# Marpa::R3 is Copyright (C) 2016, Jeffrey Kegler.
#
# This module is free software; you can redistribute it and/or modify it
# under the same terms as Perl 5.10.1. For more details, see the full text
# of the licenses in the directory LICENSES.
#
# This program is distributed in the hope that it will be
# useful, but it is provided “as is” and without any express
# or implied warranties. For details, see the full text of
# of the licenses in the directory LICENSES.
=head1 Name
Marpa::R3::Scanless::DSL - The DSL for the Scanless interface
=head1 Synopsis
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
use Marpa::R3;
my $grammar = Marpa::R3::Scanless::G->new(
{ bless_package => 'My_Nodes',
source => \(<<'END_OF_SOURCE'),
:default ::= action => [values] bless => ::lhs
lexeme default = action => [ start, length, value ]
bless => ::name
:start ::= Script
Script ::= Expression+ separator => comma
comma ~ [,]
Expression ::=
Number bless => primary
| '(' Expression ')' bless => paren assoc => group
|| Expression '**' Expression bless => exponentiate assoc => right
|| Expression '*' Expression bless => multiply
| Expression '/' Expression bless => divide
|| Expression '+' Expression bless => add
| Expression '-' Expression bless => subtract
Number ~ [\d]+
:discard ~ whitespace
whitespace ~ [\s]+
# allow comments
:discard ~ <hash comment>
<hash comment> ~ <terminated hash comment> | <unterminated
final hash comment>
<terminated hash comment> ~ '#' <hash comment body> <vertical space char>
<unterminated final hash comment> ~ '#' <hash comment body>
<hash comment body> ~ <hash comment char>*
<vertical space char> ~ [\x{A}\x{B}\x{C}\x{D}\x{2028}\x{2029}]
<hash comment char> ~ [^\x{A}\x{B}\x{C}\x{D}\x{2028}\x{2029}]
END_OF_SOURCE
}
);
=for Marpa::R3::Display::End
=head1 About this document
This is the reference document for
the domain-specific language (DSL)
of Marpa's Scanless interface (SLIF).
The SLIF's DSL is
an extension of BNF.
The SLIF DSL is used to specify other
DSL's, and is therefore a "meta-DSL".
=head1 Lexical conventions
The SLIF source string consists of a series of rules,
pseudo-rules and statements.
Whitespace separates tokens, but is otherwise ignored.
=head2 Comment
A hash ("C<#>") character starts a comment,
which continues to the end of the line.
Comments are equivalent to whitespace.
=head2 Symbol names
Symbol names can be either "bare" or enclosed in angle brackets.
Bare symbol names must consist entirely of Perl word characters
(alphanumerics, plus the underscore).
Symbol names are case-sensitive.
The angle brackets, if used, serve to "quote"
the symbol name,
and will not be part of the explicit symbol name.
If angle brackets are used, symbol names may also contain whitespace,
as in
=for Marpa::R3::Display
ignore: 1
<op comma>
=for Marpa::R3::Display::End
A whitespace sequence inside angle brackets can
include any whitespace character that is legal in Perl,
including newlines.
This allows very long symbol names to be line wrapped,
if necessary.
Unlike the angle brackets,
the whitespace in a bracketed symbol token
B<does> become part of
the explicit symbol name,
but it does so in a "normalized" form.
Leading and trailing whitespace in the name is discarded,
and all other whitespace sequences are converted to a single ASCII
space character.
This means that
=for Marpa::R3::Display
ignore: 1
< op comma >
<op comma>
< op comma>
=for Marpa::R3::Display::End
and even
=for Marpa::R3::Display
ignore: 1
<op
comma>
=for Marpa::R3::Display::End
will all be regarded as the same symbol name.
The explicit form of that symbol name
is C<< <op comma> >>, except that, again, the
angle brackets are for clarity,
and are not part of the explicit name.
Explicit, reserved and internal symbol names are often
displayed between angle brackets,
regardless of whether the symbol was originally
specified in bracketed form.
When a SLIF symbol needs to be
referred to by name in Perl code,
it is the symbol's explicit name that is used.
=head2 Single-quoted strings
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
Expression ::=
Number bless => primary
| '(' Expression ')' bless => paren assoc => group
|| Expression '**' Expression bless => exponentiate assoc => right
|| Expression '*' Expression bless => multiply
| Expression '/' Expression bless => divide
|| Expression '+' Expression bless => add
| Expression '-' Expression bless => subtract
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: Case-insensitive characters examples
partial: 1
normalize-whitespace: 1
Child ~ 'cHILd':i
=for Marpa::R3::Display::End
Single quotes can be used in prioritized
rules to indicate character strings.
The characters inside the single quote
will be matched in the input, literally and one-for-one.
Single-quoted strings can contain any characters with
the exception of
single quotes and vertical whitespace.
Single-quoted strings do not allow "escaped" characters.
A backslash ("C<\>") represents itself and has no effect
on the interpretation of the next character.
If a rule needs to match one of the forbidden characters
(single quote or vertical whitespace), it must use a
character class.
Single-quoted strings are always interpreted at the
L0 level, but they may be used in either structural
or lexical rules.
When a single-quoted string is used in a structural rule,
Marpa creates a virtual L0 rule on behalf of the application.
This is handy, but it does have a real disadvantage --
the name of the virtual rule's LHS
will be one assigned automatically by Marpa.
When tracing and debugging parses and grammars,
these virtual LHS's can be harder for a programmer
to interpret.
A modifier can appear after the string.
It must appear immediately after the string,
with no intervening whitespace.
Currently only the "C<:ic>"
and "C<:i>" modifier are availables.
These have exactly the same effect -- they make the string
match case-insensitive.
=head2 Character classes
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
<vertical space char> ~ [\x{A}\x{B}\x{C}\x{D}\x{2028}\x{2029}]
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: Case-insensitive characters examples
partial: 1
normalize-whitespace: 1
word ~ [\w]:ic +
=for Marpa::R3::Display::End
A character class in square brackets ("C<[]>")
can be used in a RHS alternative of a prioritized rule,
a quantified rule or a discard pseudo-rule.
Marpa character classes may contain anything acceptable to Perl,
and follow the same escaping conventions as Perl's character classes.
Character classes are always interpreted at the
L0 level, but they may be used in either structural
or lexical rules.
When a character class is used in a structural rule,
Marpa creates a virtual L0 rule on behalf of the application.
This is handy, but it does have a real disadvantage --
the name of the virtual rule's LHS
will be one assigned automatically by Marpa.
When tracing and debugging parses and grammars,
these virtual LHS's can be harder for a programmer
to interpret.
An implementation note: character classes are
interpreted by Perl, but this involves minimal overhead
when the parse is of any length.
Each character class is
passed to Perl to interpret exactly once and the result is
memoized in a C language structure for future use.
The modifiers allowed after single-quoted strings are also allowed allowed after character classes.
Modifiers must appear immediately after the closing square bracket,
with no intervening whitespace.
For more details, see L<the section on single-quoted strings|/"Single-quoted strings">.
=head2 Event name
The name of an event may be either a bare name,
a single-quoted event name,
or an event pseudo-name.
A bare event name must be one or more word characters,
starting with an alphabetic character.
A single-quoted event name may contain any character
except a single quote or vertical space.
The whitespace in single-quoted event names is normalized
in similar fashion to the normalization of symbol names --
leading and trailing whitespace is removed,
and all sequences of internal whitespace are changed to
a single ASCII space character.
Names containing single quotes (which, in any case,
are impossible to add using
current syntax) are reserved for future use.
A event pseudo-name is a colon,
followed by one or more word characters.
As present, there is only one event pseudo-name,
C<:symbol>.
The C<:symbol> pseudo-name is only allowed in
L<discard pseudo-rules|/"Discard pseudo-rule">,
and in the
L<discard default statement|/"Discard default statement">.
If the name of an event is specified as
the C<:symbol> pseudo-name, an event's name will be
based on the RHS of its discard rule.
This will always be either a single symbol,
or a character class.
If the RHS is a single symbol,
the actual event name will be the name
of that symbol.
If the RHS is a character class,
the actual event name will be the string
specifying that character class in the DSL.
The string specifying the event name for a character
class is taken literally from the
text specifying the DSL.
This means that the character classes
"C<[\x3B]>" and "C<[:]>" will have two different
event names,
event though both classes
specify exactly the same set of characters.
=head1 Event initializer
=for Marpa::R3::Display
name: SLIF completed event statement synopsis
partial: 1
normalize-whitespace: 1
event 'a' = completed A
event 'b'=off = completed B
event 'c'=on = completed C
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF discard event statement synopsis 2
partial: 1
normalize-whitespace: 1
:discard ~ [,] event => comma=off
:discard ~ [;] event => 'semicolon'=on
:discard ~ [.] event => period
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF nulled event statement synopsis
partial: 1
normalize-whitespace: 1
event '!a' = nulled A
event '!b'=off = nulled B
event '!c'=on = nulled C
=for Marpa::R3::Display::End
An event initializer is an event name,
optionally with an explicit initialization.
If there is an explicit initialization,
it consists of the equal sign ('C<=>')
followed by a value indicating
the event's initial activation setting
(C<on> or C<off>).
If the initialization value is C<on>,
or if there is no explicit initialization,
the event's activation
setting is initially on.
If the initialization value is C<off>,
the event's activation
setting is initially off.
=head1 L0, G1 and lexemes
In reading this document, it is important to keep in mind
the distinction, on one hand, between L0 and G1 rules and,
on the other hand, between rules and lexemes.
G1 rules have a semantics,
which can be specified as described
in this document.
L0 rules simply recognize symbols in the input.
L0 rules do not have a semantics.
Top-level L0 rules correspond to a string in the input.
The top-level L0 rules are seen by G1 as lexemes,
and the string to which a top-level L0 rule
corresponds becomes the default value of the lexeme.
The L0 grammar can be thought of as similar
in behavior to a set of regular expressions
with the lexemes being seen as similar to named captures.
Lexemes are the symbols which form the interface between
G1 and L0. Lexemes, like G1 rules, have a semantics.
The semantics of lexemes is
specified separately from the
semantics of G1 rules,
as described below.
=head1 Statements
The SLIF DSL consists of a series of statements.
The statements are of three kinds, as indicated by their declarator:
=over 4
=item * G1 rule
The BNF operator ("C<::=>"),
coming between the LHS and the first RHS alternative of a rule,
indicates that the rule is
a G1 rule.
=item * L0 rule
The match operator ("C<~>"),
coming between the LHS and the first RHS alternative
of a rule,
indicates a L0 rule.
=item * Global statements
Global statements
are signified by the assignment operator ("C<=>").
The location of a statement in the DSL source
will never affect the result.
=back
Rules differ from statements in that
the effect of a rule
is sometimes lexical --
that is, the effect may vary depending
on the position of the rule in the DSL source.
Some rules are called pseudo-rules.
Pseudo-rules do not correspond to BNF rules,
but instead use the rule format as a convenient
way to express other information.
=head2 The structure of rules
Every rule declaration consists of, in order:
=over 4
=item * A left hand side (LHS).
This will be a
symbol or a pseudo-symbol.
=item * A declaration operator ("C<::=>" or "C<~>").
=item * A right side declaration, which contains one or more RHS
alternatives.
Details of the right side declaration vary by the type of rule.
For each type of rule,
the right side declaration is described in detail below.
=back
=head2 RHS alternatives
The right side declaration of a rule will often contain one or
more RHS alternatives.
A RHS alternative is a series of RHS primaries,
where a RHS primary may be a symbol name,
a character class,
or a single-quoted string.
A list of one or more adverbs is often
associated with the
RHS alternatives.
Each adverb consists of a keyword,
the adverb operator ("C<< => >>"),
and the adverb's value.
Within an alternative, primaries may be enclosed in parentheses.
A primary enclosed in parentheses is hidden from
L<Marpa's semantics|Marpa::R3::Semantics>.
A set of parentheses may contain more than one primary,
in which case the entire sequence of primaries is hidden,
as if they had been enclosed in parentheses individually.
"Hiding" primaries in this way can be convenient for primaries whose values
the semantics will ignore, perhaps because the value is constant.
For example,
in the following rule
=for Marpa::R3::Display
ignore: 1
a ::= b (',' c) d action => ::first
=for Marpa::R3::Display::End
there is
=over 4
=item *
A LHS, in this case the symbol "C<a>".
=item *
A declarator, "C<::=>", which indicates this is
a G1 rule.
=item *
A RHS alternative consisting of four RHS primaries.
The first RHS primary is the symbol "C<b>".
The second RHS primary is a short single-quoted string C<','>.
The third and fourth RHS primaries are symbols: "C<c>" and "C<d>".
The parentheses around the second and third RHS primaries "hide"
them from the semantics.
Marpa's semantics will see this as a rule with only two RHS values.
=item *
The adverb list associated with the RHS alternative,
consisting of a single adverb.
The adverb consists of its keyword "L<C<action>|/"action">",
followed by the adverb operator ("C<< => >>"),
and the adverb's value "C<::first>".
=back
The rule in the above example is one of a very common type:
a trivial prioritized rule.
A prioritized rule is one that contains one or more prioritized RHS alternatives.
Prioritized rules are the only rules which may contain more than one RHS
alternative, but even prioritized rules usually have only one RHS alternative.
If there is only one RHS alternative, as in this case, the prioritization
is B<trivial> --
there is only one priority.
=head2 Start rule
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
:start ::= Script
=for Marpa::R3::Display::End
By default, the start symbol of the grammar is the LHS of the
first G1 rule.
This default can be make explicit or overriden by using an explicit
start rule.
The LHS of this rule is the C<:start> pseudo-symbol.
Only one RHS alternative is allowed.
This RHS alternative must contain only one symbol name,
and that symbol will be the start symbol of the G1 grammar.
No adverbs should be associated with the RHS alternative.
Start rules must be G1 rules.
=head2 Empty rule
An empty rule is a rule with an empty RHS.
The empty RHS, technically, is a RHS alternative, one with zero RHS primaries.
The L<C<action>|/"action"> and
L<C<bless>|/"bless"> adverbs are allowed
for the empty RHS alternative,
but no others.
A empty rule makes its LHS symbol a nullable symbol.
=head2 Quantified rule
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
Script ::= Expression+ separator => comma
=for Marpa::R3::Display::End
A quantified rule has only one RHS alternative,
which is followed by a quantifier.
The RHS alternative must consist of a single RHS primary.
This RHS primary must be a symbol name or a character class.
The quantifer is either
a star ("C<*>"),
or a plus sign ("C<+>")
indicating, respectively, that the sequence rule has a minimum length
of 0 or 1.
Adverbs may be associated with the RHS alternative.
The adverb list must follow the quantifier.
The adverbs allowed are L<C<action>|/"action">,
L<C<bless>|/"bless">,
L<C<proper>|/"proper"> and
L<C<separator>|/"separator">.
=head2 Prioritized rule
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
Expression ::=
Number bless => primary
| '(' Expression ')' bless => paren assoc => group
|| Expression '**' Expression bless => exponentiate assoc => right
|| Expression '*' Expression bless => multiply
| Expression '/' Expression bless => divide
|| Expression '+' Expression bless => add
| Expression '-' Expression bless => subtract
=for Marpa::R3::Display::End
A prioritized rule contains a series of one or more RHS alternatives,
separated by either the alternation operator ("C<|>")
or the loosen operators ("C<||>").
In a typical grammar, most rules are prioritized rules,
but they are often trivially prioritized,
consisting of only one RHS alternative.
For brevity, RHS alternatives are often called B<alternatives>.
Each alternative may be followed by a list of associated adverbs.
The
L<C<action>|/"action">,
L<C<assoc>|/"assoc"> and
L<C<bless>|/"bless"> adverbs are allowed.
The RHS alternatives in a prioritized right hand side proceed
from tightest (highest) priority to loosest.
The double "or" symbol ("C<||>") is the "loosen" operator --
the alternatives after it
have a looser (lower) priority than the alternatives before it.
The single "or" symbol ("C<|>") is the ordinary "alternative" operator --
alternatives on each side of it have the same priority.
Associativity is specified using adverbs, as described below.
=head2 Discard pseudo-rule
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
:discard ~ whitespace
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF discard event synopsis
partial: 1
normalize-whitespace: 1
:discard ~ whitespace event => ws
=for Marpa::R3::Display::End
A discard pseudo-rule is a rule whose LHS is
the C<:discard> pseudo-symbol,
and which has only one RHS alternative.
The RHS alternative must contain
=over 4
=item *
exactly one symbol name, or
=item *
exactly one character class.
=back
The symbol specified by the RHS of a discard pseudo-rule
is called the B<discarded symbol>.
Discard pseudo-rules indicate that the discarded symbol is a top-level L0
symbol, but one which is not a lexeme.
When a discarded symbol is recognized,
it is not passed as a lexeme to the G1 parser, but is
(as the name suggests) discarded.
Discard pseudo-rules must be L0 rules.
Only the C<event> adverb is allowed.
Its value must be an event initializer.
The format of an event initializer is
described
L<above|/"Event initializer">.
If present, it defines a discard event,
as described in
L<the document on SLIF parse
events|Marpa::R3::Event/"Discard events">.
=head2 Default pseudo-rule
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
:default ::= action => [values] bless => ::lhs
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF symbol, name array descriptor example
partial: 1
normalize-whitespace: 1
:default ::= action => [symbol, name, values]
=for Marpa::R3::Display::End
The purpose of
the default pseudo-rule is to change the defaults for
rule adverbs.
Technically, it has one RHS alternative, but this must always contain
zero RHS primaries.
Default pseudo-rules do not affect the defaults for L0 rules
or for lexemes.
There may be more than one default pseudo-rule.
The scope of default pseudo-rules is lexical, applying only to rules
that appear afterwards in the DSL source.
Currently only the L<C<action>|/"action"> and L<C<bless>|/"bless"> adverbs
can be specified in a default pseudo-rule.
Each default pseudo-rule creates a completely new
set of defaults -- if an adverb is not specified,
the default is reset to its implicit value,
the value which it had prior to any explicit settings.
=head2 Lexeme pseudo-rule
=for Marpa::R3::Display
name: SLIF lexeme rule synopsis
partial: 1
normalize-whitespace: 1
:lexeme ~ <say keyword> priority => 1
=for Marpa::R3::Display::End
The purpose of
the C<:lexeme> pseudo-rule is to allow adverbs to
change the treatment of a lexeme.
This pseudo-rule always has exactly one RHS alternative,
and that RHS alternative must contain exactly one symbol.
This RHS symbol identifies the lexeme which the adverbs will affect.
The only adverbs allowed in a C<:lexeme> rule
are
L<C<event>|/"event">,
L<C<pause>|/"pause">,
and
L<C<priority>|/"priority">.
As a side effect, a C<:lexeme> pseudo-rule
declares that its RHS symbol is expected to be a lexeme.
This declaration does not "force" lexeme status --
if the symbol does not meet the criteria for a lexeme
based on its use in L0 and G1 rules,
the result will be a fatal error.
Applications may find this ability to "declare"
lexemes useful for debugging,
and for documenting grammars.
=head2 Discard default statement
=for Marpa::R3::Display
name: SLIF default discard event statement synopsis 1
partial: 1
normalize-whitespace: 1
discard default = event => :symbol=on
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF default discard event statement synopsis 2
partial: 1
normalize-whitespace: 1
discard default = event => :symbol
=for Marpa::R3::Display::End
The discard default statement changes the defaults for
discard pseudo-rules.
Only the default for
the L<C<event>|/"event">
adverb
can be specified in a lexeme default statement.
Only one discard default statement is allowed in a grammar.
Typically in a discard default statement,
the event name will be the pseudo-name
C<:symbol>.
For details about event pseudo-names,
see L<the section on event
names|/"Event name">.
=head2 Lexeme default statement
=for Marpa::R3::Display
name: SLIF DSL synopsis
partial: 1
normalize-whitespace: 1
lexeme default = action => [ start, length, value ]
bless => ::name
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF "time flies" DSL synopsis
partial: 1
normalize-whitespace: 1
lexeme default = action => [ name, value ]
=for Marpa::R3::Display::End
The lexeme default statement changes the defaults for
lexeme adverbs.
It only changes the defaults for lexemes,
and does not affect rules.
Only the defaults for
the L<C<action>|/"action">,
and L<C<bless>|/"bless">
adverbs
can be specified in a lexeme default statement.
Only one lexeme default statement is allowed in a grammar.
=head2 Named event statement
=for Marpa::R3::Display
name: SLIF completed event statement synopsis
partial: 1
normalize-whitespace: 1
event 'a' = completed A
event 'b'=off = completed B
event 'c'=on = completed C
event 'd' = completed D
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF nulled event statement synopsis
partial: 1
normalize-whitespace: 1
event '!a' = nulled A
event '!b'=off = nulled B
event '!c'=on = nulled C
event '!d' = nulled D
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: SLIF predicted event statement synopsis
partial: 1
normalize-whitespace: 1
event '^a' = predicted A
event '^b'=off = predicted B
event '^c'=on = predicted C
event '^d' = predicted D
=for Marpa::R3::Display::End
The named event statement sets up a
SLIF parse event.
A named event statement consists of, in order
=over 4
=item *
The C<event> keyword.
=item *
An event initializer,
as described in L<the section on
event initializers|/"Event initializer">.
=item *
An equal sign ('C<=>').
=item *
A keyword, which is one of C<completed>,
C<nulled>, or C<predicted>,
to indicate the event type.
=item *
A symbol name.
=back
The SLIF's event-triggering methods are
L<the Scanless recognizer's read()|Marpa::R3::Scanless::R/"read()">,
L<resume()|Marpa::R3::Scanless::R/"resume()">,
L<lexeme_complete()|Marpa::R3::Scanless::R/"lexeme_complete()">,
and L<lexeme_read()|Marpa::R3::Scanless::R/"lexeme_read()">.
If the condition described
by the named event statement
occurs during an event-triggering method,
the method will return immediately,
with the current location at the trigger location.
Once triggered,
named events may be queried using
L<the Scanless recognizer's events()
method|Marpa::R3::Scanless::R/"events()">.
For details, see
L<the document on SLIF parse
events|Marpa::R3::Event>.
=head2 Inaccessible symbol statement
=for Marpa::R3::Display
name: inaccessible is ok statement
partial: 1
normalize-whitespace: 1
inaccessible is ok by default
=for Marpa::R3::Display::End
=for Marpa::R3::Display
name: inaccessible is fatal statement
partial: 1
normalize-whitespace: 1
inaccessible is fatal by default
=for Marpa::R3::Display::End
Inaccessible symbols are symbols
which cannot be reached from the start symbol.
Often, they
are the result of an error in grammar writing.
But inaccessible symbols can also occur for legitimate reasons --
for example,
you may have rules and symbols in grammar intended for future use.
The default can be specified
or changed with a statement of the form:
=for Marpa::R3::Display
ignore: 1
inaccessible is TREATMENT by default
=for Marpa::R3::Display
where C<TREATMENT> is one of C<warn>, C<ok>, or C<fatal>.
C<fatal> indicates that an inaccessible symbol should be
a fatal error.
C<warn> indicates that Marpa should print a warning message,
but proceed with the parse.
C<warn> is the default.
C<ok> indicates that the parse should proceed without
warning messages.
=head1 Ambiguity
Marpa parses ambiguous grammars and the design of the SLIF exploits this.
A flexible, but potentially ambiguous, syntax is used.
Actual ambiguities are obvious to the human eye,
and users will create them,
so that
the techniques of this section
will rarely be needed.
If and when an actual ambiguity does occur, an error message reports
the ambiguity and its exact location.
It will always be possible to disambiguate a SLIF DSL,
and there will always
be more than one way to do this.
=head2 Separating statements with semicolons
=for Marpa::R3::Display
name: statements separated by semicolon
partial: 1
normalize-whitespace: 1
:default ::= action => ::array
quartet ::= a a a a;
inaccessible is warn by default
a ~ 'a'
=for Marpa::R3::Display::End
A statement may be terminated with a semicolon ("C<;>").
=head2 Grouping statements in curly braces
=for Marpa::R3::Display
name: statements grouped in curly braces
partial: 1
normalize-whitespace: 1
{
:default ::= action => ::array
quartet ::= a a a a
}
inaccessible is warn by default
a ~ 'a'
=for Marpa::R3::Display::End
Statements can be grouped, using curly braces.
These do B<not> create scopes -- the curly braces
serve merely to group and to separate groups of
statements.
=head2 Other ways to disambiguate
There are many other ways to disambiguate SLIF statements.
If the ambiguity is between keywords and symbol names,
enclosing a symbol name in angle brackets will force it to be
treated only as a symbol name.
And while it is never necessary, statements can be re-ordered.
=head1 Adverbs
Adverbs consist of a keyword, the adverb operator ("C<< => >>"),
and the adverb's value.
The keyword must be one of those described in this section.
The adverb's value must be as described for each keyword.
=head2 action
The C<action> adverb is allowed for
=over 4
=item *
An RHS alternative, in which the action is for the alternative.
=item *
The default pseudo-rule, in which case the action is for all rules which do not
have their own action explicitly specified.
=item *
The lexeme default statement, in which case the action is for all lexemes.
=back
The C<action> adverb is not allowed for L0 rules.
The possible values of actions are described,
along with other details of the semantics, in
L<a separate document|Marpa::R3::Semantics>.
=head2 assoc
The C<assoc> adverb is only valid in a prioritized rule.
Its value must be one of
C<left>,
C<right> or
C<group>.
Its effect will be as described
L<below|/"Precedence">.
=head2 bless
The C<bless> adverb causes the result of the semantics to be
blessed into the class indicated by the value of the adverb.
Details of its use may be found in
L<the semantics
document|Marpa::R3::Semantics/"Blessings">.
=head2 event
=for Marpa::R3::Display
name: SLIF named lexeme event synopsis
normalize-whitespace: 1
:lexeme ~ <a> pause => before event => 'before a'
:lexeme ~ <b> pause => after event => 'after b'=on
:lexeme ~ <c> pause => before event => 'before c'=off
:lexeme ~ <d> pause => after event => 'after d'
=for Marpa::R3::Display::End
The C<event> adverb applies only to lexemes and
is only allowed in
L<a C<:lexeme> pseudo-rule|/"Lexeme pseudo-rule">.
It names the event specified by
the L<C<pause>|/"pause"> adverb.
The two adverbs must occur together:
It is a fatal error to specify the C<event> adverb
if the L<C<pause>|/"pause"> adverb is not also specified.
And
It is a fatal error to specify the C<pause> adverb
if the L<C<event>|/"event"> adverb is not also specified.
The value of the C<event> adverb is an event
initializer.
Event initializers are as described
L<above|/"Event initializer">.
SLIF parse events are described in detail in
L<a separate document|Marpa::R3::Event>.
=head2 name
=for Marpa::R3::Display
name: SLIF symbol, name array descriptor example
partial: 1
normalize-whitespace: 1
start ::= number1 number2 name => top
number1 ::= <forty two> name => 'number 1'
number2 ::= <forty three> name => 'number 2'
=for Marpa::R3::Display::End
The C<name> adverb applies only to rules and
rule alternatives.
When specified, it defines a name for that rule
alternative.
=head2 null-ranking
=for Marpa::R3::Display
name: null-ranking adverb example
partial: 1
normalize-whitespace: 1
S ::= A A A A null-ranking => high
=for Marpa::R3::Display::End
The C<null-ranking> adverb applies only to G1 rules
(L0 rules do not have a semantics) and is ignored unless
the SLIF recognizer's C<ranking_method> named argument
is set to something other than its default.
Some rule alternatives can match the same input in several ways,
depending on which symbols are nulled.
These different ways of nulling symbols in a rule
are called its null variants.
The C<null-ranking> named argument allows the application
to control the order in which null variants
are returned by the C<value()> method.
If C<null-ranking> is undefined,
the order of the null variants will be arbitrary.
This is the default,
and is acceptable to most applications.
For details on using the
C<null-ranking> adverb,
see L<the document on parse order|Marpa::R3::Semantics::Order>.
=head2 pause
=for Marpa::R3::Display
name: SLIF named lexeme event synopsis
normalize-whitespace: 1
:lexeme ~ <a> pause => before event => 'before a'
:lexeme ~ <b> pause => after event => 'after b'=on
:lexeme ~ <c> pause => before event => 'before c'=off
:lexeme ~ <d> pause => after event => 'after d'
=for Marpa::R3::Display::End
The C<pause> adverb applies only to lexemes and
is only allowed in
L<a C<:lexeme> pseudo-rule|/"Lexeme pseudo-rule">.
The C<pause> adverb declares a SLIF parse event.
The L<event adverb|/"event"> names the SLIF parse event
declared by
the C<pause> adverb.
It is a fatal error to specify the
C<pause> adverb if the C<event> adverb is
not also specified.
=head2 priority
The C<priority> adverb is only allowed in
L<a C<:lexeme>
pseudo-rule|/"Lexeme pseudo-rule">.
It sets the lexeme priority for the lexeme.
The priority must be an integer,
but it may be negative.
An increase in numerical value means a higher priority.
For example,
a priority of 1 is greater than a priority of 0.
A priority of 0, in turn, is greater than a priority of -1.
The default priority is zero.
Where more than one lexeme can be accepted at
a location, the lexeme priority limits the lexemes
that will be considered.
Only lexemes with the highest priority are considered.
If several lexemes have the same priority,
all of them will be accepted.
The only effect of the lexeme priority
is on the choice
of lexemes when
=over 4
=item * all of them would be accepted;
=item * all started at the same string location;
=item * all end at the same string location; and therefore
=item * all have the same length.
=back
Lexeme priorities only have
an effect when lexemes are accepted.
The intent of this scheme is to avoid situations where a lexeme
with a high priority is rejected, and causes a parse to fail,
even though another lower priority lexeme is acceptable
and would allow the parse to continue.
For example, suppose that "C<say>" can be both
a keyword (C<< <say keyword> >>),
and a variable name (C<< <variable> >>).
Suppose further that
the grammar specifies that C<< <say keyword> >> has a priority of 1,
and C<< <variable> >> is left at the default priority of 0.
When L0 finds a occurrence of "C<say>",
where both the C<say> keyword and a variable name would
be accepted by G1, then only the C<say> keyword is read by G1,
because of the priorities.
But, suppose instead that the parse is at a location
where G1 is not accepting
the C<< <say keyword> >>.
Since only lexeme priorites of acceptable lexemes are considered,
C<< <variable> >> lexeme
has the highest priority,
and the literal string "C<say>" will be read as a
C<< <variable> >>
token.
=head2 proper
The C<proper> keyword is only valid for a quantified right side,
and its value must be a boolean,
in the form of a binary digit (C<0> or C<1>).
It is only relevant if a separator
is defined and is 1 if proper separation
is required, and 0 if Perl separation is allowed.
"Perl separation" allows a final separator.
"Proper separation" is so called, because it requires
that separators be "proper" in the sense that they
must actually separate
sequence items.
=head2 rank
=for Marpa::R3::Display
name: rank adverb example
partial: 1
normalize-whitespace: 1
unspecial ::= ('I' 'am' 'special') words ('--' 'NOT!' ';') rank => 1
special ::= words (';') rank => -1
=for Marpa::R3::Display::End
C<rank> is ignored unless
the recognizer's C<ranking_method> named argument
is set to something other than its default.
The range allowed for C<rank> is implementation-defined,
but numbers in the range
between -134,217,727 and 134,217,727
will always be allowed.
C<rank> is 0 by default.
For details on using the
C<rank> named argument,
see L<the document on parse order|Marpa::R3::Semantics::Order>.
=head2 separator
The C<separator> keyword is only valid for a quantified right side,
and its value must be a single symbol --
either a single symbol name,
or a character class.
If specified, the separator must separate items of the sequence.
A separator may not be nullable.
=head1 Precedence
Marpa's precedence is a generalization beyond the traditional ideas of precedence.
Traditional precedence parsing required
the classification of operators as postfix, infix, etc.
Marpa's precedence parsing is NOT based on
the special treatment of operators.
For the purpose of precedence,
an operand is an occurrence in a RHS alternative
of the LHS symbol.
An operator is considered to be anything
that is not an operand.
The arity of an alternative is the number of operands that
it contains.
All arities are allowed, from zero to the arbitrary
number imposed by system limits such as memory and file size.
For example, in the synopsis, the LHS symbol is
C<Expression>. The alternative
=for Marpa::R3::Display
name: SLIF DSL Synopsis
partial: 1
normalize-whitespace: 1
'(' Expression ')'
=for Marpa::R3::Display::End
contains one occurrence of C<Expression> and therefore has an arity of one.
The C<< <op lparen> >> and C<< <op rparen> >> are considered operators.
In the RHS alternative
=for Marpa::R3::Display
name: SLIF DSL Synopsis
partial: 1
normalize-whitespace: 1
Expression '**' Expression
=for Marpa::R3::Display::End
C<Expression> occurs twice, and therefore the arity is 2.
C<< <op pow> >> is considered to be an operator.
Because for this purpose an operator is defined as anything that
is not an operand, Marpa treats some symbols as operators
that would not be considered operators in the traditional approach.
For example,
in the RHS alternative
=for Marpa::R3::Display
name: SLIF DSL Synopsis
partial: 1
normalize-whitespace: 1
Number
=for Marpa::R3::Display::End
there are no occurrences of C<Expression>, so that the alternative
has an arity of zero -- it is nullary.
The symbol C<Number> is considered to be an operator.
An alternative with arity 0 is nullary.
Precedence and associativity
are meaningless in this
case and will be ignored.
An alternative with arity 1 is unary.
Precedence will have effect,
but left and right associativity will not.
An alternative with arity 2 is binary.
Precedence will have effect,
and left and right associativity will behave
in the traditional way.
The traditional behavior for binary alternatives
is exactly as described next for the I<N>-ary case.
An alternative with an arity of I<N>,
where I<N> is 2 or greater, is I<N>-ary.
Precedence will have effect.
For left associativity,
only the leftmost operand of an I<N>-ary alternative
associates -- operands
after the first will have the
next-tightest priority level.
For right associativity,
only the rightmost operand of an I<N>-ary alternative
associates -- all operands
except the last
will have the next-tightest priority level.
Marpa also allows "group" associativity.
In "group" associativity, all operands associate
at the loosest (lowest) priority.
That is, in an alternative with group associativity,
each operand may be a full expression of the kind
defined by the prioritized rule.
"Group" associativity is used, for example,
in implementing the traditional function of parentheses
in Marpa.
Group associativity is meaningless for nullary alternatives,
and is ignored.
=head2 Precedence and ambiguous grammars
Marpa's generalization of precedence works
for all grammars that
can be defined by prioritized rules.
It is efficient (linear) for all grammars that could be
parsed by the traditional precedence parsing methods.
Marpa also allows you to define alternatives
not allowed by traditional methods.
Many of these are useful,
and most of the useful ones can be parsed efficiently.
Because of the many forms of recursion allowed,
it is possible
to define highly ambiguous grammars using the precedence mechanism.
This can occur even by accident.
The user should especially be careful with
right hand side alternatives in
which all the symbols are operands.
These can be useful.
For example, an implicit operation can be defined using
a binary alternative with no non-operands,
and this could implement, for example,
the standard notation for concatenation or multiplication.
But to do this efficiently
requires either avoiding ambiguity,
or controlling its use carefully.
Marpa does catch the case where an alternative consists
only of a single operand -- a "unit rule".
This causes a fatal error.
Unit rules are easy to define by accident in
the SLIF.
The author knows of no practical use for them,
and their presence in a grammar is usually
unintentional.
Note that, in the event an application
does find a use for a grammar with unit rules,
the NAIF and the Thin interface
can parse it.
=head1 COPYRIGHT AND LICENSE
=for Marpa::R3::Display
ignore: 1
Marpa::R3 is Copyright (C) 2016, Jeffrey Kegler.
This module is free software; you can redistribute it and/or modify it
under the same terms as Perl 5.10.1. For more details, see the full text
of the licenses in the directory LICENSES.
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.
=for Marpa::R3::Display::End
=cut
# vim: expandtab shiftwidth=4: