C Modules

C.3 LaTeXML::Package

Support for package implementations and document customization.


This package defines and exports most of the procedures users will need to customize or extend LaTeXML. The LaTeXML implementation of some package might look something like the following, but see the installed LaTeXML/Package directory for realistic examples.

  package LaTeXML::Package::pool;  # to put new subs & variables in common pool
  use LaTeXML::Package;            # to load these definitions
  use strict;                      # good style
  use warnings;
  # Load "anotherpackage"
  # A simple macro, just like in TeX
  DefMacro('\thesection', '\thechapter.\roman{section}');
  # A constructor defines how a control sequence generates XML:
  DefConstructor('\thanks{}', "<ltx:thanks>#1</ltx:thanks>");
  # And a simple environment ...
  # A math  symbol \Real to stand for the Reals:
  DefMath('\Real', "\x{211D}", role=>'ID');
  # Or a semantic floor:
  # More esoteric ...
  # Use a RelaxNG schema
  # Or use a special DocType if you have to:
  # DocType("rootelement",
  #         "-//Your Site//Your DocType",'your.dtd',
  #          prefix=>"http://whatever/");
  # Allow sometag elements to be automatically closed if needed
  Tag('prefix:sometag', autoClose=>1);
  # Don't forget this, so perl knows the package loaded.


This module provides a large set of utilities and declarations that are useful for writing ‘bindings’: LaTeXML-specific implementations of a set of control sequences such as would be defined in a LaTeX style or class file. They are also useful for controlling and customization of LaTeXML’s processing. See the See also section, below, for additional lower-level modules imported & re-exported.

To a limited extent (and currently only when explicitly enabled), LaTeXML can process the raw TeX code found in style files. However, to preserve document structure and semantics, as well as for efficiency, it is usually necessary to supply a LaTeXML-specific ‘binding’ for style and class files. For example, a binding mypackage.sty.ltxml would encode LaTeXML-specific implementations of all the control sequences in mypackage.sty so that \usepackage{mypackage} would work. Similarly for myclass.cls.ltxml. Additionally, document-specific bindings can be supplied: before processing a TeX source file, eg mydoc.tex, LaTeXML will automatically include the definitions and settings in mydoc.latexml. These .ltxml and .latexml files should be placed LaTeXML’s searchpaths, where will find them: either in the current directory or in a directory given to the –path option, or possibly added to the variable SEARCHPATHS).

Since LaTeXML mimics TeX, a familiarity with TeX’s processing model is critical. LaTeXML models: catcodes and tokens (See LaTeXML::Core::Token, LaTeXML::Core::Tokens) which are extracted from the plain source text characters by the LaTeXML::Core::Mouth; Macros, which are expanded within the LaTeXML::Core::Gullet; and Primitives, which are digested within the LaTeXML::Core::Stomach to produce LaTeXML::Core::Box, LaTeXML::Core::List. A key additional feature is the Constructors: when digested they generate a LaTeXML::Core::Whatsit which, upon absorbtion by LaTeXML::Core::Document, inserts text or XML fragments in the final document tree.

Notation: Many of the following forms take code references as arguments or options. That is, either a reference to a defined sub, eg. \&somesub, or an anonymous function sub { ... }. To document these cases, and the arguments that are passed in each case, we’ll use a notation like code($stomach,...).

Control Sequences

Many of the following forms define the behaviour of control sequences. While in TeX you’ll typically only define macros, LaTeXML is effectively redefining TeX itself, so we define Macros as well as Primitives, Registers, Constructors and Environments. These define the behaviour of these control sequences when processed during the various phases of LaTeX’s imitation of TeX’s digestive tract.


LaTeXML uses a more convienient method of specifying parameter patterns for control sequences. The first argument to each of these defining forms (DefMacro, DefPrimive, etc) is a prototype consisting of the control sequence being defined along with the specification of parameters required by the control sequence. Each parameter describes how to parse tokens following the control sequence into arguments or how to delimit them. To simplify coding and capture common idioms in TeX/LaTeX programming, latexml’s parameter specifications are more expressive than TeX’s \def or LaTeX’s \newcommand. Examples of the prototypes for familiar TeX or LaTeX control sequences are:

   DefPrimitive('\multiply Variable SkipKeyword:by Number',..
   DefPrimitive('\newcommand OptionalMatch:* DefToken[]{}', ...

The general syntax for parameter specification is


reads a regular TeX argument. spec can be omitted (ie. {}). Otherwise spec is itself a parameter specification and the argument is reparsed to accordingly. ({} is a shorthand for Plain.)


reads an LaTeX-style optional argument. spec can be omitted (ie. {}). Otherwise, if spec is of the form Default:stuff, then stuff would be the default value. Otherwise spec is itself a parameter specification and the argument, if supplied, is reparsed according to that specification. ([] is a shorthand for Optional.)


Reads an argument of the given type, where either Type has been declared, or there exists a ReadType function accessible from LaTeXML::Package::Pool. See the available types, below.


These forms invoke the parser for Type but pass additional Tokens to the reader function. Typically this would supply defaults or parameters to a match.


Similar to Type, but it is not considered an error if the reader returns undef.


Similar to OptionalType, but the value returned from the reader is ignored, and does not occupy a position in the arguments list.

The predefined argument Types are as follows.

Plain, Semiverbatim

Reads a standard TeX argument being either the next token, or if the next token is an {, the balanced token list. In the case of Semiverbatim, many catcodes are disabled, which is handy for URL’s, labels and similar.

Token, XToken

Read a single TeX Token. For XToken, if the next token is expandable, it is repeatedly expanded until an unexpandable token remains, which is returned.

Number, Dimension, Glue | MuGlue

Read an Object corresponding to Number, Dimension, Glue or MuGlue, using TeX’s rules for parsing these objects.

Until:match | XUntil:match>

Reads tokens until a match to the tokens match is found, returning the tokens preceding the match. This corresponds to TeX delimited arguments. For XUntil, tokens are expanded as they are matched and accumulated.


Reads tokens until the next open brace {. This corresponds to the peculiar TeX construct \def\foo#{....

Match:match(|match)* | Keyword:match(|match)*>

Reads tokens expecting a match to one of the token lists match, returning the one that matches, or undef. For Keyword, case and catcode of the matches are ignored. Additionally, any leading spaces are skipped.


Read tokens until a closing }, but respecting nested {} pairs.


Read a parenthesis delimited tokens, but does not balance any nested parentheses.

Undigested, Digested, DigestUntil:match

These types alter the usual sequence of tokenization and digestion in separate stages (like TeX). A Undigested parameter inhibits digestion completely and remains in token form. A Digested parameter gets digested until the (required) opening { is balanced; this is useful when the content would usually need to have been protected in order to correctly deal with catcodes. DigestUntil digests tokens until a token matching match is found.


Reads a token, expanding if necessary, and expects a control sequence naming a writable register. If such is found, it returns an array of the corresponding definition object, and any arguments required by that definition.

SkipSpaces, Skip1Space

Skips one, or any number of, space tokens, if present, but contributes nothing to the argument list.

Common Options

scope=>'local' | 'global' | scope

Most defining commands accept an option to control how the definition is stored, for global or local definitions, or using a named scope A named scope saves a set of definitions and values that can be activated at a later time.

Particularly interesting forms of scope are those that get automatically activated upon changes of counter and label. For example, definitions that have scope=>'section:1.1' will be activated when the section number is ”1.1”, and will be deactivated when that section ends.


This option controls whether this definition is locked from further changes in the TeX sources; this keeps local ’customizations’ by an author from overriding important LaTeXML definitions and breaking the conversion.


DefMacro(prototype, expansion, %options);

Defines the macro expansion for prototype; a macro control sequence that is expanded during macro expansion time in the LaTeXML::Core::Gullet. The expansion should be one of tokensstringcode($gullet,@args)>: a string will be tokenized upon first usage. Any macro arguments will be substituted for parameter indicators (eg #1) in the tokens or tokenized string and the result is used as the expansion of the control sequence. If code is used, it is called at expansion time and should return a list of tokens as its result.

DefMacro options are




specifies a definition that will only be expanded in math mode; the control sequence must be a single character.


  DefMacro('\today',sub { ExplodeText(today()); });
DefMacroI(cs, paramlist, expansion, %options);

Internal form of DefMacro where the control sequence and parameter list have already been separated; useful for definitions from within code. Also, slightly more efficient for macros with no arguments (use undef for paramlist), and useful for obscure cases like defining \begin{something*} as a Macro.


DefConditional(prototype, test, %options);

Defines a conditional for prototype; a control sequence that is processed during macro expansion time (in the LaTeXML::Core::Gullet). A conditional corresponds to a TeX \if. If the test is undef, a \newif type of conditional is defined, which is controlled with control sequences like \footrue and \foofalse. Otherwise the test should be code($gullet,@args) (with the control sequence’s arguments) that is called at expand time to determine the condition. Depending on whether the result of that evaluation returns a true or false value (in the usual Perl sense), the result of the expansion is either the first or else code following, in the usual TeX sense.

DefConditional options are




This option is only used to define \ifcase.


  DefConditional('\ifmmode',sub {
     LookupValue('IN_MATH'); });
DefConditionalI(cs, paramlist, test, %options);

Internal form of DefConditional where the control sequence and parameter list have already been parsed; useful for definitions from within code. Also, slightly more efficient for conditinal with no arguments (use undef for paramlist).


IfCondition allows you to test a conditional from within perl. Thus something like if(IfCondition('\ifmmode')){ domath } else { dotext } might be equivalent to TeX’s \ifmmode domath \else dotext \fi.


DefPrimitive(prototype, replacement, %options);

Defines a primitive control sequence; a primitive is processed during digestion (in the LaTeXML::Core::Stomach), after macro expansion but before Construction time. Primitive control sequences generate Boxes or Lists, generally containing basic Unicode content, rather than structured XML. Primitive control sequences are also executed for side effect during digestion, effecting changes to the LaTeXML::Core::State.

The replacement can be a string used as the text content of a Box to be created (using the current font). Alternatively replacement can be code($stomach,@args) (with the control sequence’s arguments) which is invoked at digestion time, probably for side-effect, but returning Boxes or Lists or nothing. replacement may also be undef, which contributes nothing to the document, but does record the TeX code that created it.

DefPrimitive options are



mode=> ('text' | 'display_math' | 'inline_math')

Changes to this mode during digestion.


Specifies the font to use (see Fonts). If the font change is to only apply to material generated within this command, you would also use <bounded=1>>; otherwise, the font will remain in effect afterwards as for a font switching command.


If true, TeX grouping (ie. {}) is enforced around this invocation.



specifies whether the given constructor can only appear, or cannot appear, in math mode.


supplies a hook to execute during digestion just before the main part of the primitive is executed (and before any arguments have been read). The code should either return nothing (return;) or a list of digested items (Box’s,List,Whatsit). It can thus change the State and/or add to the digested output.


supplies a hook to execute during digestion just after the main part of the primitive ie executed. it should either return nothing (return;) or digested items. It can thus change the State and/or add to the digested output.


indicates whether this is a prefix type of command; This is only used for the special TeX assignment prefixes, like \global.


   DefPrimitive('\begingroup',sub { $_[0]->begingroup; });
DefPrimitiveI(cs, paramlist, code($stomach,@args), %options);

Internal form of DefPrimitive where the control sequence and parameter list have already been separated; useful for definitions from within code.


DefRegister(prototype, value, %options);

Defines a register with value as the initial value (a Number, Dimension, Glue, MuGlue or Tokens — I haven’t handled Box’s yet). Usually, the prototype is just the control sequence, but registers are also handled by prototypes like \count{Number}. DefRegister arranges that the register value can be accessed when a numeric, dimension, … value is being read, and also defines the control sequence for assignment.

Options are


specifies if it is not allowed to change this value.



By default value is stored in the State’s Value table under a name concatenating the control sequence and argument values. These options allow other means of fetching and storing the value.


DefRegisterI(cs, paramlist, value, %options);

Internal form of DefRegister where the control sequence and parameter list have already been parsed; useful for definitions from within code.


DefConstructor(prototype, $replacement, %options);

The Constructor is where LaTeXML really starts getting interesting; invoking the control sequence will generate an arbitrary XML fragment in the document tree. More specifically: during digestion, the arguments will be read and digested, creating a LaTeXML::Core::Whatsit to represent the object. During absorbtion by the LaTeXML::Core::Document, the Whatsit will generate the XML fragment according to replacement. The replacement can be code($document,@args,%properties) which is called during document absorbtion to create the appropriate XML (See the methods of LaTeXML::Core::Document).

More conveniently, replacement can be an pattern: simply a bit of XML as a string with certain substitutions to be made. The substitutions are of the following forms:

#1, #2 ... #name

These are replaced by the corresponding argument (for #1) or property (for #name) stored with the Whatsit. Each are turned into a string when it appears as in an attribute position, or recursively processed when it appears as content.


Another form of substituted value is prefixed with & which invokes a function. For example, &func(#1) would invoke the function func on the first argument to the control sequence; what it returns will be inserted into the document.


or ?test(ifpattern)(elsepattern)

Patterns can be conditionallized using this form. The test is any of the above expressions (eg. #1), considered true if the result is non-empty. Thus ?#1(<foo/>) would add the empty element foo if the first argument were given.


If the constuctor begins with ^, the XML fragment is allowed to float up to a parent node that is allowed to contain it, according to the Document Type.

The Whatsit property font is defined by default. Additional properties body and trailer are defined when captureBody is true, or for environments. By using $whatsit->setProperty(key=>$value); within afterDigest, or by using the properties option, other properties can be added.

DefConstructor options are








These options are the same as for Primitives


specifies the reversion of the invocation back into TeX tokens (if the default reversion is not appropriate). The textstring string can include #1, #2… The code is called with the $whatsit and digested arguments and must return a list of Token’s.


provides a control sequence to be used in the reversion instead of the one defined in the prototype. This is a convenient alternative for reversion when a ’public’ command conditionally expands into an internal one, but the reversion should be for the public command.


specifies how to compute (approximate) the displayed size of the object, if that size is ever needed (typically needed for graphics generation). If a string is given, it should contain only a sequence of #1 or #name to access arguments and properties of the Whatsit: the size is computed from these items layed out side-by-side. If code is given, it should return the three Dimensions (width, height and depth). If neither is given, and the reversion specification is of suitible format, it will be used for the sizer.

properties=>{%properties} | code($stomach,#1,#2...)

supplies additional properties to be set on the generated Whatsit. In the first form, the values can be of any type, but if a value is a code references, it takes the same args ($stomach,#1,#2,…) and should return the value; it is executed before creating the Whatsit. In the second form, the code should return a hash of properties.


supplies a hook to execute during digestion just before the Whatsit is created. The code should either return nothing (return;) or a list of digested items (Box’s,List,Whatsit). It can thus change the State and/or add to the digested output.


supplies a hook to execute during digestion just after the Whatsit is created (and so the Whatsit already has its arguments and properties). It should either return nothing (return;) or digested items. It can thus change the State, modify the Whatsit, and/or add to the digested output.


supplies a hook to execute before constructing the XML (generated by replacement).


Supplies code to execute after constructing the XML.


if true, arbitrary following material will be accumulated into a ‘body’ until the current grouping level is reverted, or till the Token is encountered if the option is a Token. This body is available as the body property of the Whatsit. This is used by environments and math.


This gives a number of args for cases where it can’t be infered directly from the prototype (eg. when more args are explicitly read by hooks).

DefConstructorI(cs, paramlist, replacement, %options);

Internal form of DefConstructor where the control sequence and parameter list have already been separated; useful for definitions from within code.

DefMath(prototype, tex, %options);

A common shorthand constructor; it defines a control sequence that creates a mathematical object, such as a symbol, function or operator application. The options given can effectively create semantic macros that contribute to the eventual parsing of mathematical content. In particular, it generates an XMDual using the replacement tex for the presentation. The content information is drawn from the name and options

DefMath accepts the options:










These options are the same as for Constructors


gives a name attribute for the object


gives the OpenMath content dictionary that name is from.


adds a grammatical role attribute to the object; this specifies the grammatical role that the object plays in surrounding expressions. This direly needs documentation!

mathstyle=>('display' | 'text' | 'script' | 'scriptscript')

Controls whether the this object will be presented in a specific mathstyle, or according to the current setting of mathstyle.

scriptpos=>('mid' | 'post')

Controls the positioning of any sub and super-scripts relative to this object; whether they be stacked over or under it, or whether they will appear in the usual position. TeX.pool defines a function doScriptpos() which is useful for operators like \sum in that it sets to mid position when in displaystyle, otherwise post.


Whether or not the object is stretchy when displayed.




These three are similar to role, scriptpos and stretchy, but are used in unusual cases. These apply to the given attributes to the operator token in the content branch.


Normally, these commands are digested with an implicit grouping around them, localizing changes to fonts, etc; noggroup=>1 inhibits this.


     role=>'ID', meaning=>'infinity');
DefMathI(cs, paramlist, tex, %options);

Internal form of DefMath where the control sequence and parameter list have already been separated; useful for definitions from within code.


DefEnvironment(prototype, replacement, %options);

Defines an Environment that generates a specific XML fragment. replacement is of the same form as for DefConstructor, but will generally include reference to the #body property. Upon encountering a \begin{env}: the mode is switched, if needed, else a new group is opened; then the environment name is noted; the beforeDigest hook is run. Then the Whatsit representing the begin command (but ultimately the whole environment) is created and the afterDigestBegin hook is run. Next, the body will be digested and collected until the balancing \end{env}. Then, any afterDigest hook is run, the environment is ended, finally the mode is ended or the group is closed. The body and \end{env} whatsit are added to the \begin{env}’s whatsit as body and trailer, respectively.

DefEnvironment takes the following options:







These options are the same as for Primitives






These options are the same as for DefConstructor


This hook is similar to that for DefConstructor, but it applies to the \begin{environment} control sequence.


This hook is similar to DefConstructor’s afterDigest but it applies to the \begin{environment} control sequence. The Whatsit is the one for the beginning control sequence, but represents the environment as a whole. Note that although the arguments and properties are present in the Whatsit, the body of the environment is not yet available!


This hook is similar to DefConstructor’s beforeDigest but it applies to the \end{environment} control sequence.


This hook is simlar to DefConstructor’s afterDigest but it applies to the \end{environment} control sequence. Note, however that the Whatsit is only for the ending control sequence, not the Whatsit for the environment as a whole.


This option supplies a hook to be executed during digestion after the ending control sequence has been digested (and all the 4 other digestion hook have executed) and after the body of the environment has been obtained. The Whatsit is the (useful) one representing the whole environment, and it now does have the body and trailer available, stored as a properties.


     "<ltx:emph>#1</ltx:emph", mode=>'text');
DefEnvironmentI(name, paramlist, replacement, %options);

Internal form of DefEnvironment where the control sequence and parameter list have already been separated; useful for definitions from within code.

Inputing Content and Definitions

FindFile(name, %options);

Find an appropriate file with the given name in the current directories in SEARCHPATHS. If a file ending with .ltxml is found, it will be preferred.

Note that if the name starts with a recognized protocol (currently one of (literal|http|https|ftp)) followed by a colon, the name is returned, as is, and no search for files is carried out.

The options are:


specifies the file type. If not set, it will search for both name.tex and name.


inhibits searching for a LaTeXML binding (name.type.ltxml) to use instead of the file itself.


inhibits searching for raw tex version of the file. That is, it will only search for the LaTeXML binding.

InputContent(request, %options);

InputContent is used for cases when the file (or data) is plain TeX material that is expected to contribute content to the document (as opposed to pure definitions). A Mouth is opened onto the file, and subsequent reading and/or digestion will pull Tokens from that Mouth until it is exhausted, or closed.

In some circumstances it may be useful to provide a string containing the TeX material explicitly, rather than referencing a file. In this case, the literal pseudo-protocal may be used:


If a file named $request.latexml exists, it will be read in as if it were a latexml binding file, before processing. This can be used for adhoc customization of the conversion of specific files, without modifying the source, or creating more elaborate bindings.

The only option to InputContent is:


Inhibits signalling an error if no appropriate file is found.


Input is analogous to LaTeX’s \input, and is used in cases where it isn’t completely clear whether content or definitions is expected. Once a file is found, the approach specified by InputContent or InputDefinitions is used, depending on which type of file is found.

InputDefinitions(request, %options);

InputDefinitions is used for loading definitions, ie. various macros, settings, etc, rather than document content; it can be used to load LaTeXML’s binding files, or for reading in raw TeX definitions or style files. It reads and processes the material completely before returning, even in the case of TeX definitions. This procedure optionally supports the conventions used for standard LaTeX packages and classes (see RequirePackage and LoadClass).

Options for InputDefinitions are:


the file type to search for.


inhibits searching for a LaTeXML binding; only raw TeX files will be sought and loaded.


inhibits searching for raw TeX files, only a LaTeXML binding will be sought and loaded.


inhibits reporting an error if no appropriate file is found.

The following options are primarily useful when InputDefinitions is supporting standard LaTeX package and class loading.


indicates whether to pass in any options from the calling class or package.


indicates whether options processing should be handled.


specifies a list of options (in the ’package options’ sense) to be passed (possibly in addition to any provided by the calling class or package).


provides tokens or code to be processed by a name.type-h@@k macro.


fishy option that indicates that this definitions file should be treated as if it were defining a class; typically shows up in latex compatibility mode, or AMSTeX.

A handy method to use most of the TeX distribution’s raw TeX definitions for a package, but override only a few with LaTeXML bindings is by defining a binding file, say tikz.sty.ltxml, to contain

  InputDefinitions('tikz', type => 'sty', noltxml => 1);

which would find and read in tizk.sty, and then follow it by a couple of strategic LaTeXML definitions, DefMacro, etc.

Class and Packages

RequirePackage(package, %options);

Finds and loads a package implementation (usually package.sty.ltxml, unless noltxml is specified)for the requested package. It returns the pathname of the loaded package. The options are:


specifies the file type (default sty.


specifies a list of package options.


inhibits searching for the LaTeXML binding for the file (ie. name.type.ltxml


inhibits searching for raw tex version of the file. That is, it will only search for the LaTeXML binding.

LoadClass(class, %options);

Finds and loads a class definition (usually class.cls.ltxml). It returns the pathname of the loaded class. The only option is


specifies a list of class options.

LoadPool(pool, %options);

Loads a pool file (usually pool.pool.ltxml), one of the top-level definition files, such as TeX, LaTeX or AMSTeX. It returns the pathname of the loaded file.

DeclareOption(option, tokensstringcode($stomach));

Declares an option for the current package or class. The 2nd argument can be a string (which will be tokenized and expanded) or tokens (which will be macro expanded), to provide the value for the option, or it can be a code reference which is treated as a primitive for side-effect.

If a package or class wants to accomodate options, it should start with one or more DeclareOptions, followed by ProcessOptions().

PassOptions(name, ext, @options);

Causes the given @options (strings) to be passed to the package (if ext is sty) or class (if ext is cls) named by name.


Processes the options that have been passed to the current package or class in a fashion similar to LaTeX. The only option (to ProcessOptions is inorder=>boolean indicating whehter the (package) options are processed in the order they were used, like ProcessOptions*.


Process the options given explicitly in @options.


Arranges for @stuff to be carried out after the preamble, at the beginning of the document. @stuff should typically be macro-level stuff, but carried out for side effect; it should be tokens, tokens lists, strings (which will be tokenized), or code($gullet) which would yeild tokens to be expanded.

This operation is useful for style files loaded with --preload or document specific customization files (ie. ending with .latexml); normally the contents would be executed before LaTeX and other style files are loaded and thus can be overridden by them. By deferring the evaluation to begin-document time, these contents can override those style files. This is likely to only be meaningful for LaTeX documents.


Arranges for @stuff to be carried out just before \\end{document}. These tokens can be used for side effect, or any content they generate will appear as the last children of the document.

Counters and IDs

NewCounter(ctr, within, %options);

Defines a new counter, like LaTeX’s \newcounter, but extended. It defines a counter that can be used to generate reference numbers, and defines \thectr, etc. It also defines an ”uncounter” which can be used to generate ID’s (xml:id) for unnumbered objects. ctr is the name of the counter. If defined, within is the name of another counter which, when incremented, will cause this counter to be reset. The options are


Specifies a prefix to be used to generate ID’s when using this counter


Not sure that this is even sane.

$num = CounterValue($ctr);

Fetches the value associated with the counter $ctr.

$tokens = StepCounter($ctr);

Analog of \stepcounter, steps the counter and returns the expansion of \the$ctr. Usually you should use RefStepCounter($ctr) instead.

$keys = RefStepCounter($ctr);

Analog of \refstepcounter, steps the counter and returns a hash containing the keys refnum=$refnum, id=>$id>. This makes it suitable for use in a properties option to constructors. The id is generated in parallel with the reference number to assist debugging.

$keys = RefStepID($ctr);

Like to RefStepCounter, but only steps the ”uncounter”, and returns only the id; This is useful for unnumbered cases of objects that normally get both a refnum and id.


Resets the counter $ctr to zero.


Generates an ID for nodes during the construction phase, useful for cases where the counter based scheme is inappropriate. The calling pattern makes it appropriate for use in Tag, as in

   Tag('ltx:para',afterClose=>sub { GenerateID(@_,'p'); })

If $node doesn’t already have an xml:id set, it computes an appropriate id by concatenating the xml:id of the closest ancestor with an id (if any), the prefix (if any) and a unique counter.

Document Model

Constructors define how TeX markup will generate XML fragments, but the Document Model is used to control exactly how those fragments are assembled.

Tag(tag, %properties);

Declares properties of elements with the name tag. Note that Tag can set or add properties to any element from any binding file, unlike the properties set on control by DefPrimtive, DefConstructor, etc.. And, since the properties are recorded in the current Model, they are not subject to TeX grouping; once set, they remain in effect until changed or the end of the document.

The tag can be specified in one of three forms:

   prefix:name matches specific name in specific namespace
   prefix:*    matches any tag in the specific namespace;
   *           matches any tag in any namespace.

There are two kinds of properties:

Scalar properties

For scalar properties, only a single value is returned for a given element. When the property is looked up, each of the above forms is considered (the specific element name, the namespace, and all elements); the first defined value is returned.

The recognized scalar properties are:


Specifies whether tag can be automatically opened if needed to insert an element that can only be contained by tag. This property can help match the more SGML-like LaTeX to XML.


Specifies whether this tag can be automatically closed if needed to close an ancestor node, or insert an element into an ancestor. This property can help match the more SGML-like LaTeX to XML.

Code properties

These properties provide a bit of code to be run at the times of certain events associated with an element. All the code bits that match a given element will be run, and since they can be added by any binding file, and be specified in a random orders, a little bit of extra control is desirable.

Firstly, any early codes are run (eg afterOpen:early), then any normal codes (without modifier) are run, and finally any late codes are run (eg. afterOpen:late).

Within each of those groups, the codes assigned for an element’s specific name are run first, then those assigned for its package and finally the generic one (*); that is, the most specific codes are run first.

When code properties are accumulated by Tag for normal or late events, the code is appended to the end of the current list (if there were any previous codes added); for early event, the code is prepended.

The recognized code properties are:


Provides code to be run whenever a node with this tag is opened. It is called with the document being constructed, and the initiating digested object as arguments. It is called after the node has been created, and after any initial attributes due to the constructor (passed to openElement) are added.

afterOpen:early or afterOpen:late can be used in place of afterOpen; these will be run as a group bfore, or after (respectively) the unmodified blocks.


Provides code to be run whenever a node with this tag is closed. It is called with the document being constructed, and the initiating digested object as arguments.

afterClose:early or afterClose:late can be used in place of afterClose; these will be run as a group bfore, or after (respectively) the unmodified blocks.


Specifies the schema to use for determining document model. You can leave off the extension; it will look for schemaname.rng (and maybe eventually, .rnc if that is ever implemented).

RegisterNamespace(prefix, URL);

Declares the prefix to be associated with the given URL. These prefixes may be used in ltxml files, particularly for constructors, xpath expressions, etc. They are not necessarily the same as the prefixes that will be used in the generated document Use the prefix #default for the default, non-prefixed, namespace. (See RegisterDocumentNamespace, as well as DocType or RelaxNGSchema).

RegisterDocumentNamespace(prefix, URL);

Declares the prefix to be associated with the given URL used within the generated XML. They are not necessarily the same as the prefixes used in code (RegisterNamespace). This function is less rarely needed, as the namespace declarations are generally obtained from the DTD or Schema themselves Use the prefix #default for the default, non-prefixed, namespace. (See DocType or RelaxNGSchema).

DocType(rootelement, publicid, systemid, %namespaces);

Declares the expected rootelement, the public and system ID’s of the document type to be used in the final document. The hash %namespaces specifies the namespaces prefixes that are expected to be found in the DTD, along with each associated namespace URI. Use the prefix #default for the default namespace (ie. the namespace of non-prefixed elements in the DTD).

The prefixes defined for the DTD may be different from the prefixes used in implementation CODE (eg. in ltxml files; see RegisterNamespace). The generated document will use the namespaces and prefixes defined for the DTD.

Document Rewriting

During document construction, as each node gets closed, the text content gets simplfied. We’ll call it applying ligatures, for lack of a better name.

DefLigature(regexp, %options);

Apply the regular expression (given as a string: ”/fa/fa/” since it will be converted internally to a true regexp), to the text content. The only option is fontTest=>code($font); if given, then the substitution is applied only when fontTest returns true.

Predefined Ligatures combine sequences of ”.” or single-quotes into appropriate Unicode characters.


A Math Ligature typically combines a sequence of math tokens (XMTok) into a single one. A simple example is

   DefMathLigature(":=" => ":=", role => 'RELOP', meaning => 'assign');

replaces the two tokens for colon and equals by a token representing assignment. The options are those characterising an XMTok, namely: role, meaning and name.

For more complex cases (recognizing numbers, for example), you may supply a function matcher=CODE($document,$node)>, which is passed the current document and the last math node in the sequence. It should examine $node and any preceding nodes (using previousSibling) and return a list of ($n,$string,%attributes) to replace the $n nodes by a new one with text content being $string content and the given attributes. If no replacement is called for, CODE should return undef.

After document construction, various rewriting and augmenting of the document can take place.



These two declarations define document rewrite rules that are applied to the document tree after it has been constructed, but before math parsing, or any other postprocessing, is done. The %specification consists of a sequence of key/value pairs with the initial specs successively narrowing the selection of document nodes, and the remaining specs indicating how to modify or replace the selected nodes.

The following select portions of the document:


Selects the part of the document with label=$label


The scope could be ”label:foo” or ”section:1.2.3” or something similar. These select a subtree labelled ’foo’, or a section with reference number ”1.2.3”


Select those nodes matching an explicit xpath expression.


Selects nodes that look like what the processing of tex would produce.


Selects text nodes that match the regular expression.

The following act upon the selected node:


Adds the attributes given in the hash reference to the node.


Interprets replacement as TeX code to generate nodes that will replace the selected nodes.

Mid-Level support

$tokens = Expand($tokens);

Expands the given $tokens according to current definitions.

$boxes = Digest($tokens);

Processes and digestes the $tokens. Any arguments needed by control sequences in $tokens must be contained within the $tokens itself.

@tokens = Invocation($cs,@args);

Constructs a sequence of tokens that would invoke the token $cs on the arguments.

RawTeX('... tex code ...');

RawTeX is a convenience function for including chunks of raw TeX (or LaTeX) code in a Package implementation. It is useful for copying portions of the normal implementation that can be handled simply using macros and primitives.


Gives $token1 the same ‘meaning’ (definition) as $token2; like TeX’s \let.

StartSemiVerbatim(); ... ; EndSemiVerbatim();

Disable disable most TeX catcodes.

$tokens = Tokenize($string);

Tokenizes the $string using the standard catcodes, returning a LaTeXML::Core::Tokens.

$tokens = TokenizeInternal($string);

Tokenizes the $string according to the internal cattable (where @ is a letter), returning a LaTeXML::Core::Tokens.

Argument Readers


Reads from $gullet the tokens corresponding to $spec (a Parameters object).

DefParameterType(type, code($gullet,@values), %options);

Defines a new Parameter type, type, with code for its reader.

Options are:


This code is responsible for converting a previously parsed argument back into a sequence of Token’s.


whether it is an error if no matching input is found.


whether the value returned should contribute to argument lists, or simply be passed over.


whether the catcode table should be modified before reading tokens.

<DefColumnType(proto, expansion);

Defines a new column type for tabular and arrays. proto is the prototype for the pattern, analogous to the pattern used for other definitions, except that macro being defined is a single character. The expansion is a string specifying what it should expand into, typically more verbose column specification.

DefKeyVal(keyset, key, type, default);

Defines a keyword key used in keyval arguments for the set keyset. If type is given, it defines the type of value that must be supplied, such as 'Dimension'. If default is given, that value will be used when key is used without an equals and explicit value in a keyvals argument.

Access to State

$value = LookupValue($name);

Lookup the current value associated with the the string $name.


Assign $value to be associated with the the string $name, according to the given scoping rule.

Values are also used to specify most configuration parameters (which can therefor also be scoped). The recognized configuration parameters are:

 VERBOSITY         : the level of verbosity for debugging
                     output, with 0 being default.
 STRICT            : whether errors (eg. undefined macros)
                     are fatal.
 INCLUDE_COMMENTS  : whether to preserve comments in the
                     source, and to add occasional line
                     number comments. (Default true).
 PRESERVE_NEWLINES : whether newlines in the source should
                     be preserved (not 100% TeX-like).
                     By default this is true.
 SEARCHPATHS       : a list of directories to search for
                     sources, implementations, etc.

This function, along with the next three are like AssignValue, but maintain a global list of values. PushValue pushes the provided values onto the end of a list. The data stored for $name is global and must be a LIST reference; it is created if needed.


Similar to PushValue, but pushes a value onto the front of the list. The data stored for $name is global and must be a LIST reference; it is created if needed.


Removes and returns the value on the end of the list named by $name. The data stored for $name is global and must be a LIST reference. Returns undef if there is no data in the list.


Removes and returns the first value in the list named by $name. The data stored for $name is global and must be a LIST reference. Returns undef if there is no data in the list.


This function maintains a hash association named by $name. It returns the value associated with $key within that mapping. The data stored for $name is global and must be a HASH reference. Returns undef if there is no data associated with $key in the mapping, or the mapping is not (yet) defined.


This function associates $value with $key within the mapping named by $name. The data stored for $name is global and must be a HASH reference; it is created if needed.

$value = LookupCatcode($char);

Lookup the current catcode associated with the the character $char.


Set $char to have the given $catcode, with the assignment made according to the given scoping rule.

This method is also used to specify whether a given character is active in math mode, by using math:$char for the character, and using a value of 1 to specify that it is active.

$meaning = LookupMeaning($token);

Looks up the current meaning of the given $token which may be a Definition, another token, or the token itself if it has not otherwise been defined.

$defn = LookupDefinition($token);

Looks up the current definition, if any, of the $token.


Install the Definition $defn into $STATE under its control sequence.


Tests whether the two tokens are equal in the sense that they are either equal tokens, or if defined, have the same definition.



Set the current font by merging the font style attributes with the current font. The %fontspec specifies the properties of the desired font. Likely values include (the values aren’t required to be in this set):

 family : serif, sansserif, typewriter, caligraphic,
          fraktur, script
 series : medium, bold
 shape  : upright, italic, slanted, smallcaps
 size   : tiny, footnote, small, normal, large,
          Large, LARGE, huge, Huge
 color  : any named color, default is black

Some families will only be used in math. This function returns nothing so it can be easily used in beforeDigest, afterDigest.


Declares a font map for the encoding $name. The map $map is an array of 128 or 256 entries, each element is either a unicode string for the representation of that codepoint, or undef if that codepoint is not supported by this encoding. The only option currently is family used because some fonts (notably cmr!) have different glyphs in some font families, such as family=’typewriter’>.


Returns the unicode string representing the given codepoint $code (an integer) in the given font encoding $encoding. If $encoding is undefined, the usual case, the current font encoding and font family is used for the lookup. Explicit decoding is used when \\char or similar are invoked ($implicit is false), and the codepoint must be represented in the fontmap, otherwise undef is returned. Implicit decoding (ie. $implicit is true) occurs within the Stomach when a Token’s content is being digested and converted to a Box; in that case only the lower 128 codepoints are converted; all codepoints above 128 are assumed to already be Unicode.

The font map for $encoding is automatically loaded if it has not already been loaded.


Returns the unicode string resulting from decoding the individual characters in $string according to FontDecode, above.


Finds and loads the font map for the encoding named $encoding, if it hasn’t been loaded before. It looks for encoding.fontmap.ltxml, which would typically define the font map using DeclareFontMap, possibly including extra maps for families like typewriter.



Lookup the color object associated with $name.


Associates the $name with the given $color (a color object), with the given scoping.


Defines a color model $model that is derived from the core color model $coremodel. The two functions $tocore and $fromcore convert a color object in that model to the core model, or from the core model to the derived model. Core models are rgb, cmy, cmyk, hsb and gray.

Low-level Functions


Cleans an $id of disallowed characters, trimming space.


Cleans a $label of disallowed characters, trimming space. The prefix $prefix is prepended (or LABEL, if none given).


Cleans an index key, so it can be used as an ID.


Cleans a bibliographic citation key, so it can be used as an ID.


Cleans a url.


Generates a UTF character, handy for the the 8 bit characters. For example, UTF(0xA0) generates the non-breaking space.

@tokens = roman($number);

Formats the $number in (lowercase) roman numerals, returning a list of the tokens.

@tokens = Roman($number);

Formats the $number in (uppercase) roman numerals, returning a list of the tokens.

See also