peg 0.0.8 peg: ^0.0.8 copied to clipboard
PEG (Parsing expression grammar) parsers generator.
#peg #
PEG (Parsing expression grammar) parsers generator.
Version: 0.0.8
Status: Experimental
Latest version always can be found at https://github.com/mezoni/peg
Features:
- Generation of detailed comments
- Generated parsers has no dependencies
- Grammar analytics
- Grammar reporting
- Grammar statistics
- High quality generated source code
- Lookahead mapping tables
- Memoization
- Possibility to trace parsing
- Powerful error and mistakes detection
- Printing grammar
- Terminal and nonterminal symbol recognition
In the future, will be added generator of an alternative PEG (a state machine based) parser.
Error detection
- Infinite loops
- Left recursive rules
- Optional expression in choices
Trace
Trace information are useful for diagnose the problems.
Trace displayed in the following format:
column, line:state rule padding code position
Eg:
94, 8: F* OPEN '-' Char { $$ = [$1, $3]; (2343)
94, 8: > Literal '-' Char { $$ = [$1, $3]; (2343)
State:
Cache : Match : Direction
Cache:
- 'C' - Cache
- ' ' - Not cache
Match:
- 'F' - Failed
- ' ' - Succeed
Direction:
- '>' - Enter
- '<' - Leave
- 'S' - Skip (lookahead)
Examples:
- ' >' Enter
- ' <' Leave, success
- ' F<' Leave, failed
- 'C <' Leave, succeed, uses cached result
- 'CF<' Leave, failed, uses cached result
- ' S' Skip (lookahead), succeed
- ' FS' Skip (lookahead), failed
Grammar
Grammar <- SPACING Globals? Members? Definition+ EOF
Globals <- "%{" GlobalsBody* "}%" SPACING
GlobalsBody <- !"}%" .
Members <- "{" ActionBody* "}" SPACING
Action <- "{" ActionBody* "}" SPACING
ActionBody <- Action / !"}" .
Definition <- IDENTIFIER LEFTARROW Expression
Expression <- Sequence (SLASH Sequence)*
Sequence <- Prefix+
Prefix <- (AND / NOT)? Suffix Action?
Suffix <- Primary (QUESTION / STAR / PLUS)?
Primary <- IDENTIFIER !LEFTARROW / OPEN Expression CLOSE / Literal / Class / DOT
Literal <- "\'" (!"\'" Char)* "\'" SPACING / "\"" (!"\"" Char)* "\"" SPACING
Class <- "[" (!"]" Range)* "]" SPACING
Range <- Char "-" Char / Char
Char <- "\\" ["'\-\[-\]nrt] / HEX_NUMBER / !"\\" !EOL .
EOF <- !.
IDENTIFIER <- IDENT_START IDENT_CONT* SPACING
IDENT_START <- [A-Z_a-z]
IDENT_CONT <- IDENT_START / [0-9]
LEFTARROW <- "<-" SPACING
SLASH <- "/" SPACING
AND <- "&" SPACING
NOT <- "!" SPACING
QUESTION <- "?" SPACING
STAR <- "*" SPACING
PLUS <- "+" SPACING
OPEN <- "(" SPACING
CLOSE <- ")" SPACING
DOT <- "." SPACING
HEX_NUMBER <- "\\u" [0-9A-Fa-f]+
SPACING <- (SPACE / COMMENT)*
COMMENT <- "#" (!EOL .)* EOL?
SPACE <- [\t ] / EOL
EOL <- "\r\n" / [\n\r]
Example
Arithmetic grammar
%{
part of peg.example.arithmetic;
num _binop(num left, num right, String op) {
switch(op) {
case "+":
return left + right;
case "-":
return left - right;
case "*":
return left * right;
case "/":
return left / right;
default:
throw "Unsupported operation $op";
}
}
}%
Expr <-
Sentence EOF { $$ = $1; }
Sentence <-
SPACES Term (PLUS / MINUS) Sentence { $$ = _binop($2, $4, $3); }
/ Term
Term <-
Atom (MUL / DIV) Term { $$ = _binop($1, $3, $2); }
/ Atom
Atom <-
NUMBER
/ OPEN Sentence CLOSE { $$ = $2; }
# Tokens
CLOSE <-
')' SPACES
DIV <-
'/' SPACES { $$ = $1; }
EOF <-
!.
MINUS <-
'-' SPACES { $$ = $1; }
MUL <-
'*' SPACES { $$ = $1; }
NUMBER <-
[0-9]+ SPACES { $$ = int.parse($1.join()); }
OPEN <-
'(' SPACES
PLUS <-
'+' SPACES { $$ = $1; }
SPACES <-
WS*
WS <-
[ \n\t\r] / '\r\n'
Source code of the generated parser for arithmetic grammar
Such (as an arithmetic) grammars parsed faster with a memoization.
peg general --comment --lookahead --memoize arithmetic.peg
// This code was generated by a tool.
// Processing tool available at https://github.com/mezoni/peg
part of peg.example.arithmetic;
num _binop(num left, num right, String op) {
switch(op) {
case "+":
return left + right;
case "-":
return left - right;
case "*":
return left * right;
case "/":
return left / right;
default:
throw "Unsupported operation $op";
}
}
class ArithmeticParser {
static const int EOF = -1;
static final List<bool> _lookahead = _unmap([0x800013, 0x3ff01]);
// '\t', '\n', '\r', ' '
static final List<bool> _mapping0 = _unmap([0x800013]);
bool success;
List _cache;
int _cachePos;
List<int> _cacheRule;
List<int> _cacheState;
int _ch;
int _column;
int _cursor;
List<String> _expected;
int _failurePos;
int _flag;
int _inputLen;
int _line;
int _testing;
String _input;
ArithmeticParser(String text) {
if (text == null) {
throw new ArgumentError('text: $text');
}
_input = text;
_inputLen = _input.length;
if (_inputLen >= 0x3fffffe8 / 32) {
throw new StateError('File size to big: $_inputLen');
}
reset(0);
}
int get column {
if (_column == -1) {
_calculatePos(_failurePos);
}
return _column;
}
int get line {
if (_line == -1) {
_calculatePos(_failurePos);
}
return _line;
}
dynamic parse_Atom() {
// NONTERMINAL
// Atom <- NUMBER / OPEN Sentence CLOSE
var $$;
while (true) {
// NUMBER
$$ = null;
success = _ch >= 48 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (NUMBER)
if (success) $$ = parse_NUMBER();
if (!success) {
if (_cursor > _testing) _failure(const ["NUMBER"]);
}
if (success) break;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// OPEN
$$ = null;
success = _ch == 40; // '('
// Lookahead (OPEN)
if (success) $$ = parse_OPEN();
if (!success) {
if (_cursor > _testing) _failure(const ["("]);
break;
}
var seq = new List(3)..[0] = $$;
// Sentence
$$ = null;
success = _ch >= 9 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Sentence)
if (success) $$ = parse_Sentence();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
seq[1] = $$;
// CLOSE
$$ = null;
success = _ch == 41; // ')'
// Lookahead (CLOSE)
if (success) $$ = parse_CLOSE();
if (!success) {
if (_cursor > _testing) _failure(const [")"]);
break;
}
seq[2] = $$;
$$ = seq;
if (success) {
// OPEN
final $1 = seq[0];
// Sentence
final $2 = seq[1];
// CLOSE
final $3 = seq[2];
$$ = $2;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
break;
}
return $$;
}
dynamic parse_CLOSE() {
// TERMINAL
// CLOSE <- ")" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// ")"
$$ = _matchString(')', const [")"]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_DIV() {
// TERMINAL
// DIV <- "/" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// "/"
$$ = _matchString('/', const ["/"]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
if (success) {
// "/"
final $1 = seq[0];
// SPACES
final $2 = seq[1];
$$ = $1;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_EOF() {
// TERMINAL
// EOF <- !.
var $$;
// !.
var ch0 = _ch;
var pos0 = _cursor;
var testing0 = _testing;
_testing = _inputLen + 1;
// .
$$ = _matchAny();
_ch = ch0;
_cursor = pos0;
_testing = testing0;
$$ = null;
success = !success;
if (!success) {
if (_cursor > _testing) _failure();
}
return $$;
}
dynamic parse_Expr() {
// NONTERMINAL
// Expr <- Sentence EOF
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// Sentence
$$ = null;
success = _ch >= 9 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Sentence)
if (success) $$ = parse_Sentence();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
var seq = new List(2)..[0] = $$;
// EOF
$$ = parse_EOF();
if (!success) break;
seq[1] = $$;
$$ = seq;
if (success) {
// Sentence
final $1 = seq[0];
// EOF
final $2 = seq[1];
$$ = $1;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_MINUS() {
// TERMINAL
// MINUS <- "-" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// "-"
$$ = _matchString('-', const ["-"]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
if (success) {
// "-"
final $1 = seq[0];
// SPACES
final $2 = seq[1];
$$ = $1;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_MUL() {
// TERMINAL
// MUL <- "*" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// "*"
$$ = _matchString('*', const ["*"]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
if (success) {
// "*"
final $1 = seq[0];
// SPACES
final $2 = seq[1];
$$ = $1;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_NUMBER() {
// TERMINAL
// NUMBER <- [0-9]+ SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// [0-9]+
var testing0;
for (var first = true, reps; ;) {
// [0-9]
$$ = _matchRange(48, 57);
if (success) {
if (first) {
first = false;
reps = [$$];
testing0 = _testing;
} else {
reps.add($$);
}
_testing = _cursor;
} else {
success = !first;
if (success) {
_testing = testing0;
$$ = reps;
} else $$ = null;
break;
}
}
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
if (success) {
// [0-9]+
final $1 = seq[0];
// SPACES
final $2 = seq[1];
$$ = int.parse($1.join());
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_OPEN() {
// TERMINAL
// OPEN <- "(" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// "("
$$ = _matchString('(', const ["("]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_PLUS() {
// TERMINAL
// PLUS <- "+" SPACES
var $$;
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// "+"
$$ = _matchString('+', const ["+"]);
if (!success) break;
var seq = new List(2)..[0] = $$;
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
seq[1] = $$;
$$ = seq;
if (success) {
// "+"
final $1 = seq[0];
// SPACES
final $2 = seq[1];
$$ = $1;
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
return $$;
}
dynamic parse_SPACES() {
// TERMINAL
// SPACES <- WS*
var $$;
var pos = _cursor;
if(pos <= _cachePos) {
$$ = _getFromCache(12);
}
if($$ != null) {
return $$[0];
}
// WS*
var testing0 = _testing;
for (var reps = []; ; ) {
_testing = _cursor;
// WS
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (WS)
if (success) $$ = parse_WS();
if (!success) {
if (_cursor > _testing) _failure(const ["WS"]);
}
if (success) {
reps.add($$);
} else {
success = true;
_testing = testing0;
$$ = reps;
break;
}
}
_addToCache($$, pos, 12);
return $$;
}
dynamic parse_Sentence() {
// NONTERMINAL
// Sentence <- SPACES Term (PLUS / MINUS) Sentence / Term
var $$;
var pos = _cursor;
if(pos <= _cachePos) {
$$ = _getFromCache(1);
}
if($$ != null) {
return $$[0];
}
while (true) {
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// SPACES
$$ = null;
success = _ch >= 9 && _ch <= 32 && _lookahead[_ch + -9];
// Lookahead (SPACES is optional)
if (success) $$ = parse_SPACES();
else success = true;
var seq = new List(4)..[0] = $$;
// Term
$$ = null;
success = _ch >= 40 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Term)
if (success) $$ = parse_Term();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
seq[1] = $$;
while (true) {
// PLUS
$$ = null;
success = _ch == 43; // '+'
// Lookahead (PLUS)
if (success) $$ = parse_PLUS();
if (!success) {
if (_cursor > _testing) _failure(const ["+"]);
}
if (success) break;
// MINUS
$$ = null;
success = _ch == 45; // '-'
// Lookahead (MINUS)
if (success) $$ = parse_MINUS();
if (!success) {
if (_cursor > _testing) _failure(const ["-"]);
}
break;
}
if (!success) break;
seq[2] = $$;
// Sentence
$$ = null;
success = _ch >= 9 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Sentence)
if (success) $$ = parse_Sentence();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
seq[3] = $$;
$$ = seq;
if (success) {
// SPACES
final $1 = seq[0];
// Term
final $2 = seq[1];
// PLUS / MINUS
final $3 = seq[2];
// Sentence
final $4 = seq[3];
$$ = _binop($2, $4, $3);
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
if (success) break;
// Term
$$ = null;
success = _ch >= 40 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Term)
if (success) $$ = parse_Term();
if (!success) {
if (_cursor > _testing) _failure(null);
}
break;
}
_addToCache($$, pos, 1);
return $$;
}
dynamic parse_Term() {
// NONTERMINAL
// Term <- Atom (MUL / DIV) Term / Atom
var $$;
var pos = _cursor;
if(pos <= _cachePos) {
$$ = _getFromCache(2);
}
if($$ != null) {
return $$[0];
}
while (true) {
var ch0 = _ch;
var pos0 = _cursor;
while (true) {
// Atom
$$ = null;
success = _ch >= 40 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Atom)
if (success) $$ = parse_Atom();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
var seq = new List(3)..[0] = $$;
while (true) {
// MUL
$$ = null;
success = _ch == 42; // '*'
// Lookahead (MUL)
if (success) $$ = parse_MUL();
if (!success) {
if (_cursor > _testing) _failure(const ["*"]);
}
if (success) break;
// DIV
$$ = null;
success = _ch == 47; // '/'
// Lookahead (DIV)
if (success) $$ = parse_DIV();
if (!success) {
if (_cursor > _testing) _failure(const ["/"]);
}
break;
}
if (!success) break;
seq[1] = $$;
// Term
$$ = null;
success = _ch >= 40 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Term)
if (success) $$ = parse_Term();
if (!success) {
if (_cursor > _testing) _failure(null);
break;
}
seq[2] = $$;
$$ = seq;
if (success) {
// Atom
final $1 = seq[0];
// MUL / DIV
final $2 = seq[1];
// Term
final $3 = seq[2];
$$ = _binop($1, $3, $2);
}
break;
}
if (!success) {
_ch = ch0;
_cursor = pos0;
}
if (success) break;
// Atom
$$ = null;
success = _ch >= 40 && _ch <= 57 && _lookahead[_ch + -9];
// Lookahead (Atom)
if (success) $$ = parse_Atom();
if (!success) {
if (_cursor > _testing) _failure(null);
}
break;
}
_addToCache($$, pos, 2);
return $$;
}
dynamic parse_WS() {
// TERMINAL
// WS <- [\t-\n\r ] / "\r\n"
var $$;
while (true) {
// [\t-\n\r ]
$$ = _matchMapping(9, 32, _mapping0);
if (success) break;
// "\r\n"
$$ = _matchString('\r\n', const ["\\r\\n"]);
break;
}
return $$;
}
void _addToCache(dynamic result, int start, int id) {
var cached = _cache[start];
if (cached == null) {
_cacheRule[start] = id;
_cache[start] = [result, _cursor, success];
} else {
var slot = start >> 5;
var r1 = (slot << 5) & 0x3fffffff;
var mask = 1 << (start - r1);
if ((_cacheState[slot] & mask) == 0) {
_cacheState[slot] |= mask;
cached = [new List.filled(2, 0), new Map<int, List>()];
_cache[start] = cached;
}
slot = id >> 5;
r1 = (slot << 5) & 0x3fffffff;
mask = 1 << (id - r1);
cached[0][slot] |= mask;
cached[1][id] = [result, _cursor, success];
}
if (_cachePos < start) {
_cachePos = start;
}
}
void _calculatePos(int pos) {
if (pos == null || pos < 0 || pos > _inputLen) {
return;
}
_line = 1;
_column = 1;
for (var i = 0; i < _inputLen && i < pos; i++) {
var c = _input.codeUnitAt(i);
if (c == 13) {
_line++;
_column = 1;
if (i + 1 < _inputLen && _input.codeUnitAt(i + 1) == 10) {
i++;
}
} else if (c == 10) {
_line++;
_column = 1;
} else {
_column++;
}
}
}
void _failure([List<String> expected]) {
if (_failurePos > _cursor) {
return;
}
if (_cursor > _failurePos) {
_expected = [];
_failurePos = _cursor;
}
if (expected != null) {
_expected.addAll(expected);
}
}
List _flatten(dynamic value) {
if (value is List) {
var result = [];
var length = value.length;
for (var i = 0; i < length; i++) {
var element = value[i];
if (element is Iterable) {
result.addAll(_flatten(element));
} else {
result.add(element);
}
}
return result;
} else if (value is Iterable) {
var result = [];
for (var element in value) {
if (element is! List) {
result.add(element);
} else {
result.addAll(_flatten(element));
}
}
}
return [value];
}
dynamic _getFromCache(int id) {
var result = _cache[_cursor];
if (result == null) {
return null;
}
var slot = _cursor >> 5;
var r1 = (slot << 5) & 0x3fffffff;
var mask = 1 << (_cursor - r1);
if ((_cacheState[slot] & mask) == 0) {
if (_cacheRule[_cursor] == id) {
_cursor = result[1];
success = result[2];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return result;
} else {
return null;
}
}
slot = id >> 5;
r1 = (slot << 5) & 0x3fffffff;
mask = 1 << (id - r1);
if ((result[0][slot] & mask) == 0) {
return null;
}
var data = result[1][id];
_cursor = data[1];
success = data[2];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return data;
}
String _matchAny() {
success = _cursor < _inputLen;
if (success) {
var result = _input[_cursor++];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return result;
}
if (_cursor > _testing) {
_failure();
}
return null;
}
String _matchChar(int ch, List<String> expected) {
success = _ch == ch;
if (success) {
var result = _input[_cursor++];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return result;
}
if (_cursor > _testing) {
_failure(expected);
}
return null;
}
String _matchMapping(int start, int end, List<bool> mapping) {
success = _ch >= start && _ch <= end;
if (success) {
if(mapping[_ch - start]) {
var result = _input[_cursor++];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return result;
}
success = false;
}
if (_cursor > _testing) {
_failure();
}
return null;
}
String _matchRange(int start, int end) {
success = _ch >= start && _ch <= end;
if (success) {
var result = _input[_cursor++];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return result;
}
if (_cursor > _testing) {
_failure();
}
return null;
}
String _matchRanges(List<int> ranges) {
var length = ranges.length;
for (var i = 0; i < length; i += 2) {
if (_ch <= ranges[i + 1]) {
if (_ch >= ranges[i]) {
var result = _input[_cursor++];
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
success = true;
return result;
}
} else break;
}
if (_cursor > _testing) {
_failure();
}
success = false;
return null;
}
String _matchString(String string, List<String> expected) {
success = _input.startsWith(string, _cursor);
if (success) {
_cursor += string.length;
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
return string;
}
if (_cursor > _testing) {
_failure(expected);
}
return null;
}
void _nextChar([int count = 1]) {
success = true;
_cursor += count;
if (_cursor < _inputLen) {
_ch = _input.codeUnitAt(_cursor);
} else {
_ch = EOF;
}
}
bool _testChar(int c, int flag) {
if (c < 0 || c > 127) {
return false;
}
int slot = (c & 0xff) >> 6;
int mask = 1 << c - ((slot << 6) & 0x3fffffff);
if ((flag & mask) != 0) {
return true;
}
return false;
}
bool _testInput(int flag) {
if (_cursor >= _inputLen) {
return false;
}
var c = _input.codeUnitAt(_cursor);
if (c < 0 || c > 127) {
return false;
}
int slot = (c & 0xff) >> 6;
int mask = 1 << c - ((slot << 6) & 0x3fffffff);
if ((flag & mask) != 0) {
return true;
}
return false;
}
static List<bool> _unmap(List<int> mapping) {
var length = mapping.length;
var result = new List<bool>(length * 31);
var offset = 0;
for (var i = 0; i < length; i++) {
var v = mapping[i];
for (var j = 0; j < 31; j++) {
result[offset++] = v & (1 << j) == 0 ? false : true;
}
}
return result;
}
List<String> get expected {
var set = new Set<String>();
set.addAll(_expected);
if (set.contains(null)) {
set.clear();
}
var result = set.toList();
result.sort();
return result;
}
void reset(int pos) {
if (pos == null) {
throw new ArgumentError('pos: $pos');
}
if (pos < 0 || pos > _inputLen) {
throw new RangeError('pos');
}
success = true;
_cache = new List(_inputLen + 1);
_cachePos = -1;
_cacheRule = new List(_inputLen + 1);
_cacheState = new List.filled(((_inputLen + 1) >> 5) + 1, 0);
_ch = EOF;
_column = -1;
_expected = [];
_failurePos = -1;
_flag = 0;
_cursor = pos;
_line = -1;
_testing = -1;
if (pos < _inputLen) {
_ch = _input.codeUnitAt(pos);
}
}
String get unexpected {
if (_failurePos < 0 || _failurePos >= _inputLen) {
return '';
}
return _input[_failurePos];
}
}