bullet
           When an "=over ... =back" block is parsed where the items are a bulletted list, it will produce
           this event structure:

             <over-bullet indent="4" start_line="543">
               <item-bullet start_line="545">
                 ...Stuff...
               </item-bullet>
               ...more item-bullets...
             </over-bullet>

           The value of the indent attribute is whatever value is after the "=over" directive, as in "=over
           8".  If no such value is specified in the directive, then the indent attribute has the value "4".

           For example, this Pod source:

             =over

             =item *

             Stuff

             =item *

             Bar I<baz>!

             =back

           produces this event structure:

             <over-bullet indent="4" start_line="10">
               <item-bullet start_line="12">
                 Stuff
               </item-bullet>
               <item-bullet start_line="14">
                 Bar <I>baz</I>!
               </item-bullet>
             </over-bullet>

       events with an element_name of over-number
           When an "=over ... =back" block is parsed where the items are a numbered list, it will produce
           this event structure:

             <over-number indent="4" start_line="543">
               <item-number number="1" start_line="545">
                 ...Stuff...
               </item-number>
               ...more item-number...
             </over-bullet>

           This is like the "over-bullet" event structure; but note that the contents are "item-number"
           instead of "item-bullet", and note that they will have a "number" attribute, which some
           formatters/processors may ignore (since, for example, there's no need for it in HTML when
           producing an "<UL><LI>...</LI>...</UL>" structure), but which any processor may use.

           Note that the values for the number attributes of "item-number" elements in a given "over-number"
           area will start at 1 and go up by one each time.  If the Pod source doesn't follow that order
           (even though it really should should!), whatever numbers it has will be ignored (with the correct
           values being put in the number attributes), and an error message might be issued to the user.

       events with an element_name of over-text
           These events are are somewhat unlike the other over-* structures, as far as what their contents
           are.  When an "=over ... =back" block is parsed where the items are a list of text "subheadings",
           it will produce this event structure:

             <over-text indent="4" start_line="543">
               <item-text>
                 ...stuff...
               </item-text>
               ...stuff (generally Para or Verbatim elements)...
               <item-text>
               ...more item-text and/or stuff...
             </over-text>

           The indent attribute is as with the other over-* events.

           For example, this Pod source:

             =over

             =item Foo

             Stuff

             =item Bar I<baz>!

             Quux

             =back

           produces this event structure:

             <over-text indent="4" start_line="20">
               <item-text start_line="22">
                 Foo
               </item-text>
               <Para start_line="24">
                 Stuff
               </Para>
               <item-text start_line="26">
                 Bar
                   <I>
                     baz
                   </I>
                 !
               </item-text>
               <Para start_line="28">
                 Quux
               </Para>
             </over-text>

       events with an element_name of over-block
           These events are are somewhat unlike the other over-* structures, as far as what their contents
           are.  When an "=over ... =back" block is parsed where there are no items, it will produce this
           event structure:

             <over-block indent="4" start_line="543">
               ...stuff (generally Para or Verbatim elements)...
             </over-block>

           The indent attribute is as with the other over-* events.

           For example, this Pod source:

             =over

             For cutting off our trade with all parts of the world

             For transporting us beyond seas to be tried for pretended offenses

             He is at this time transporting large armies of foreign mercenaries to
             complete the works of death, desolation and tyranny, already begun with
             circumstances of cruelty and perfidy scarcely paralleled in the most
             barbarous ages, and totally unworthy the head of a civilized nation.

             =cut

           will produce this event structure:

             <over-block indent="4" start_line="2">
               <Para start_line="4">
                 For cutting off our trade with all parts of the world
               </Para>
               <Para start_line="6">
                 For transporting us beyond seas to be tried for pretended offenses
               </Para>
               <Para start_line="8">
                 He is at this time transporting large armies of [...more text...]
               </Para>
             </over-block>

       events with an element_name of item-bullet
           See "events with an element_name of over-bullet", above.

       events with an element_name of item-number
           See "events with an element_name of over-number", above.

       events with an element_name of item-text
           See "events with an element_name of over-text", above.

       events with an element_name of for
           TODO...

       events with an element_name of Data
           TODO...

More Pod::Simple Methods
       Pod::Simple provides a lot of methods that aren't generally interesting to the end user of an
       existing Pod formatter, but some of which you might find useful in writing a Pod formatter. They are
       listed below. The first several methods (the accept_* methods) are for declaring the capabilites of
       your parser, notably what "=for targetname" sections it's interested in, what extra N<...> codes it
       accepts beyond the ones described in the perlpod.

       "$parser->accept_targets( SOMEVALUE )"
           As the parser sees sections like:

               =for html  <img src="fig1.jpg">

           or

               =begin html

                 <img src="fig1.jpg">

               =end html

           ...the parser will ignore these sections unless your subclass has specified that it wants to see
           sections targetted to "html" (or whatever the formatter name is).

           If you want to process all sections, even if they're not targetted for you, call this before you
           start parsing:

             $parser->accept_targets('*');

       "$parser->accept_targets_as_text(  SOMEVALUE  )"
           This is like accept_targets, except that it specifies also that the content of sections for this
           target should be treated as Pod text even if the target name in "=for targetname" doesn't start
           with a ":".

           At time of writing, I don't think you'll need to use this.

       "$parser->accept_codes( Codename, Codename...  )"
           This tells the parser that you accept additional formatting codes, beyond just the standard ones
           (I B C L F S X, plus the two weird ones you don't actually see in the parse tree, Z and E). For
           example, to also accept codes "N", "R", and "W":

               $parser->accept_codes( qw( N R W ) );

           TODO: document how this interacts with =extend, and long element names

       "$parser->accept_directive_as_data( directive_name )"
       "$parser->accept_directive_as_verbatim( directive_name )"
       "$parser->accept_directive_as_processed( directive_name )"
           In the unlikely situation that you need to tell the parser that you will accept additional
           directives ("=foo" things), you need to first set the parset to treat its content as data (i.e.,
           not really processed at all), or as verbatim (mostly just expanding tabs), or as processed text
           (parsing formatting codes like B<...>).

           For example, to accept a new directive "=method", you'd presumably use:

               $parser->accept_directive_as_processed("method");

           so that you could have Pod lines like:

               =method I<$whatever> thing B<um>

           Making up your own directives breaks compatibility with other Pod formatters, in a way that using
           "=for target ..." lines doesn't; however, you may find this useful if you're making a Pod
           superset format where you don't need to worry about compatibility.

       "$parser->nbsp_for_S( BOOLEAN );"
           Setting this attribute to a true value (and by default it is false) will turn "S<...>" sequences
           into sequences of words separated by "\xA0" (non-breaking space) characters. For example, it will
           take this:

               I like S<Dutch apple pie>, don't you?

           and treat it as if it were:

               I like DutchE<nbsp>appleE<nbsp>pie, don't you?

           This is handy for output formats that don't have anything quite like an "S<...>" code, but which
           do have a code for non-breaking space.

           There is currently no method for going the other way; but I can probably provide one upon
           request.

       "$parser->version_report()"
           This returns a string reporting the $VERSION value from your module (and its classname) as well
           as the $VERSION value of Pod::Simple.  Note that perlpodspec requires output formats (wherever
           possible) to note this detail in a comment in the output format.  For example, for some kind of
           SGML output format:

               print OUT "<!-- \n", $parser->version_report, "\n -->";

       "$parser->pod_para_count()"
           This returns the count of Pod paragraphs seen so far.

       "$parser->line_count()"
           This is the current line number being parsed. But you might find the "line_number" event
           attribute more accurate, when it is present.

       "$parser->nix_X_codes(  SOMEVALUE  )"
           This attribute, when set to a true value (and it is false by default) ignores any "X<...>"
           sequences in the document being parsed.  Many formats don't actually use the content of these
           codes, so have no reason to process them.

       "$parser->merge_text(  SOMEVALUE  )"
           This attribute, when set to a true value (and it is false by default) makes sure that only one
           event (or token, or node) will be created for any single contiguous sequence of text.  For
           example, consider this somewhat contrived example:

               I just LOVE Z<>hotE<32>apple pie!

           When that is parsed and events are about to be called on it, it may actually seem to be four
           different text events, one right after another: one event for "I just LOVE ", one for "hot", one
           for " ", and one for "apple pie!". But if you have merge_text on, then you're guaranteed that it
           will be fired as one text event:  "I just LOVE hot apple pie!".

       "$parser->code_handler(  CODE_REF  )"
           This specifies code that should be called when a code line is seen (i.e., a line outside of the
           Pod).  Normally this is undef, meaning that no code should be called.  If you provide a routine,
           it should start out like this:

               sub get_code_line {  # or whatever you'll call it
                 my($line, $line_number, $parser) = @_;
                 ...
               }

           Note, however, that sometimes the Pod events aren't processed in exactly the same order as the
           code lines are -- i.e., if you have a file with Pod, then code, then more Pod, sometimes the code
           will be processed (via whatever you have code_handler call) before the all of the preceding Pod
           has been processed.

       "$parser->cut_handler(  CODE_REF  )"
           This is just like the code_handler attribute, except that it's for "=cut" lines, not code lines.
           The same caveats apply. "=cut" lines are unlikely to be interesting, but this is included for
           completeness.

       "$parser->whine( linenumber, complaint string )"
           This notes a problem in the Pod, which will be reported to in the "Pod Errors" section of the
           document and/or send to STDERR, depending on the values of the attributes "no_whining",
           "no_errata_section", and "complain_stderr".

       "$parser->scream( linenumber, complaint string )"
           This notes an error like "whine" does, except that it is not suppressable with "no_whining". This
           should be used only for very serious errors.

       "$parser->source_dead(1)"
           This aborts parsing of the current document, by switching on the flag that indicates that EOF has
           been seen.  In particularly drastic cases, you might want to do this.  It's rather nicer than
           just calling "die"!

       "$parser->hide_line_numbers( SOMEVALUE )"
           Some subclasses that indescriminately dump event attributes (well, except for ones beginning with
           "~") can use this object attribute for refraining to dump the "start_line" attribute.

       "$parser->no_whining( SOMEVALUE )"
           This attribute, if set to true, will suppress reports of non-fatal error messages.  The default
           value is false, meaning that complaints are reported.  How they get reported depends on the
           values of the attributes "no_errata_section" and "complain_stderr".

       "$parser->no_errata_section( SOMEVALUE )"
           This attribute, if set to true, will suppress generation of an errata section.  The default value
           is false -- i.e., an errata section will be generated.

       "$parser->complain_stderr( SOMEVALUE )"
           This attribute, if set to true will send complaints to STDERR.  The default value is false --
           i.e., complaints do not go to STDERR.

       "$parser->bare_output( SOMEVALUE )"
           Some formatter subclasses use this as a flag for whether output should have prologue and epilogue
           code omitted. For example, setting this to true for an HTML formatter class should omit the
           "<html><head><title>...</title><body>..." prologue and the "</body></html>" epilogue.

           If you want to set this to true, you should probably also set "no_whining" or at least
           "no_errata_section" to true.

       "$parser->preserve_whitespace( SOMEVALUE )"
           If you set this attribute to a true value, the parser will try to preserve whitespace in the
           output.  This means that such formatting conventions as two spaces after periods will be
           preserved by the parser.  This is primarily useful for output formats that treat whitespace as
           significant (such as text or *roff, but not HTML).

SEE ALSO
       Pod::Simple -- event-based Pod-parsing framework

       Pod::Simple::Methody -- like Pod::Simple, but each sort of event calls its own method (like
       "start_head3")

       Pod::Simple::PullParser -- a Pod-parsing framework like Pod::Simple, but with a token-stream
       interface

       Pod::Simple::SimpleTree -- a Pod-parsing framework like Pod::Simple, but with a tree interface

       Pod::Simple::Checker -- a simple Pod::Simple subclass that reads documents, and then makes a
       plaintext report of any errors found in the document

       Pod::Simple::DumpAsXML -- for dumping Pod documents as tidily indented XML, showing each event on its
       own line

       Pod::Simple::XMLOutStream -- dumps a Pod document as XML (without introducing extra whitespace as
       Pod::Simple::DumpAsXML does).

       Pod::Simple::DumpAsText -- for dumping Pod documents as tidily indented text, showing each event on
       its own line

       Pod::Simple::LinkSection -- class for objects representing the values of the TODO and TODO attributes
       of L<...> elements

       Pod::Escapes -- the module the Pod::Simple uses for evaluating E<...> content

       Pod::Simple::Text -- a simple plaintext formatter for Pod

       Pod::Simple::TextContent -- like Pod::Simple::Text, but makes no effort for indent or wrap the text
       being formatted

       perlpod

       perlpodspec

       perldoc

COPYRIGHT AND DISCLAIMERS
       Copyright (c) 2002 Sean M. Burke.  All rights reserved.

       This library is free software; you can redistribute it and/or modify it under the same terms as Perl
       itself.

       This program is distributed in the hope that it will be useful, but without any warranty; without
       even the implied warranty of merchantability or fitness for a particular purpose.

AUTHOR
       Sean M. Burke "sburke@cpan.org"



perl v5.8.8                                      2006-07-07                      Pod::Simple::Subclassing(3)

Did this document help you?
Yes: Tell us what works for you.
It’s good, but: Report typos, inaccuracies, and so forth.
It wasn’t helpful: Tell us what would have helped.