Files
meteor/packages/jsparse/parser.js
2012-11-26 18:28:07 -08:00

843 lines
29 KiB
JavaScript

///// JAVASCRIPT PARSER
// What we don't support from ECMA-262 5.1:
// - object literal trailing comma
// - object literal get/set
(function () {
var expecting = Parser.expecting;
var assertion = Parsers.assertion;
var node = Parsers.node;
var or = Parsers.or;
var and = Parsers.and;
var not = Parsers.not;
var list = Parsers.list;
var seq = Parsers.seq;
var opt = Parsers.opt;
var constant = Parsers.constant;
var mapResult = Parsers.mapResult;
var makeSet = function (array) {
var s = {};
for (var i = 0, N = array.length; i < N; i++)
s[array[i]] = true;
return s;
};
JSParser = function (code, options) {
this.lexer = new JSLexer(code);
this.oldToken = null;
this.newToken = null;
this.pos = 0;
this.isLineTerminatorHere = false;
this.includeComments = false;
// the last COMMENT lexeme between oldToken and newToken
// that we've consumed, if any.
this.lastCommentConsumed = null;
options = options || {};
// pass {tokens:'strings'} to get strings for
// tokens instead of token objects
if (options.tokens === 'strings') {
this.tokenFunc = function (tok) {
return tok.text();
};
} else {
this.tokenFunc = function (tok) {
return tok;
};
}
// pass {includeComments: true} to include comments in the AST. For
// a comment to be included, it must occur where a series of
// statements could occur, and it must be preceded by only comments
// and whitespace on the same line.
if (options.includeComments) {
this.includeComments = true;
}
};
JSParser.prototype.consumeNewToken = function () {
var self = this;
var lexer = self.lexer;
self.oldToken = self.newToken;
self.isLineTerminatorHere = false;
var lex;
do {
lex = lexer.next();
if (lex.isError())
throw new Error("Bad token at " +
JSLexer.prettyOffset(lexer.code, lex.startPos()) +
", text `" + lex.text() + "`");
else if (lex.type() === "NEWLINE")
self.isLineTerminatorHere = true;
else if (lex.type() === "COMMENT" && ! /^.*$/.test(lex.text()))
// multiline comments containing line terminators count
// as line terminators.
self.isLineTerminatorHere = true;
} while (! lex.isEOF() && ! lex.isToken());
self.newToken = lex;
self.pos = lex.startPos();
self.lastCommentConsumed = null;
};
JSParser.prototype.getParseError = function (expecting, found) {
var msg = (expecting ? "Expected " + expecting : "Unexpected token");
if (this.oldToken)
msg += " after " + this.oldToken;
var pos = this.pos;
msg += " at " + JSLexer.prettyOffset(this.lexer.code, pos);
msg += ", found " + (found || this.newToken);
return new Error(msg);
};
JSParser.prototype.getSyntaxTree = function () {
var self = this;
self.consumeNewToken();
var NIL = new ParseNode('nil', []);
var booleanFlaggedParser = function (parserConstructFunc) {
return {
'false': parserConstructFunc(false),
'true': parserConstructFunc(true)
};
};
// Takes a space-separated list of either punctuation or keyword tokens
var lookAheadToken = function (tokens) {
var type = (/\w/.test(tokens) ? 'KEYWORD' : 'PUNCTUATION');
var textSet = makeSet(tokens.split(' '));
return expecting(
tokens.split(' ').join(', '),
assertion(function (t) {
return (t.newToken.type() === type && textSet[t.newToken.text()]);
}));
};
var lookAheadTokenType = function (type) {
return expecting(type, assertion(function (t) {
return t.newToken.type() === type;
}));
};
// Takes a space-separated list of either punctuation or keyword tokens
var token = function (tokens) {
var type = (/\w/.test(tokens) ? 'KEYWORD' : 'PUNCTUATION');
var textSet = makeSet(tokens.split(' '));
return new Parser(
tokens.split(' ').join(', '),
function (t) {
if (t.newToken.type() === type && textSet[t.newToken.text()]) {
t.consumeNewToken();
return self.tokenFunc(t.oldToken);
}
return null;
});
};
var tokenType = function (type) {
return new Parser(type, function (t) {
if (t.newToken.type() === type) {
t.consumeNewToken();
return self.tokenFunc(t.oldToken);
}
return null;
});
};
var noLineTerminatorHere = expecting(
'noLineTerminator', assertion(function (t) {
return ! t.isLineTerminatorHere;
}));
var nonLHSExpressionNames = makeSet(
'unary binary postfix ternary assignment comma'.split(' '));
var isExpressionLHS = function (exprNode) {
return ! nonLHSExpressionNames[exprNode.name];
};
// Like token, but marks tokens that need to defy the lexer's
// heuristic about whether the next '/' is a division or
// starts a regex.
var preSlashToken = function (text, divisionNotRegex) {
var inner = token(text);
return new Parser(
inner.expecting,
function (t) {
// temporarily set divisionPermitted,
// restoring it if we don't match.
var oldValue = t.lexer.divisionPermitted;
var result;
try {
t.lexer.divisionPermitted = divisionNotRegex;
result = inner.parse(t);
return result;
} finally {
if (! result)
t.lexer.divisionPermitted = oldValue;
}
});
};
// Mark some productions "lazy" to allow grammar circularity, i.e. accessing
// later parsers from earlier ones.
// These lazy versions will be replaced with real ones, which they will
// access when run.
var expressionMaybeNoIn = {
'false': Parsers.lazy(
'expression',
function () { return expressionMaybeNoIn[false]; }),
'true': Parsers.lazy(
'expression',
function () { return expressionMaybeNoIn[true]; })
};
var expression = expressionMaybeNoIn[false];
var assignmentExpressionMaybeNoIn = {
'false': Parsers.lazy(
'expression',
function () { return assignmentExpressionMaybeNoIn[false]; }),
'true': Parsers.lazy(
'expression',
function () { return assignmentExpressionMaybeNoIn[true]; })
};
var assignmentExpression = assignmentExpressionMaybeNoIn[false];
var functionBody = Parsers.lazy(
'statement', function () { return functionBody; });
var statement = Parsers.lazy(
'statement', function () { return statement; });
////
var arrayLiteral =
node('array',
seq(token('['),
opt(list(token(','))),
or(
lookAheadToken(']'),
list(
expecting(
'expression',
or(assignmentExpression,
// count a peeked-at ']' as an expression
// to support elisions at end, e.g.
// `[1,2,3,,,,,,]`.
lookAheadToken(']'))),
// list seperator is one or more commas
// to support elision
list(token(',')))),
token(']')));
// "IdentifierName" in ES5 allows reserved words, like in a property access
// or a key of an object literal.
// Put IDENTIFIER last so it shows up in the error message.
var identifierName = or(tokenType('NULL'), tokenType('BOOLEAN'),
tokenType('KEYWORD'), tokenType('IDENTIFIER'));
var propertyName = expecting('propertyName', or(
node('idPropName', identifierName),
node('numPropName', tokenType('NUMBER')),
node('strPropName', tokenType('STRING'))));
var nameColonValue = expecting(
'propertyName',
node('prop', seq(propertyName, token(':'), assignmentExpression)));
// Allow trailing comma in object literal, per ES5. Trailing comma
// must follow a `name:value`, that is, `{,}` is invalid.
//
// We can't just use a normal comma list(), because it will seize
// on the comma as a sign that the list continues. Instead,
// we specify a list of either ',' or nameColonValue, using positive
// and negative lookAheads to constrain the sequence. The grammar
// is ordered so that error messages will always say
// "Expected propertyName" or "Expected ," as appropriate, not
// "Expected ," when the look-ahead is negative or "Expected }".
var objectLiteral =
node('object',
seq(token('{'),
or(lookAheadToken('}'),
and(not(lookAheadToken(',')),
list(or(seq(token(','),
expecting('propertyName',
not(lookAheadToken(',')))),
seq(nameColonValue,
or(lookAheadToken('}'),
lookAheadToken(','))))))),
expecting('propertyName', token('}'))));
var functionMaybeNameRequired = booleanFlaggedParser(
function (nameRequired) {
return seq(token('function'),
(nameRequired ? tokenType('IDENTIFIER') :
or(tokenType('IDENTIFIER'),
and(lookAheadToken('('), constant(NIL)))),
token('('),
or(lookAheadToken(')'),
list(tokenType('IDENTIFIER'), token(','))),
token(')'),
token('{'),
functionBody,
token('}'));
});
var functionExpression = node('functionExpr',
functionMaybeNameRequired[false]);
var primaryOrFunctionExpression =
expecting('expression',
or(node('this', token('this')),
node('identifier', tokenType('IDENTIFIER')),
node('number', tokenType('NUMBER')),
node('boolean', tokenType('BOOLEAN')),
node('null', tokenType('NULL')),
node('regex', tokenType('REGEX')),
node('string', tokenType('STRING')),
node('parens',
seq(token('('), expression, token(')'))),
arrayLiteral,
objectLiteral,
functionExpression));
var dotEnding = seq(token('.'), identifierName);
var bracketEnding = seq(token('['), expression, token(']'));
var callArgs = seq(token('('),
or(lookAheadToken(')'),
list(assignmentExpression,
token(','))),
token(')'));
var newKeyword = token('new');
// This is a completely equivalent refactor of the spec's production
// for a LeftHandSideExpression.
//
// An lhsExpression is basically an expression that can serve as
// the left-hand-side of an assignment, though function calls and
// "new" invocation are included because they have the same
// precedence. Actually, the spec technically allows a function
// call to "return" a valid l-value, as in `foo(bar) = baz`,
// though no built-in or user-specifiable call has this property
// (it would have to be defined by a browser or other "host").
var lhsExpression = new Parser(
'expression',
function (t) {
// Accumulate all initial "new" keywords, not yet knowing
// if they have a corresponding argument list later.
var news = [];
var n;
while ((n = newKeyword.parse(t)))
news.push(n);
// Read the primaryOrFunctionExpression that will be the "core"
// of this lhsExpression. It is preceded by zero or more `new`
// keywords, and followed by any sequence of (...), [...],
// and .foo add-ons.
// if we have 'new' keywords, we are committed and must
// match an expression or error.
var result = primaryOrFunctionExpression.parseRequiredIf(t, news.length);
if (! result)
return null;
// Our plan of attack is to apply each dot, bracket, or call
// as we come across it. Whether a call is a `new` call depends
// on whether there are `new` keywords we haven't used. If so,
// we pop one off the stack.
var done = false;
while (! done) {
var r;
if ((r = dotEnding.parse(t))) {
result = new ParseNode('dot', [result].concat(r));
} else if ((r = bracketEnding.parse(t))) {
result = new ParseNode('bracket', [result].concat(r));
} else if ((r = callArgs.parse(t))) {
if (news.length)
result = new ParseNode('newcall', [news.pop(), result].concat(r));
else
result = new ParseNode('call', [result].concat(r));
} else {
done = true;
}
}
// There may be more `new` keywords than calls, which is how
// paren-less constructions (`new Date`) are parsed. We've
// already handled `new foo().bar()`, now handle `new new foo().bar`.
while (news.length)
result = new ParseNode('new', [news.pop(), result]);
return result;
});
var postfixToken = token('++ --');
var postfixLookahead = lookAheadToken('++ --');
var postfixExpression = expecting(
'expression',
mapResult(seq(lhsExpression,
opt(and(noLineTerminatorHere,
postfixLookahead,
postfixToken))),
function (v) {
if (v.length === 1)
return v[0];
return new ParseNode('postfix', v);
}));
var unaryExpression = Parsers.unary(
'unary', postfixExpression,
or(token('delete void typeof'),
preSlashToken('++ -- + - ~ !', false)));
// The "noIn" business is all to facilitate parsing
// of for-in constructs, though the cases that make
// this required are quite obscure.
// The `for(var x in y)` form is allowed to take
// an initializer for `x` (which is only useful for
// its side effects, or if `y` has no properties).
// So an example might be:
// `for(var x = a().b in c);`
// In this example, `var x = a().b` is parsed without
// the `in`, which would otherwise be part of the
// varDecl, using varDeclNoIn.
// Our binaryExpression is the spec's LogicalORExpression,
// which includes all the higher-precendence operators.
var binaryExpressionMaybeNoIn = booleanFlaggedParser(
function (noIn) {
// high to low precedence
var binaryOps = [token('* / %'),
token('+ -'),
token('<< >> >>>'),
or(token('< > <= >='),
noIn ? token('instanceof') :
token('instanceof in')),
token('== != === !=='),
token('&'),
token('^'),
token('|'),
token('&&'),
token('||')];
return expecting(
'expression',
Parsers.binaryLeft('binary', unaryExpression, binaryOps));
});
var binaryExpression = binaryExpressionMaybeNoIn[false];
var conditionalExpressionMaybeNoIn = booleanFlaggedParser(
function (noIn) {
return expecting(
'expression',
mapResult(
seq(binaryExpressionMaybeNoIn[noIn],
opt(seq(
token('?'),
assignmentExpression, token(':'),
assignmentExpressionMaybeNoIn[noIn]))),
function (v) {
if (v.length === 1)
return v[0];
return new ParseNode('ternary', v);
}));
});
var conditionalExpression = conditionalExpressionMaybeNoIn[false];
var assignOp = token('= *= /= %= += -= <<= >>= >>>= &= ^= |=');
assignmentExpressionMaybeNoIn = booleanFlaggedParser(
function (noIn) {
return new Parser(
'expression',
function (t) {
var r = conditionalExpressionMaybeNoIn[noIn].parse(t);
if (! r)
return null;
// Assignment is right-associative.
// Plan of attack: make a list of all the parts
// [expression, op, expression, op, ... expression]
// and then fold them up at the end.
var parts = [r];
var op;
while (isExpressionLHS(r) &&(op = assignOp.parse(t)))
parts.push(op,
conditionalExpressionMaybeNoIn[noIn].parseRequired(t));
var result = parts.pop();
while (parts.length) {
op = parts.pop();
var lhs = parts.pop();
result = new ParseNode('assignment', [lhs, op, result]);
}
return result;
});
});
assignmentExpression = assignmentExpressionMaybeNoIn[false];
expressionMaybeNoIn = booleanFlaggedParser(
function (noIn) {
return expecting(
'expression',
mapResult(
list(assignmentExpressionMaybeNoIn[noIn], token(',')),
function (v) {
if (v.length === 1)
return v[0];
return new ParseNode('comma', v);
}));
});
expression = expressionMaybeNoIn[false];
// STATEMENTS
var comment = node('comment', new Parser(null, function (t) {
if (! t.includeComments)
return null;
// Match a COMMENT lexeme between oldToken and newToken.
//
// This is an unusual Parser because it doesn't match and consume
// newToken, but instead uses the next()/prev() API on lexemes.
// It assumes it can walk the linked list backwards from newToken
// (though not necessarily forwards).
//
// We start at the last comment we've visited for this
// oldToken/newToken pair, if any, or else oldToken, or else the
// beginning of the token stream. We ignore comments that are
// preceded by any non-comment source code on the same line.
var lexeme = (t.lastCommentConsumed || t.oldToken || null);
if (! lexeme) {
// no oldToken, must be on first token. walk backwards
// to start with first lexeme (which may be a comment
// or whitespace)
lexeme = t.newToken;
while (lexeme.prev())
lexeme = lexeme.prev();
} else {
// start with lexeme after last token or comment consumed
lexeme = lexeme.next();
}
var seenNewline = ((! t.oldToken) || t.lastCommentConsumed || false);
while (lexeme !== t.newToken) {
var type = lexeme.type();
if (type === "NEWLINE") {
seenNewline = true;
} else if (type === "COMMENT") {
t.lastCommentConsumed = lexeme;
if (seenNewline)
return lexeme;
}
lexeme = lexeme.next();
}
return null;
}));
var statements = list(or(comment, statement));
// implements JavaScript's semicolon "insertion" rules
var maybeSemicolon = expecting(
'semicolon',
or(token(';'),
and(
or(
lookAheadToken('}'),
lookAheadTokenType('EOF'),
assertion(function (t) {
return t.isLineTerminatorHere;
})),
constant(new ParseNode(';', [])))));
var expressionStatement = node(
'expressionStmnt',
and(
not(or(lookAheadToken('{'), lookAheadToken('function'))),
seq(expression,
expecting('semicolon',
or(maybeSemicolon,
// allow presence of colon to terminate
// statement legally, for the benefit of
// expressionOrLabelStatement. Basically assume
// an implicit semicolon. This
// is safe because a colon can never legally
// follow a semicolon anyway.
and(lookAheadToken(':'),
constant(new ParseNode(';', []))))))));
// it's hard to parse statement labels, as in
// `foo: x = 1`, because we can't tell from the
// first token whether we are looking at an expression
// statement or a label statement. To work around this,
// expressionOrLabelStatement parses the expression and
// then rewrites the result if it is an identifier
// followed by a colon.
var labelColonAndStatement = seq(token(':'), statement);
var noColon = expecting(
'semicolon', not(lookAheadToken(':')));
var expressionOrLabelStatement = new Parser(
null,
function (t) {
var exprStmnt = expressionStatement.parse(t);
if (! exprStmnt)
return null;
var expr = exprStmnt.children[0];
var maybeSemi = exprStmnt.children[1];
if (expr.name !== 'identifier' ||
! (maybeSemi instanceof ParseNode)) {
// We either have a non-identifier expression or a present
// semicolon. This is not a label.
//
// Fail now if we are looking at a colon, causing an
// error message on input like `1+1:` of the same kind
// you'd get without statement label parsing.
noColon.parseRequired(t);
return exprStmnt;
}
var rest = labelColonAndStatement.parse(t);
if (! rest)
return exprStmnt;
return new ParseNode('labelStmnt',
[expr.children[0]].concat(rest));
});
var emptyStatement = node('emptyStmnt', token(';')); // required semicolon
var blockStatement = expecting('block', node('blockStmnt', seq(
token('{'), or(lookAheadToken('}'), statements),
token('}'))));
var varDeclMaybeNoIn = booleanFlaggedParser(function (noIn) {
return node(
'varDecl',
seq(tokenType('IDENTIFIER'),
opt(seq(token('='),
assignmentExpressionMaybeNoIn[noIn]))));
});
var varDecl = varDeclMaybeNoIn[false];
var variableStatement = node(
'varStmnt',
seq(token('var'), list(varDecl, token(',')),
maybeSemicolon));
// A paren that may be followed by a statement
// beginning with a regex literal.
var closeParenBeforeStatement = preSlashToken(')', false);
var ifStatement = node(
'ifStmnt',
seq(token('if'), token('('), expression,
closeParenBeforeStatement, statement,
opt(seq(token('else'), statement))));
var secondThirdClauses = expecting(
'semicolon',
and(lookAheadToken(';'),
seq(
expecting('semicolon', token(';')),
or(and(lookAheadToken(';'),
constant(NIL)),
expression),
expecting('semicolon', token(';')),
or(and(lookAheadToken(')'),
constant(NIL)),
expression))));
var inExpr = seq(token('in'), expression);
var inExprExpectingSemi = expecting('semicolon',
seq(token('in'), expression));
var forSpec = mapResult(node(
'forSpec',
or(seq(token('var'),
varDeclMaybeNoIn[true],
expecting(
'commaOrIn',
or(inExpr,
seq(
or(
lookAheadToken(';'),
seq(token(','),
list(varDeclMaybeNoIn[true], token(',')))),
secondThirdClauses)))),
// get the case where the first clause is empty out of the way.
// the lookAhead's return value is the empty placeholder for the
// missing expression.
seq(and(lookAheadToken(';'),
constant(NIL)), secondThirdClauses),
// custom parser the non-var case because we have to
// read the first expression before we know if there's
// an "in".
new Parser(
null,
function (t) {
var firstExpr = expressionMaybeNoIn[true].parse(t);
if (! firstExpr)
return null;
var rest = secondThirdClauses.parse(t);
if (! rest) {
// we need a left-hand-side expression for a
// `for (x in y)` loop.
if (! isExpressionLHS(firstExpr))
throw t.getParseError("semicolon");
// if we don't see 'in' at this point, it's probably
// a missing semicolon
rest = inExprExpectingSemi.parseRequired(t);
}
return [firstExpr].concat(rest);
}))),
function (clauses) {
// There are four kinds of for-loop, and we call the
// part between the parens one of forSpec, forVarSpec,
// forInSpec, and forVarInSpec. Having parsed it
// already, we rewrite the node name based on how
// many items came out. forIn and forVarIn always
// have 3 and 4 items respectively. for has 5
// (the optional expressions are present as nils).
// forVar has 6 or more, because `for(var x;;);`
// produces [`var` `x` `;` nil `;` nil].
var numChildren = clauses.children.length;
if (numChildren === 3)
return new ParseNode('forInSpec', clauses.children);
else if (numChildren === 4)
return new ParseNode('forVarInSpec', clauses.children);
else if (numChildren >= 6)
return new ParseNode('forVarSpec', clauses.children);
return clauses;
});
var iterationStatement = or(
node('doStmnt', seq(token('do'), statement, token('while'),
token('('), expression, token(')'),
maybeSemicolon)),
node('whileStmnt', seq(token('while'), token('('), expression,
closeParenBeforeStatement, statement)),
// semicolons must be real, not maybeSemicolons
node('forStmnt', seq(
token('for'), token('('), forSpec, closeParenBeforeStatement,
statement)));
var returnStatement = node(
'returnStmnt',
seq(token('return'), or(
and(noLineTerminatorHere, expression), constant(NIL)),
maybeSemicolon));
var continueStatement = node(
'continueStmnt',
seq(token('continue'), or(
and(noLineTerminatorHere, tokenType('IDENTIFIER')), constant(NIL)),
maybeSemicolon));
var breakStatement = node(
'breakStmnt',
seq(token('break'), or(
and(noLineTerminatorHere, tokenType('IDENTIFIER')), constant(NIL)),
maybeSemicolon));
var throwStatement = node(
'throwStmnt',
seq(token('throw'),
and(or(noLineTerminatorHere,
// If there is a line break here and more tokens after,
// we want to error appropriately. `throw \n e` should
// complain about the "end of line", not the `e`.
and(not(lookAheadTokenType("EOF")),
new Parser(null,
function (t) {
throw t.getParseError('expression', 'end of line');
}))),
expression),
maybeSemicolon));
var withStatement = node(
'withStmnt',
seq(token('with'), token('('), expression, closeParenBeforeStatement,
statement));
var switchCase = node(
'case',
seq(token('case'), expression, token(':'),
or(lookAheadToken('}'),
lookAheadToken('case default'),
statements)));
var switchDefault = node(
'default',
seq(token('default'), token(':'),
or(lookAheadToken('}'),
lookAheadToken('case'),
statements)));
var switchStatement = node(
'switchStmnt',
seq(token('switch'), token('('), expression, token(')'),
token('{'),
or(lookAheadToken('}'),
lookAheadToken('default'),
list(switchCase)),
opt(seq(switchDefault,
opt(list(switchCase)))),
token('}')));
var catchFinally = expecting(
'catch',
and(lookAheadToken('catch finally'),
seq(
or(node(
'catch',
seq(token('catch'), token('('), tokenType('IDENTIFIER'),
token(')'), blockStatement)),
constant(NIL)),
or(node(
'finally',
seq(token('finally'), blockStatement)),
constant(NIL)))));
var tryStatement = node(
'tryStmnt',
seq(token('try'), blockStatement, catchFinally));
var debuggerStatement = node(
'debuggerStmnt', seq(token('debugger'), maybeSemicolon));
statement = expecting('statement',
or(expressionOrLabelStatement,
emptyStatement,
blockStatement,
variableStatement,
ifStatement,
iterationStatement,
returnStatement,
continueStatement,
breakStatement,
withStatement,
switchStatement,
throwStatement,
tryStatement,
debuggerStatement));
// PROGRAM
var functionDecl = node(
'functionDecl', functionMaybeNameRequired[true]);
// Look for statement before functionDecl, to catch comments in
// includeComments mode. A statement can't start with 'function'
// anyway, so the order doesn't matter otherwise.
var sourceElement = or(statement, functionDecl);
var sourceElements = list(or(comment, sourceElement));
functionBody = expecting(
'functionBody', or(lookAheadToken('}'), sourceElements));
var program = node(
'program',
seq(opt(sourceElements),
// If not at EOF, complain "expecting statement"
expecting('statement', lookAheadTokenType("EOF"))));
return program.parse(this);
};
})();