Jean-Damien Durand
and 2 contributors


MarpaX::ESLIF::BNF - MarpaX::ESLIF's BNF


version 2.0.11


MarpaX::ESLIF is a Scanless Interface expressed in a BNF format, that is using marpaWrapper, itself being a thin interface on top of libmarpa parser.


The MarpaX::ESLIF BNF is composed of unicode characters, in any encoding supported by the underlying convertor (ICU or iconv, in order of preference). Unsignificant whitespaces, Perl-like comments and C++-like comments are discarded.

Symbol names

They consist of bare names, or can be enclosed in angle brackets if whitespace if desired. They are case sensitive, and can be composed only of ASCII characters. There is no attempt to discard any leading, trailing, or repeated whitespace in angle brackets version, i.e. all the followings are different symbol names:

  <this >
  < this

The grammar can contain multiple levels, the level syntax being:

  ::=         # Alias for level 0
    ~         # Alias for level 1
  :[\d]+:=    # General form

The level 0 must exist. We will use only ::= and/or ~ in the rest of this document for conveniene, though we are adressing any possible level.


There are three types of explicit terminals.

  • Strings

    They can be single or double-quoted, the content is any valid unicode character, and the \ character can be used to escape the quote or \ itself. The :i modifier can be used to force case-insensitive match:

  • Character classes

    They are always enclosed with left and right brackets []. Modifiers can start after a : character. A character class class is nothing else but a lexically restricted regular expression.

  • Regular expression

    They are always enclosed within slashes //, and the content must be valid as per the PCRE2 Perl Compatible Regular Expression library. Modifiers can start after the slash on the right. Regular expression patterns are by default anchored. The slash character itself must be preceeded by a backslash, i.e. \/ in the string seen by the parser (so, in practice, it is coded like this: "\\/").

The PCRE2 syntax is supported in its entirety, this include any PCRE2 add-on. Character classes and regular expression share the same set of modifiers, executed in order of appearance, that are:

  Modifiers   Explanation
  e           Unset back-references in the pattern will match to empty strings
  i           Case-insensitive
  j           \u, \U and \x and unset back-references will act as JavaScript standard
  m           Multi-line regex
  n           Enable Unicode properties and extend meaning of meta-characters
  s           A dot meta-character in the pattern matches all characters, including newlines
  x           Enable comments. This has some limitation due MarpaX::ESLIF semantics
  D           A dollar meta-character matches only at the end of the subject string
  J           Allow duplicate names for sub-patterns
  U           Inverts the "greediness" of the quantifiers
  a           Meta-characters will be limited to their ASCII equivalent
  u           Forces support of large codepoints
  b           Could mean "forced binary" mode
  c           Could mean "forced unicode character" mode
  A           Remove the systematic anchoring

Internally this correspond to this set of options in PCRE2:

  Modifiers         PCRE2 flag unset   PCR2 flag set
  e                                    PCRE2_MATCH_UNSET_BACKREF
  i                                    PCRE2_CASELESS
  j                                    PCRE2_ALT_BSUX|PCRE2_MATCH_UNSET_BACKREF
  m                                    PCRE2_MULTILINE
  n                                    PCRE2_UCP
  s                                    PCRE2_DOTALL
  x                                    PCRE2_EXTENDED
  D                                    PCRE2_DOLLAR_ENDONLY
  J                                    PCRE2_DUPNAMES
  U                                    PCRE2_UNGREEDY
  a                 PCRE2_UTF
  N                 PCRE2_UCP
  u                                    PCRE2_UTF
  b                 PCRE2_UTF          PCRE2_NEVER_UTF
  c                 PCRE2_NEVER_UTF    PCRE2_UTF
  A                 PCRE2_ANCHORED

Lexemes are meta-symbols that does appear as the LHS symbol anywhere within the current grammar. Therefore they behave like terminals, except that their definition is not in the current grammar. By default such meta-symbol is looked up at the next level. For example:

  rule      ::= something
  something   ~ [\d]

say that symbol something at grammar level 0 is a reference to something at grammar level 1.

Lexemes can be references:


Without any indication, a lexeme is always assumed to be at the grammar of the next level

explicity by grammar description
  X ::= Y@'Grammar Description'

This is working because a grammar description is unique across all sub-grammars.

explicitely by relative level
  X ::= Y@+1
  X ::= Y@-2
  X ::= Y@3

The signed integer is interpreted as a delta with current grammar level.

explicitely by absolute level
  X ::= Y@=1

The unsigned integer is interpreted as an explicit grammar level.


Everytime expected terminals cannot be match, MarpaX::ESLIF will try to match the special rule :discard. The :discard rule also have precedence if it matches longer than the longest acceptable lexeme. and can not be ambiguous (else discard silently fail).

Grammar meta settings

Start rule

By default, the first symbol of a grammar of level n is its start symbol. This can be set once with e.g.:

  :start ::= symbolname
Grammar description

By default, a grammar of level n has the description Grammar level n. This can be set once with e.g.:

  :desc ::= 'A single-quoted string'

By default, symbol action is ::transfer and rule action is ::concat, i.e. the parse tree value of a grammar is a binary concatenation of every input representation (see the representation section below), without the eventual discard. Stack manipulation may require the trigger of a free function, and this has no default. Only expected terminals or lexemes are looked up, this is the Longest Acceptable Token Match (LATM) setting, defaulting to a true value. You should not change that. Defaults can be set once, for example like this:

  :default ::= action        => defaultRuleAction
               latm          => 1
               symbol-action => defaultSymbolAction
               free-action   => defaultFreeAction

Predefined actions are available for rules and symbols. Please refer the API documentation to know more about value types.

The free-action adverb is required when the end-user is pushing an opaque value of type PTR or ARRAY that is not declared shallowed. The internal stack manipulation may decide it does not need this value anymore: the free action is then called. This is used in low-level programming, in practice interface to high-level languages like Perl, Java etc... should hide that, making the presence of this keyword in the grammar meaningless.

The symbol-action adverb is how a match within a sub-grammar is transfered.


Creates a value of type UNDEF.

Meaningful for both rule and symbol actions.


Creates a value of type PTR from the right-hand side representation, guaranteed to be a NUL byte terminated sequence of ASCII characters, or UNDEF if representation is empty. Please refer to the representation section below.

Meaningful for both rule and symbol actions.


Creates a value of type ARRAY from the right-hand side representation encoded in the charset specified within the brackets, or UNDEF if representation is empty. iconv convention is used for the charset, i.e. a charset name, followed by eventual options like //TRANSLIT and/or //IGNORE. Any other option depend on how the tconv library is built, and may probably not be supported. Note that using the iconv notation does not mean that this is iconv running behind.

Please refer to the representation section below.

Meaningful for both rule and symbol actions.


Creates a value of type ARRAY from the binary concatenatation all the RHS's representation, or UNDEF if representation is empty.

Please refer to the representation section below.

Meaningful for both rule and symbol actions, and is the default rule action.


Copies the RHS number x (first RHS is at indice 0), putting UNDEF if it does not exist. This action is the only one that guarantees that the nature of the RHS value is unchanged.

Meaningful only for rule actions.


Alias for ::copy[0].


Copies the single RHS number value. This action guarantees that the nature of the RHS value is unchanged.

Meaningful only for symbol actions, and is the default symbol action.


The :discard symbol, despite belonging to a given grammar, is not accessible directly, and can only be set as a meta setting. An event can be associated upon discard completion, there can be multiple :discard statements:

  :discard ::= symbolname1 event => discard_symbolname1$
  :discard ::= symbolname2 event => discard_symbolname2$

Note than when an event is set, this will be triggered only on the :discard's RHS completion, therefore the RHS of the :discard must be an LHS in the same grammar when there is an event setting.

  • Event names

    They are composed of a restricted set of the ASCII graph characters. Special cases are:


    Transformed to the symbol name for which the event is triggered.


    Hook that is disabling :discard rule for the current recognizer. Equivalent to a call to MarpaX::ESLIF::recognizer_hook_discardb(1). Not propagated. Take care, this is a permanent setting.


    Hook that is enabling :discard rule for the current recognizer. Equivalent to a call to MarpaX::ESLIF::recognizer_hook_discardb(0). Not propagated. Take care, this is a permanent setting.

    Please note that the :discard[on] and :discard[off] events will always happen if specified in the grammar, at any level of parsing, regardless if you the grammar's parse() method. This is because these are internal events, not visible by the user.

  • Event initializers

    By default, events are on, this is equivalent to appending =on after the event name. The =off characters are putting event off at startup.

Lexemes are different than non-lexeme symbols because they are treated in the grammar are terminals, others are not.

Lexeme events

Meta symbols that are lexemes can have pause events, before mean that the scanning recognized them, after mean they have been consumed, e.g.:

  :lexeme ::= symbolname1 pause => before event => ^symbolname1
  :lexeme ::= symbolname2 pause => after  event =>  symbolname2$2

It is not allowed to set a lexeme event on a symbol that is not a lexeme.

Non-lexeme events

Completion, predicted or nulled events are supported, targetting a symbol name.

For example:

  event a     = completed  symbolname
  event b=off = nulled     symbolname
  event c=on  = predicted ^symbolname

It is not allowed to a a non-lexeme event on a symbol that is a lexeme.


Rules can be autoranked, the higest of a set of alternative having the highest rank, default is off:

  autorank is on by default
  autorank is off by default
Inaccessible statements

Inaccessible statements can generate warnings, can be ignored, or be error on demand, default is to ignore them:

  inaccessible is warn by default
  inaccessible is ok by default
  inaccessible is fatal by default


A statement have a symbol name on the left-hand side (LHS) and zero or more symbol names, or terminals, on the right-hand side (RHS):

  LHS ::= RHS1 RHS2 etc...

There are two exceptions:

The exception statement

Its semantic is a single symbol name following by another single symbol name, with - in the middle:

  LHS ::= RHS1 - RHS2
The sequence statement

This is a single symbol name following by the * or the + character:

  LHS1 ::= RHS1*
  LHS2 ::= RHS2+

Empty rule have no RHS:


Eventual ambiguities in the grammar itself may be solved by adding the ; character at the end of a rule, or by enclosing zero or more statements within { and } characters:

    LHS1 ::= RHS1
    LHS2 ::= RHS2 - RHS3

There are two types of alternatives: the standard | meaning this is an or, or the loosen character || meaning that this is an alternative starting a prioritized group of alternatives, for example the calculator grammar is:

  Expression  ::=  /[\d]+/
                | '(' Expression ')'              assoc => group
                ||    Expression '**' Expression  assoc => right
                ||    Expression  '*' Expression
                |     Expression  '/' Expression
                ||    Expression  '+' Expression
                |     Expression  '-' Expression

which is strictly equivalent, in traditional BNF syntax to:

  Expression  ::= Expression0
  Expression0 ::= Expression1
  Expression1 ::= Expression2
  Expression2 ::= Expression3

  Expression3 ::= /[\d]+/
                | '(' Expression0 ')'
  Expression2 ::=  Expression3 '**' Expression2
  Expression1 ::=  Expression1  '*' Expression2
                |  Expression1  '/' Expression2
  Expression0 ::=  Expression0  '+' Expression1
  Expression0 ::=  Expression0  '-' Expression1

As you can see statements has been grouped at every occurence of || operator. Therefore the loosen operator || is a convenience operator, it is always possible to write an equivalent grammar without it, though this can become quite tedious. The assoc adverb has a meaning only in the presence of prioritized alternatives, else it has no effect.

The following is copied almost verbatim from the Marpa::R2 section on precedence:

In prioritized statements, every alternative has an arity. The arity is the number of times an operand appears on the RHS. A RHS symbol is an operand if and only if it is the same as the LHS symbol. Anything else is considered as an operator. When the arity is 0, precedence and associativy are meaningless and ignored. When the arity is 1, precedence has effect, but not left nor right associativity.

If arity is 2 or more and the alternative is left associative, the leftmost operand associates and operands after the first will have the next-tighest priority level. If arity is 2 or more and the alternative is right associative, the last operand associates and operands before the last will have the next-tighest priority level. In group associativity, all operands associate at the lowest priority.


Any rule can be followed by zero or more of these adverbs, if an adverb appears more than once, the latest is the winner:


During valuation, a specific action can be associated to a rule:

  action => my_action
Left association

In a prioritized statement, associate with the left-most operand:

  assoc => left
Right association

In a prioritized statement, associate with the right-most operand:

  assoc => right
Group association

All operands associate at the lowest priority:

  assoc => group

Sequence rules can have a separator, that can be a symbol name, a string, a character class or a regular expression.

  separator => comma
  separator => ','
  separator => [,]
  separator => /,/

Modifiers are allowed after string, character class or regular expressions.

Proper specification

Sequence rules can be proper, i.e. without trailing separator:

  proper => 1
Rank specification

During valuation, rules can have a rank to get prioritized. Rank is a signed integer and default to 0:

  rank => -2

Any other value but 0 is not allowed if autoranking is set to a true value.

Null-ranking specification

Nulling symbols can rank high low, the default is low.

  null-ranking => 'low'
  null-ranking => 'high'
Priority specification

Lexemes can be prioritized, using a signed integer:

  priority => 15
Pause specification

Scanner can be paused before a lexeme is recognized, or just after it has been completed:

  pause => before
  pause => after
Event specification

Events can be specified, with an eventual initializer, given that default initialization is =on:

  event => eventName
  event => eventName=on
  event => eventName=off

A name can be associated to the rule, in the form:

  name => something
  name => 'quoted name'  # No modifier is allowed after the string
  name => "quoted name"  # No modifier is allowed after the string



MarpaX::ESLIF BNF can be expressed in itself:

   * **********************
   * Meta-grammar settings:
   * **********************
  :discard                       ::= whitespace
  :discard                       ::= <perl comment> 
  :discard                       ::= <cplusplus comment> 

   * ******
   * Rules:
   * ******
  <statements>                   ::= statement*

  <statement>                    ::= <start rule>
                                   | <desc rule>
                                   | <empty rule>
                                   | <null statement>
                                   | <statement group>
                                   | <priority rule>
                                   | <quantified rule>
                                   | <discard rule>
                                   | <default rule>
                                   | <lexeme rule>
                                   | <completion event declaration>
                                   | <nulled event declaration>
                                   | <prediction event declaration>
                                   | <inaccessible statement>
                                   | <exception statement>
                                   | <autorank statement>

  <start rule>                   ::= ':start' <op declare> symbol
  <desc rule>                    ::= ':desc' <op declare> <quoted name>
  <empty rule>                   ::= <lhs> <op declare> <adverb list>
  <null statement>               ::= ';'
  <statement group>              ::= '{' statements '}'
  <priority rule>                ::= lhs <op declare> <priorities>
  <quantified rule>              ::= lhs <op declare> <rhs primary> <quantifier> <adverb list>
  <discard rule>                 ::= ':discard' <op declare> <rhs primary> <adverb list>
  <default rule>                 ::= ':default' <op declare> <adverb list>
  <lexeme rule>                  ::= ':lexeme' <op declare> symbol <adverb list>
  <completion event declaration> ::= 'event' <event initialization> '=' 'completed' <symbol name>
                                   | 'event' <event initialization> <op declare> 'completed' <symbol name>
  <nulled event declaration>     ::= 'event' <event initialization> '=' 'nulled' <symbol name>
                                   | 'event' <event initialization> <op declare> 'nulled' <symbol name>
  <prediction event declaration> ::= 'event' <event initialization> '=' 'predicted' <symbol name>
                                   | 'event' <event initialization> <op declare> 'predicted' <symbol name>
  <inaccessible statement>       ::= 'inaccessible' 'is' <inaccessible treatment> 'by' 'default'
  <inaccessible treatment>       ::= 'warn'
  <inaccessible treatment>       ::= 'ok'
  <inaccessible treatment>       ::= 'fatal'
  <exception statement>          ::= lhs <op declare> <rhs primary> '-' <rhs primary> <adverb list>
  <autorank statement>           ::= 'autorank' 'is' <on or off> 'by' 'default'
  <op declare>                   ::= <op declare top grammar>
                                   | <op declare lex grammar>
                                   | <op declare any grammar>
  <priorities>                   ::= <alternatives>+ separator => <op loosen> proper => 1
  <alternatives>                 ::= <alternative>+ separator => <op equal priority> proper => 1
  <alternative>                  ::= rhs <adverb list>
  <adverb list>                  ::= <adverb list items>
  <adverb list items>            ::= <adverb item>*
  <adverb item>                  ::= <action>
                                   | <left association>
                                   | <right association>
                                   | <group association>
                                   | <separator specification>
                                   | <proper specification>
                                   | <rank specification>
                                   | <null ranking specification>
                                   | <priority specification>
                                   | <pause specification>
                                   | <latm specification>
                                   | naming
                                   | <null adverb>
                                   | <symbol action>
                                   | <free action>
                                   | <event specification>
  <action>                       ::= 'action' '=>' <action name>
  <left association>             ::= 'assoc' '=>' 'left'
  <right association>            ::= 'assoc' '=>' 'right'
  <group association>            ::= 'assoc' '=>' 'group'
  <separator specification>      ::= 'separator' '=>' <single symbol>
  <proper specification>         ::= 'proper' '=>' false
                                   | 'proper' '=>' true
  <rank specification>           ::= 'rank' '=>' <signed integer>
  <null ranking specification>   ::= 'null-ranking' '=>' <null ranking constant>
                                   | 'null' 'rank' '=>' <null ranking constant>
  <null ranking constant>        ::= 'low'
                                   | 'high'
  <priority specification>       ::= 'priority' '=>' <signed integer>
  <pause specification>          ::= 'pause' '=>' 'before'
                                   | 'pause' '=>' 'after'
  <event specification>          ::= 'event' '=>' <event initialization>
  <event initialization>         ::= <event name> <event initializer>
  <event initializer>            ::= '=' <on or off>
  <event initializer>            ::=
  <on or off>                    ::= 'on'
                                   | 'off'
  <latm specification>           ::= 'latm' '=>' <false>
                                   | 'latm' '=>' <true>
  naming                         ::= 'name' '=>' <alternative name>
  <null adverb>                  ::= ','
  <symbol action>                ::= 'symbol-action' '=>' <symbol action name>
  <free action>                  ::= 'free-action' '=>' <free name>
  <alternative name>             ::= <standard name>
  <alternative name>             ::= <quoted name>
  <event name>                   ::= <restricted ascii graph name>
                                   | ':symbol'
                                   | ':discard[on]'
                                   | ':discard[off]'
  lhs                            ::= <symbol name>
  rhs                            ::= <rhs primary>+
  <rhs primary>                  ::= <single symbol>
                                   | <symbol name> '@' <grammar reference>
  <single symbol>                ::= symbol
                                   | <character class>
                                   | <regular expression>
                                   | <quoted string>
  symbol                         ::= <symbol name>
  <symbol name>                  ::= <bare name>
                                   | <bracketed name>
  <action name>                  ::= <restricted ascii graph name>
                                   | '::shift'
                                   | '::undef'
                                   | '::ascii'
                                   | /::convert\[[^\]]+\]/
                                   | '::concat'
                                   | /::copy\[\d+\]/
  <free name>                    ::= <restricted ascii graph name>
  <symbol action name>           ::= <restricted ascii graph name>
                                   | '::transfer'
                                   | '::undef'
                                   | '::ascii'
                                   | /::convert\[[^\]]+\]/
                                   | '::concat'
  <quantifier>                   ::= '*'
                                   | '+'
  <signed integer>               ::= /[+-]?\d+/
  <unsigned integer>             ::= /\d+/
  <grammar reference>            ::= <quoted string>
                                   | <signed integer>
                                   | '=' <unsigned integer>

  # ---------------------------------------
  # Lexemes of the grammar given above are:
  # ---------------------------------------
  # <op declare any grammar> ::= <op declare any grammar>@+1
  # <op declare top grammar> ::= <op declare top grammar>@+1
  # <op declare lex grammar> ::= <op declare lex grammar>@+1
  # <op loosen> ::= <op loosen>@+1
  # <op equal priority> ::= <op equal priority>@+1
  # <false> ::= <false>@+1
  # <true> ::= <true>@+1
  # <standard name> ::= <standard name>@+1
  # <quoted name> ::= <quoted name>@+1
  # <quoted string> ::= <quoted string>@+1
  # <character class> ::= <character class>@+1
  # <regular expression> ::= <regular expression>@+1
  # <bare name> ::= <bare name>@+1
  # <bracketed name> ::= <bracketed name>@+1
  # <restricted ascii graph name> ::= <restricted ascii graph name>@+1
  # <whitespace> ::= <whitespace>@+1
  # <perl comment> ::= <perl comment>@+1
  # <cplusplus comment> ::= <cplusplus comment>@+1

   * *************
   * Lexeme rules:
   * *************
  whitespace                       ~ /[\s]+/
  <perl comment>                   ~ /(?:(?:#)(?:[^\n]*)(?:\n|\z))/u
  <cplusplus comment>              ~ /(?:(?:(?:\/\/)(?:[^\n]*)(?:\n|\z))|(?:(?:\/\*)(?:(?:[^\*]+|\*(?!\/))*)(?:\*\/)))/u
  <op declare any grammar>         ~ /:\[(\d+)\]:=/
  <op declare top grammar>         ~ '::='
  <op declare lex grammar>         ~ '~'
  <op loosen>                      ~ '||'
  <op equal priority>              ~ '|'
  <true>                           ~ '1'
  <false>                          ~ '0'
  <word character>                 ~ [\w]
  <one or more word characters>    ~ <word character>+ proper => 1
  <zero or more word characters>   ~ <word character>* proper => 1
  <restricted ascii graph name>    ~ /[-!#$%&()*+.\/;<>?@\[\\\]^_`|~A-Za-z0-9][-!#$%&()*+.\/:;<>?@\[\\\]^_`|~A-Za-z0-9]*/
  <bare name>                      ~ <word character>+ proper => 1
  <standard name>                  ~ [a-zA-Z] <zero or more word characters>
  <bracketed name>                 ~ '<' <bracketed name string> '>'
  <bracketed name string>          ~ /[\s\w]+/
  <quoted string>                  ~ /(?:(?|(?:')(?:[^\\']*(?:\\.[^\\']*)*)(?:')|(?:")(?:[^\\"]*(?:\\.[^\\"]*)*)(?:")))/su
                                   | /(?:(?|(?:')(?:[^\\']*(?:\\.[^\\']*)*)(?:')|(?:")(?:[^\\"]*(?:\\.[^\\"]*)*)(?:")))/su ':' /ic?/
  <quoted name>                    ~ /(?:(?|(?:')(?:[^\\']*(?:\\.[^\\']*)*)(?:')|(?:")(?:[^\\"]*(?:\\.[^\\"]*)*)(?:")))/su
  <character class>                ~ /((?:\[(?:(?>[^\[\]]+)|(?-1))*\]))/
                                   | /((?:\[(?:(?>[^\[\]]+)|(?-1))*\]))/ ':' /[eijmnsxDJUuaNbcA]+/
  <regular expression>             ~ /(?:(?|(?:\/)(?:[^\\\/]*(?:\\.[^\\\/]*)*)(?:\/)))/su
                                   | /(?:(?|(?:\/)(?:[^\\\/]*(?:\\.[^\\\/]*)*)(?:\/)))/su /[eijmnsxDJUuaNbcA]+/


marpaESLIF, tconv, ICU


Jean-Damien Durand <>


This software is copyright (c) 2017 by Jean-Damien Durand.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.