X-Git-Url: http://git.megacz.com/?p=org.ibex.core.git;a=blobdiff_plain;f=src%2Forg%2Fxwt%2Fjs%2FParser.java;h=b67b59c9bff43c08c1a02f799d4300b6e1704fbd;hp=addf0f03b0b3864a50f5c235bc48f4d02dc0cdb4;hb=6261c41b2ac9d182d8c3541e8e0e5fd00062fa43;hpb=70fdbe545b400b82aebbd9165607b3599c3e0113 diff --git a/src/org/xwt/js/Parser.java b/src/org/xwt/js/Parser.java index addf0f0..b67b59c 100644 --- a/src/org/xwt/js/Parser.java +++ b/src/org/xwt/js/Parser.java @@ -4,28 +4,78 @@ package org.xwt.js; import org.xwt.util.*; import java.io.*; -/** Parses a stream of lexed tokens into a tree of ByteCodeBlock's */ +/** + * Parses a stream of lexed tokens into a tree of JSFunction's. + * + * There are three kinds of things we parse: blocks, statements, and + * expressions. + * + * - Expressions are a special type of statement that evaluates to a + * value (for example, "break" is not an expression, * but "3+2" + * is). Some tokens sequences start expressions (for * example, + * literal numbers) and others continue an expression which * has + * already been begun (for example, '+'). Finally, some * + * expressions are valid targets for an assignment operation; after + * * each of these expressions, continueExprAfterAssignable() is + * called * to check for an assignment operation. + * + * - A statement ends with a semicolon and does not return a value. + * + * - A block is a single statement or a sequence of statements + * surrounded by curly braces. + * + * Each parsing method saves the parserLine before doing its actual + * work and restores it afterwards. This ensures that parsing a + * subexpression does not modify the line number until a token + * *after* the subexpression has been consumed by the parent + * expression. + * + * Technically it would be a better design for this class to build an + * intermediate parse tree and use that to emit bytecode. Here's the + * tradeoff: + * + * Advantages of building a parse tree: + * - easier to apply optimizations + * - would let us handle more sophisticated languages than JavaScript + * + * Advantages of leaving out the parse tree + * - faster compilation + * - less load on the garbage collector + * - much simpler code, easier to understand + * - less error-prone + * + * Fortunately JS is such a simple language that we can get away with + * the half-assed approach and still produce a working, complete + * compiler. + * + * The bytecode language emitted doesn't really cause any appreciable + * semantic loss, and is itself a parseable language very similar to + * Forth or a postfix variant of LISP. This means that the bytecode + * can be transformed into a parse tree, which can be manipulated. + * So if we ever want to add an optimizer, it could easily be done by + * producing a parse tree from the bytecode, optimizing that tree, + * and then re-emitting the bytecode. The parse tree node class + * would also be much simpler since the bytecode language has so few + * operators. + * + * Actually, the above paragraph is slightly inaccurate -- there are + * places where we push a value and then perform an arbitrary number + * of operations using it before popping it; this doesn't parse well. + * But these cases are clearly marked and easy to change if we do + * need to move to a parse tree format. + */ class Parser extends Lexer implements ByteCodes { // Constructors ////////////////////////////////////////////////////// - public Parser(Reader r, String sourceName, int line) throws IOException { - super(r); - this.sourceName = sourceName; - this.line = line; - } + public Parser(Reader r, String sourceName, int line) throws IOException { super(r, sourceName, line); } /** for debugging */ public static void main(String[] s) throws Exception { - Parser p = new Parser(new InputStreamReader(System.in), "stdin", 0); - while(true) { - ByteCodeBlock block = new ByteCodeBlock(0, "stdin"); - p.parseStatement(false, block); - if (block == null) return; - System.out.println(block); - if (p.peekToken() == -1) return; - } + JSFunction block = JSFunction.fromReader("stdin", 0, new InputStreamReader(System.in)); + if (block == null) return; + System.out.println(block); } @@ -33,604 +83,753 @@ class Parser extends Lexer implements ByteCodes { static byte[] precedence = new byte[MAX_TOKEN + 1]; static boolean[] isRightAssociative = new boolean[MAX_TOKEN + 1]; + // Use this as the precedence when we want anything up to the comma + private final static int NO_COMMA = 2; static { - isRightAssociative[ASSIGN] = true; - - precedence[ASSIGN] = 1; - precedence[HOOK] = 2; - precedence[COMMA] = 3; - precedence[OR] = precedence[AND] = 4; - precedence[GT] = precedence[GE] = 5; - precedence[BITOR] = 6; - precedence[BITXOR] = 7; - precedence[BITAND] = 8; - precedence[EQ] = precedence[NE] = 9; - precedence[LT] = precedence[LE] = 10; - precedence[SHEQ] = precedence[SHNE] = 11; - precedence[LSH] = precedence[RSH] = precedence[URSH] = 12; - precedence[ADD] = precedence[SUB] = 13; - precedence[MUL] = precedence[DIV] = precedence[MOD] = 14; - precedence[BITNOT] = 15; - precedence[INC] = precedence[DEC] = 16; - precedence[LP] = 17; - precedence[LB] = 18; - precedence[DOT] = 19; + isRightAssociative[ASSIGN] = + isRightAssociative[ASSIGN_BITOR] = + isRightAssociative[ASSIGN_BITXOR] = + isRightAssociative[ASSIGN_BITAND] = + isRightAssociative[ASSIGN_LSH] = + isRightAssociative[ASSIGN_RSH] = + isRightAssociative[ASSIGN_URSH] = + isRightAssociative[ASSIGN_ADD] = + isRightAssociative[ASSIGN_SUB] = + isRightAssociative[ASSIGN_MUL] = + isRightAssociative[ASSIGN_DIV] = + isRightAssociative[ASSIGN_MOD] = true; + + precedence[COMMA] = 1; + // 2 is intentionally left unassigned. we use minPrecedence==2 for comma separated lists + precedence[ASSIGN] = + precedence[ASSIGN_BITOR] = + precedence[ASSIGN_BITXOR] = + precedence[ASSIGN_BITAND] = + precedence[ASSIGN_LSH] = + precedence[ASSIGN_RSH] = + precedence[ASSIGN_URSH] = + precedence[ASSIGN_ADD] = + precedence[ASSIGN_SUB] = + precedence[ASSIGN_MUL] = + precedence[ASSIGN_DIV] = + precedence[ASSIGN_MOD] = 3; + precedence[HOOK] = 4; + precedence[OR] = 5; + precedence[AND] = 6; + precedence[BITOR] = 7; + precedence[BITXOR] = 8; + precedence[BITAND] = 9; + precedence[EQ] = precedence[NE] = precedence[SHEQ] = precedence[SHNE] = 10; + precedence[LT] = precedence[LE] = precedence[GT] = precedence[GE] = 11; + precedence[LSH] = precedence[RSH] = precedence[URSH] = 12; + precedence[ADD] = precedence[SUB] = 12; + precedence[MUL] = precedence[DIV] = precedence[MOD] = 13; + precedence[BITNOT] = precedence[BANG] = precedence[TYPEOF] = 14; + precedence[DOT] = precedence[LB] = precedence[LP] = precedence[INC] = precedence[DEC] = 15; } // Parsing Logic ///////////////////////////////////////////////////////// - private ByteCodeBlock newbb(int line) { return new ByteCodeBlock(line, sourceName); } - /** gets a token and throws an exception if it is not code */ - public void consume(int code) throws IOException { - if (getToken() != code) - throw new ParserException("expected " + codeToString[code] + ", got " + (op == -1 ? "EOF" : codeToString[op])); + private void consume(int code) throws IOException { + if (getToken() != code) { + if(code == NAME) switch(op) { + case RETURN: case TYPEOF: case BREAK: case CONTINUE: case TRY: case THROW: + case ASSERT: case NULL: case TRUE: case FALSE: case IN: case IF: case ELSE: + case SWITCH: case CASE: case DEFAULT: case WHILE: case VAR: case WITH: + case CATCH: case FINALLY: + throw pe("Bad variable name; '" + codeToString[op].toLowerCase() + "' is a javascript keyword"); + } + throw pe("expected " + codeToString[code] + ", got " + (op == -1 ? "EOF" : codeToString[op])); + } } - /** append the largest expression beginning with prefix containing no operators of precedence below minPrecedence */ - public ByteCodeBlock startExpr() throws IOException { return startExpr(-1); } - public void startExpr(ByteCodeBlock block) throws IOException { startExpr(-1, block); } - public ByteCodeBlock startExpr(int minPrecedence) throws IOException { - ByteCodeBlock ret = new ByteCodeBlock(line, sourceName); - startExpr(minPrecedence, ret); - return ret.size() == 0 ? null : ret; + /** + * Parse the largest possible expression containing no operators + * of precedence below minPrecedence and append the + * bytecodes for that expression to appendTo; the + * appended bytecodes MUST grow the stack by exactly one element. + */ + private void startExpr(JSFunction appendTo, int minPrecedence) throws IOException { + int saveParserLine = parserLine; + _startExpr(appendTo, minPrecedence); + parserLine = saveParserLine; } - - public void startExpr(int minPrecedence, ByteCodeBlock appendTo) throws IOException { - int tok = getToken(); - int curLine = line; - if (tok == -1) return; - if (minPrecedence > 0 && precedence[tok] != 0) - if (precedence[tok] < minPrecedence || (precedence[tok] == minPrecedence && !isRightAssociative[tok])) - { pushBackToken(); return; } - - ByteCodeBlock b = appendTo; - - switch (tok) { - case NUMBER: continueExpr(b.add(ByteCodeBlock.LITERAL, number), minPrecedence); return; - case STRING: continueExpr(b.add(ByteCodeBlock.LITERAL, string), minPrecedence); return; - case THIS: continueExpr(b.add(TOPSCOPE, null), minPrecedence); return; - case NULL: continueExpr(b.add(ByteCodeBlock.LITERAL, null), minPrecedence); return; - case TRUE: case FALSE: continueExpr(b.add(ByteCodeBlock.LITERAL, new Boolean(tok == TRUE)), minPrecedence); return; - case LB: { - b.add(ARRAY, new Integer(0)); - int i = 0; - while(true) { - int size = b.size(); - startExpr(b); - if (size == b.size()) - if (peekToken() == RB) { consume(RB); continueExpr(b, minPrecedence); return; } - b.add(LITERAL, new Integer(i++)); - if (size == b.size()) b.add(LITERAL, null); - b.add(PUT); - b.add(POP); - if (peekToken() == RB) { consume(RB); continueExpr(b, minPrecedence); return; } - consume(COMMA); - } - } - case SUB: { - consume(NUMBER); - continueExpr(b.add(ByteCodeBlock.LITERAL, new Double(number.doubleValue() * -1)), minPrecedence); - return; - } - case LP: { - startExpr(b); - consume(RP); - continueExpr(b, minPrecedence); - return; - } - case INC: case DEC: { - // prefix - startExpr(precedence[tok], b); - b.set(b.size() - 1, tok, new Boolean(true)); - continueExpr(b, minPrecedence); - return; - } - case BANG: case BITNOT: case TYPEOF: { - startExpr(precedence[tok], b); - b.add(tok); - continueExpr(b, minPrecedence); - return; - } - case LC: { - b.add(OBJECT, null); - if (peekToken() != RC) while(true) { - if (peekToken() != NAME && peekToken() != STRING) throw new Error("expected NAME or STRING"); - getToken(); - b.add(LITERAL, string); - consume(COLON); - startExpr(b); - b.add(PUT); - b.add(POP); - if (peekToken() == RC) break; - consume(COMMA); - if (peekToken() == RC) break; - } - consume(RC); - continueExpr(b, minPrecedence); - return; - } - case NAME: { - String name = string; - if (peekToken() == ASSIGN) { - consume(ASSIGN); - b.add(TOPSCOPE); - b.add(ByteCodeBlock.LITERAL, name); - startExpr(minPrecedence, b); - b.add(ByteCodeBlock.PUT); - b.add(ByteCodeBlock.SWAP); - b.add(ByteCodeBlock.POP); - } else { - b.add(TOPSCOPE); - b.add(ByteCodeBlock.LITERAL, name); - b.add(ByteCodeBlock.GET); - } - continueExpr(b, minPrecedence); - return; - } - case FUNCTION: { - consume(LP); - int numArgs = 0; - ByteCodeBlock b2 = newbb(curLine); - b2.add(TOPSCOPE); - b2.add(SWAP); - b2.add(LITERAL, "arguments"); - b2.add(LITERAL, "arguments"); - b2.add(DECLARE); - b2.add(SWAP); - b2.add(PUT); - b2.add(SWAP); - b2.add(POP); - if (peekToken() == RP) consume(RP); - else while(true) { - if (peekToken() == COMMA) { - consume(COMMA); - - } else { - consume(NAME); - - // declare the name - b2.add(LITERAL, string); - b2.add(DECLARE); - - // retrieve it from the arguments array - b2.add(LITERAL, new Integer(numArgs)); - b2.add(GET_PRESERVE); - b2.add(SWAP); - b2.add(POP); - - // put it to the current scope - b2.add(TOPSCOPE); - b2.add(SWAP); - b2.add(LITERAL, string); - b2.add(SWAP); - b2.add(PUT); - - // clean the stack - b2.add(POP); - b2.add(POP); - - if (peekToken() == RP) { consume(RP); break; } - consume(COMMA); - } - numArgs++; - } - // pop off the arguments array - b2.add(POP); - parseStatement(true, b2); - b2.add(LITERAL, null); - b2.add(RETURN); - continueExpr(b.add(NEWFUNCTION, b2), minPrecedence); - return; - } - default: pushBackToken(); return; - } + private void _startExpr(JSFunction appendTo, int minPrecedence) throws IOException { + int tok = getToken(); + JSFunction b = appendTo; + + switch (tok) { + case -1: throw pe("expected expression"); + + // all of these simply push values onto the stack + case NUMBER: b.add(parserLine, LITERAL, number); break; + case STRING: b.add(parserLine, LITERAL, string); break; + case NULL: b.add(parserLine, LITERAL, null); break; + case TRUE: case FALSE: b.add(parserLine, LITERAL, JS.B(tok == TRUE)); break; + + // (.foo) syntax + case DOT: { + consume(NAME); + b.add(parserLine, TOPSCOPE); + b.add(parserLine, LITERAL, ""); + b.add(parserLine, GET); + b.add(parserLine, LITERAL, string); + b.add(parserLine, GET); + continueExpr(b, minPrecedence); + break; + } + + case LB: { + b.add(parserLine, ARRAY, JS.ZERO); // push an array onto the stack + int size0 = b.size; + int i = 0; + if (peekToken() != RB) + while(true) { // iterate over the initialization values + int size = b.size; + b.add(parserLine, LITERAL, JS.N(i++)); // push the index in the array to place it into + if (peekToken() == COMMA || peekToken() == RB) + b.add(parserLine, LITERAL, null); // for stuff like [1,,2,] + else + startExpr(b, NO_COMMA); // push the value onto the stack + b.add(parserLine, PUT); // put it into the array + b.add(parserLine, POP); // discard the value remaining on the stack + if (peekToken() == RB) break; + consume(COMMA); + } + b.set(size0 - 1, JS.N(i)); // back at the ARRAY instruction, write the size of the array + consume(RB); + break; + } + case SUB: { // negative literal (like "3 * -1") + consume(NUMBER); + b.add(parserLine, LITERAL, JS.N(number.doubleValue() * -1)); + break; + } + case LP: { // grouping (not calling) + startExpr(b, -1); + consume(RP); + break; + } + case INC: case DEC: { // prefix (not postfix) + startExpr(b, precedence[tok]); + int prev = b.size - 1; + if (b.get(prev) == GET && b.getArg(prev) != null) + b.set(prev, LITERAL, b.getArg(prev)); + else if(b.get(prev) == GET) + b.pop(); + else + throw pe("prefixed increment/decrement can only be performed on a valid assignment target"); + b.add(parserLine, GET_PRESERVE, Boolean.TRUE); + b.add(parserLine, LITERAL, JS.N(1)); + b.add(parserLine, tok == INC ? ADD : SUB, JS.N(2)); + b.add(parserLine, PUT, null); + b.add(parserLine, SWAP, null); + b.add(parserLine, POP, null); + break; + } + case BANG: case BITNOT: case TYPEOF: { + startExpr(b, precedence[tok]); + b.add(parserLine, tok); + break; + } + case LC: { // object constructor + b.add(parserLine, OBJECT, null); // put an object on the stack + if (peekToken() != RC) + while(true) { + if (peekToken() != NAME && peekToken() != STRING) + throw pe("expected NAME or STRING"); + getToken(); + b.add(parserLine, LITERAL, string); // grab the key + consume(COLON); + startExpr(b, NO_COMMA); // grab the value + b.add(parserLine, PUT); // put the value into the object + b.add(parserLine, POP); // discard the remaining value + if (peekToken() == RC) break; + consume(COMMA); + if (peekToken() == RC) break; // we permit {,,} -- I'm not sure if ECMA does + } + consume(RC); + break; + } + case NAME: { + b.add(parserLine, TOPSCOPE); + b.add(parserLine, LITERAL, string); + continueExprAfterAssignable(b,minPrecedence); + break; + } + case FUNCTION: { + consume(LP); + int numArgs = 0; + JSFunction b2 = new JSFunction(sourceName, parserLine, null); + b.add(parserLine, NEWFUNCTION, b2); + + // function prelude; arguments array is already on the stack + b2.add(parserLine, TOPSCOPE); + b2.add(parserLine, SWAP); + b2.add(parserLine, DECLARE, "arguments"); // declare arguments (equivalent to 'var arguments;') + b2.add(parserLine, SWAP); // set this.arguments and leave the value on the stack + b2.add(parserLine, PUT); + + while(peekToken() != RP) { // run through the list of argument names + numArgs++; + if (peekToken() == NAME) { + consume(NAME); // a named argument + String varName = string; + + b2.add(parserLine, DUP); // dup the args array + b2.add(parserLine, GET, JS.N(numArgs - 1)); // retrieve it from the arguments array + b2.add(parserLine, TOPSCOPE); + b2.add(parserLine, SWAP); + b2.add(parserLine, DECLARE, varName); // declare the name + b2.add(parserLine, SWAP); + b2.add(parserLine, PUT); + b2.add(parserLine, POP); // pop the value + b2.add(parserLine, POP); // pop the scope + } + if (peekToken() == RP) break; + consume(COMMA); + } + consume(RP); + + b2.numFormalArgs = numArgs; + b2.add(parserLine, POP); // pop off the arguments array + b2.add(parserLine, POP); // pop off TOPSCOPE + + if(peekToken() != LC) + throw pe("JSFunctions must have a block surrounded by curly brackets"); + + parseBlock(b2, null); // the function body + + b2.add(parserLine, LITERAL, null); // in case we "fall out the bottom", return NULL + b2.add(parserLine, RETURN); + + break; + } + default: throw pe("expected expression, found " + codeToString[tok] + ", which cannot start an expression"); + } + + // attempt to continue the expression + continueExpr(b, minPrecedence); } - public void continueExpr(ByteCodeBlock prefix, int minPrecedence) throws IOException { - int tok = getToken(); - int curLine = line; - if (tok == -1) return; - if (minPrecedence > 0 && precedence[tok] != 0) - if (precedence[tok] < minPrecedence || (precedence[tok] == minPrecedence && !isRightAssociative[tok])) - { pushBackToken(); return; } - - if (prefix == null) throw new Error("got null prefix"); - ByteCodeBlock b = prefix; - - switch (tok) { + /** + * Assuming that a complete assignable (lvalue) has just been + * parsed and the object and key are on the stack, + * continueExprAfterAssignable will attempt to parse an + * expression that modifies the assignable. This method always + * decreases the stack depth by exactly one element. + */ + private void continueExprAfterAssignable(JSFunction b,int minPrecedence) throws IOException { + int saveParserLine = parserLine; + _continueExprAfterAssignable(b,minPrecedence); + parserLine = saveParserLine; + } + private void _continueExprAfterAssignable(JSFunction b,int minPrecedence) throws IOException { + if (b == null) throw new Error("got null b; this should never happen"); + int tok = getToken(); + if (minPrecedence != -1 && (precedence[tok] < minPrecedence || (precedence[tok] == minPrecedence && !isRightAssociative[tok]))) + // force the default case + tok = -1; + switch(tok) { case ASSIGN_BITOR: case ASSIGN_BITXOR: case ASSIGN_BITAND: case ASSIGN_LSH: case ASSIGN_RSH: case ASSIGN_URSH: - case ASSIGN_ADD: case ASSIGN_SUB: case ASSIGN_MUL: case ASSIGN_DIV: case ASSIGN_MOD: { - prefix.set(prefix.size() - 1, b.GET_PRESERVE, new Boolean(true)); - startExpr(precedence[tok - 1], b); - prefix.add(tok - 1); - prefix.add(PUT); - prefix.add(SWAP); - prefix.add(POP); - continueExpr(b, minPrecedence); - return; - } - - case INC: case DEC: { - // postfix - b.set(b.size() - 1, tok, new Boolean(false)); - continueExpr(b, minPrecedence); - return; - } - - case LP: { - // invocation - int i = 0; - while(peekToken() != RP) { - i++; - startExpr(b); - if (peekToken() == RP) break; - consume(COMMA); - } - consume(RP); - b.add(CALL, new Integer(i)); - continueExpr(b, minPrecedence); - return; - } + case ASSIGN_MUL: case ASSIGN_DIV: case ASSIGN_MOD: case ASSIGN_ADD: case ASSIGN_SUB: { + b.add(parserLine, GET_PRESERVE); + startExpr(b, precedence[tok]); + int size = b.size; + if (tok == ASSIGN_ADD || tok == ASSIGN_SUB) { + b.add(parserLine, tok); + } + // tok-1 is always s/^ASSIGN_// (0 is BITOR, 1 is ASSIGN_BITOR, etc) + b.add(parserLine, tok - 1, tok-1==ADD ? JS.N(2) : null); + b.add(parserLine, PUT); + b.add(parserLine, SWAP); + b.add(parserLine, POP); + if (tok == ASSIGN_ADD || tok == ASSIGN_SUB) b.set(size, tok, JS.N(b.size - size)); + break; + } + case INC: case DEC: { // postfix + b.add(parserLine, GET_PRESERVE, Boolean.TRUE); + b.add(parserLine, LITERAL, JS.N(1)); + b.add(parserLine, tok == INC ? ADD : SUB, JS.N(2)); + b.add(parserLine, PUT, null); + b.add(parserLine, SWAP, null); + b.add(parserLine, POP, null); + b.add(parserLine, LITERAL, JS.N(1)); + b.add(parserLine, tok == INC ? SUB : ADD, null); // undo what we just did, since this is postfix + break; + } + case ASSIGN: { + startExpr(b, precedence[tok]); + b.add(parserLine, PUT); + b.add(parserLine, SWAP); + b.add(parserLine, POP); + break; + } + case LP: { + + // Method calls are implemented by doing a GET_PRESERVE + // first. If the object supports method calls, it will + // return JS.METHOD + int n = parseArgs(b, 2); + b.add(parserLine, GET_PRESERVE); + b.add(parserLine, CALLMETHOD, JS.N(n)); + break; + } + default: { + pushBackToken(); + if(b.get(b.size-1) == LITERAL && b.getArg(b.size-1) != null) + b.set(b.size-1,GET,b.getArg(b.size-1)); + else + b.add(parserLine, GET); + return; + } + } + } + + /** + * Assuming that a complete expression has just been parsed, + * continueExpr will attempt to extend this expression by + * parsing additional tokens and appending additional bytecodes. + * + * No operators with precedence less than minPrecedence + * will be parsed. + * + * If any bytecodes are appended, they will not alter the stack + * depth. + */ + private void continueExpr(JSFunction b, int minPrecedence) throws IOException { + int saveParserLine = parserLine; + _continueExpr(b, minPrecedence); + parserLine = saveParserLine; + } + private void _continueExpr(JSFunction b, int minPrecedence) throws IOException { + if (b == null) throw new Error("got null b; this should never happen"); + int tok = getToken(); + if (tok == -1) return; + if (minPrecedence != -1 && (precedence[tok] < minPrecedence || (precedence[tok] == minPrecedence && !isRightAssociative[tok]))) { + pushBackToken(); + return; + } + + switch (tok) { + case LP: { // invocation (not grouping) + int n = parseArgs(b, 1); + b.add(parserLine, CALL, JS.N(n)); + break; + } case BITOR: case BITXOR: case BITAND: case SHEQ: case SHNE: case LSH: - case RSH: case URSH: case ADD: case MUL: case DIV: case MOD: - case GT: case GE: case EQ: case NE: case LT: case LE: case SUB: { - startExpr(precedence[tok], b); - b.add(tok); - continueExpr(b, minPrecedence); - return; - } - - case OR: case AND: { - b.add(tok == AND ? b.JF : b.JT, new Integer(0)); - int size = b.size(); - startExpr(precedence[tok], b); - b.set(size - 1, new Integer(b.size() - size + 2)); - b.add(JMP, new Integer(2)); - b.add(LITERAL, tok == AND ? new Boolean(false) : new Boolean(true)); - continueExpr(b, minPrecedence); - return; - } - - case DOT: { - consume(NAME); - String target = string; - if (peekToken() == ASSIGN) { - consume(ASSIGN); - b.add(LITERAL, target); - startExpr(b); - b.add(PUT); - b.add(SWAP); - b.add(POP); - } else { - b.add(LITERAL, target); - b.add(GET); - } - continueExpr(b, minPrecedence); - return; - } - - case LB: { - startExpr(b); - consume(RB); - if (peekToken() == ASSIGN) { - consume(ASSIGN); - startExpr(b); - b.add(PUT); - b.add(SWAP); - b.add(POP); - } else { - b.add(GET); - } - continueExpr(b, minPrecedence); - return; - } - - case HOOK: { - b.add(JF, new Integer(0)); - int size = b.size(); - startExpr(b); - b.set(size - 1, new Integer(b.size() - size + 2)); - b.add(JMP, new Integer(0)); - consume(COLON); - size = b.size(); - startExpr(b); - b.set(size - 1, new Integer(b.size() - size + 1)); - continueExpr(b, minPrecedence); - return; - } - - default: { pushBackToken(); return; } - } + case RSH: case URSH: case MUL: case DIV: case MOD: + case GT: case GE: case EQ: case NE: case LT: case LE: case SUB: { + startExpr(b, precedence[tok]); + b.add(parserLine, tok); + break; + } + case ADD: { + int count=1; + int nextTok; + do { + startExpr(b,precedence[tok]); + count++; + nextTok = getToken(); + } while(nextTok == tok); + pushBackToken(); + b.add(parserLine, tok, JS.N(count)); + break; + } + case OR: case AND: { + b.add(parserLine, tok == AND ? b.JF : b.JT, JS.ZERO); // test to see if we can short-circuit + int size = b.size; + startExpr(b, precedence[tok]); // otherwise check the second value + b.add(parserLine, JMP, JS.N(2)); // leave the second value on the stack and jump to the end + b.add(parserLine, LITERAL, tok == AND ? + JS.B(false) : JS.B(true)); // target of the short-circuit jump is here + b.set(size - 1, JS.N(b.size - size)); // write the target of the short-circuit jump + break; + } + case DOT: { + // support foo..bar syntax for foo[""].bar + if (peekToken() == DOT) { + string = ""; + } else { + consume(NAME); + } + b.add(parserLine, LITERAL, string); + continueExprAfterAssignable(b,minPrecedence); + break; + } + case LB: { // subscripting (not array constructor) + startExpr(b, -1); + consume(RB); + continueExprAfterAssignable(b,minPrecedence); + break; + } + case HOOK: { + b.add(parserLine, JF, JS.ZERO); // jump to the if-false expression + int size = b.size; + startExpr(b, minPrecedence); // write the if-true expression + b.add(parserLine, JMP, JS.ZERO); // if true, jump *over* the if-false expression + b.set(size - 1, JS.N(b.size - size + 1)); // now we know where the target of the jump is + consume(COLON); + size = b.size; + startExpr(b, minPrecedence); // write the if-false expression + b.set(size - 1, JS.N(b.size - size + 1)); // this is the end; jump to here + break; + } + case COMMA: { + // pop the result of the previous expression, it is ignored + b.add(parserLine,POP); + startExpr(b,-1); + break; + } + default: { + pushBackToken(); + return; + } + } + + continueExpr(b, minPrecedence); // try to continue the expression } - /** a block is either a single statement or a list of statements surrounded by curly braces; all expressions are also statements */ - public ByteCodeBlock parseStatement() throws IOException { - ByteCodeBlock ret = new ByteCodeBlock(line, sourceName); - ret.add(ret.LITERAL, null); - parseStatement(false, ret); - if (ret.size() == 1) return null; - return ret; + // parse a set of comma separated function arguments, assume LP has already been consumed + // if swap is true, (because the function is already on the stack) we will SWAP after each argument to keep it on top + private int parseArgs(JSFunction b, int pushdown) throws IOException { + int i = 0; + while(peekToken() != RP) { + i++; + if (peekToken() != COMMA) { + startExpr(b, NO_COMMA); + b.add(parserLine, SWAP, JS.N(pushdown)); + if (peekToken() == RP) break; + } + consume(COMMA); + } + consume(RP); + return i; } - - public void parseStatement(boolean requireBraces) throws IOException { - parseStatement(requireBraces, new ByteCodeBlock(line, sourceName)); + + /** Parse a block of statements which must be surrounded by LC..RC. */ + void parseBlock(JSFunction b) throws IOException { parseBlock(b, null); } + void parseBlock(JSFunction b, String label) throws IOException { + int saveParserLine = parserLine; + _parseBlock(b, label); + parserLine = saveParserLine; } - public void parseStatement(boolean requireBraces, ByteCodeBlock b) throws IOException { - int tok = peekToken(); - if (tok == -1) return; - boolean braced = tok == LC; - if (requireBraces && !braced) throw new ParserException("expected {, got " + codeToString[tok]); - if (braced) consume(LC); - int curLine = line; - while(true) { - switch(tok = peekToken()) { - - case THROW: case RETURN: case ASSERT: { - getToken(); - if (tok == RETURN && peekToken() == SEMI) b.add(LITERAL, null); - else startExpr(b); - consume(SEMI); - b.add(tok); - break; - } - - case BREAK: case CONTINUE: { - getToken(); - if (peekToken() == NAME) consume(NAME); - b.add(tok, string); - consume(SEMI); - break; - } - - case SEMI: - consume(SEMI); - if (!braced) return; - break; - - case VAR: { - consume(VAR); - b.add(TOPSCOPE); // push the current scope - while(true) { - consume(NAME); - String name = string; - b.add(LITERAL, name); // push the name to be declared - b.add(DECLARE); // declare it - if (peekToken() == ASSIGN) { // if there is an '=' after the variable name - b.add(LITERAL, name); // put the var name back on the stack - consume(ASSIGN); - startExpr(b); - b.add(PUT); - b.add(POP); - } - if (peekToken() != COMMA) break; - consume(COMMA); - } - b.add(POP); - if (peekToken() == SEMI) consume(SEMI); - break; - } - - case IF: { - consume(IF); - consume(LP); - startExpr(b); - consume(RP); - - b.add(JF, new Integer(0)); - int size = b.size(); - parseStatement(false, b); - - if (peekToken() == ELSE) { - consume(ELSE); - b.set(size - 1, new Integer(2 + b.size() - size)); - b.add(JMP, new Integer(0)); - size = b.size(); - parseStatement(false, b); - } - b.set(size - 1, new Integer(1 + b.size() - size)); - break; - } - - case WHILE: { - consume(WHILE); - consume(LP); - ByteCodeBlock loop = newbb(curLine); - b.add(LOOP, loop); - - loop.add(POP); - startExpr(loop); - loop.add(JT, new Integer(2)); - loop.add(BREAK); - consume(RP); - parseStatement(false, loop); - - // if we fall out of the end, definately continue - loop.add(CONTINUE); - break; - } - - case SWITCH: { - consume(SWITCH); - consume(LP); - ByteCodeBlock loop = newbb(curLine); - b.add(LOOP, loop); - startExpr(loop); - consume(RP); - consume(LC); - while(true) - if (peekToken() == CASE) { - consume(CASE); - loop.add(DUP); - startExpr(loop); - consume(COLON); - loop.add(EQ); - loop.add(JF, new Integer(0)); - int size = loop.size(); - while(peekToken() != CASE && peekToken() != DEFAULT && peekToken() != RC) { - int size2 = loop.size(); - parseStatement(false, loop); - if (size2 == loop.size()) break; - } - loop.set(size - 1, new Integer(1 + loop.size() - size)); - } else if (peekToken() == DEFAULT) { - consume(DEFAULT); - consume(COLON); - while(peekToken() != CASE && peekToken() != DEFAULT && peekToken() != RC) { - int size2 = loop.size(); - parseStatement(false, loop); - if (size2 == loop.size()) break; - } - } else if (peekToken() == RC) { - consume(RC); - loop.add(BREAK); - break; - } else { - throw new ParserException("expected CASE, DEFAULT, or RC; got " + codeToString[peekToken()]); - } - break; - } - - case DO: { - consume(DO); - ByteCodeBlock loop = newbb(curLine); - b.add(LOOP, loop); - - parseStatement(false, loop); - consume(WHILE); - consume(LP); - startExpr(loop); - loop.add(JT, new Integer(2)); - loop.add(BREAK); - loop.add(CONTINUE); - consume(RP); - consume(SEMI); - break; - } - - case TRY: { - // We deliberately allow you to omit braces in catch{}/finally{} if they are single statements... - consume(TRY); - parseStatement(true, b); - - if (peekToken() == CATCH) { - getToken(); - consume(LP); - consume(NAME); - consume(RP); - parseStatement(); // just discard the catchblock - } - - if (peekToken() == FINALLY) { - consume(FINALLY); - parseStatement(false, b); - } - break; - } - - case FOR: { - consume(FOR); - consume(LP); - - tok = getToken(); - if (tok == VAR) tok = getToken(); - String varName = string; - boolean forIn = peekToken() == IN; - pushBackToken(tok, varName); - - if (forIn) { - consume(NAME); - consume(IN); - startExpr(b); - b.add(PUSHKEYS); - b.add(LITERAL, "length"); - b.add(GET); - consume(RP); - ByteCodeBlock b2 = newbb(curLine); - b.add(SCOPE, b2); - b2.add(LITERAL, new Integer(1)); - b2.add(SUB); - b2.add(DUP); - b2.add(LITERAL, new Integer(0)); - b2.add(LT); - b2.add(JT, new Integer(7)); - b2.add(GET_PRESERVE); - b2.add(LITERAL, varName); - b2.add(LITERAL, varName); - b2.add(DECLARE); - b2.add(PUT); - parseStatement(false, b2); - break; - - } else { - ByteCodeBlock b2 = newbb(curLine); - b.add(SCOPE, b2); - b.add(POP); - - int size = b2.size(); - startExpr(b2); - if (b2.size() - size > 0) b2.add(POP); - consume(SEMI); - ByteCodeBlock e2 = startExpr(); - consume(SEMI); - if (e2 == null) e2 = newbb(curLine).add(b.LITERAL, null); - - ByteCodeBlock b3 = newbb(curLine); - b2.add(LOOP, b3); - b2.add(LITERAL, null); - - b3.add(JT, new Integer(0)); - size = b3.size(); - startExpr(b3); - consume(RP); - if (b3.size() - size > 0) b3.add(POP); - b3.set(size - 1, new Integer(b3.size() - size + 1)); - - b3.paste(e2); - b3.add(JT, new Integer(2)); - b3.add(BREAK); - parseStatement(false, b3); - b3.add(BREAK); - break; - } - } - - case NAME: { - consume(NAME); - String name = string; - if (peekToken() == COLON) { - consume(COLON); - b.add(ByteCodeBlock.LABEL, string); - break; - } else { - pushBackToken(NAME, name); - // fall through to default case - } - } - // fall through - case RC: - if (tok == RC && braced) { consume(RC); return; } - // fall through - default: { - int size = b.size(); - startExpr(b); - if (size == b.size()) return; - b.add(POP); - if (peekToken() == SEMI) consume(SEMI); - break; - } - } - - if (!braced) return; - } + void _parseBlock(JSFunction b, String label) throws IOException { + if (peekToken() == -1) return; + else if (peekToken() != LC) parseStatement(b, null); + else { + consume(LC); + while(peekToken() != RC && peekToken() != -1) parseStatement(b, null); + consume(RC); + } } - - class ParserException extends RuntimeException { - public ParserException(String s) { super(sourceName + ":" + line + " " + s); } + + /** Parse a single statement, consuming the RC or SEMI which terminates it. */ + void parseStatement(JSFunction b, String label) throws IOException { + int saveParserLine = parserLine; + _parseStatement(b, label); + parserLine = saveParserLine; + } + void _parseStatement(JSFunction b, String label) throws IOException { + int tok = peekToken(); + if (tok == -1) return; + switch(tok = getToken()) { + + case THROW: case ASSERT: case RETURN: { + if (tok == RETURN && peekToken() == SEMI) + b.add(parserLine, LITERAL, null); + else + startExpr(b, -1); + b.add(parserLine, tok); + consume(SEMI); + break; + } + case BREAK: case CONTINUE: { + if (peekToken() == NAME) consume(NAME); + b.add(parserLine, tok, string); + consume(SEMI); + break; + } + case VAR: { + b.add(parserLine, TOPSCOPE); // push the current scope + while(true) { + consume(NAME); + b.add(parserLine, DECLARE, string); // declare it + if (peekToken() == ASSIGN) { // if there is an '=' after the variable name + consume(ASSIGN); + startExpr(b, NO_COMMA); + b.add(parserLine, PUT); // assign it + b.add(parserLine, POP); // clean the stack + } else { + b.add(parserLine, POP); // pop the string pushed by declare + } + if (peekToken() != COMMA) break; + consume(COMMA); + } + b.add(parserLine, POP); // pop off the topscope + if ((mostRecentlyReadToken != RC || peekToken() == SEMI) && peekToken() != -1 && mostRecentlyReadToken != SEMI) consume(SEMI); + break; + } + case IF: { + consume(LP); + startExpr(b, -1); + consume(RP); + + b.add(parserLine, JF, JS.ZERO); // if false, jump to the else-block + int size = b.size; + parseStatement(b, null); + + if (peekToken() == ELSE) { + consume(ELSE); + b.add(parserLine, JMP, JS.ZERO); // if we took the true-block, jump over the else-block + b.set(size - 1, JS.N(b.size - size + 1)); + size = b.size; + parseStatement(b, null); + } + b.set(size - 1, JS.N(b.size - size + 1)); // regardless of which branch we took, b[size] needs to point here + break; + } + case WHILE: { + consume(LP); + if (label != null) b.add(parserLine, LABEL, label); + b.add(parserLine, LOOP); + int size = b.size; + b.add(parserLine, POP); // discard the first-iteration indicator + startExpr(b, -1); + b.add(parserLine, JT, JS.N(2)); // if the while() clause is true, jump over the BREAK + b.add(parserLine, BREAK); + consume(RP); + parseStatement(b, null); + b.add(parserLine, CONTINUE); // if we fall out of the end, definately continue + b.set(size - 1, JS.N(b.size - size + 1)); // end of the loop + break; + } + case SWITCH: { + consume(LP); + if (label != null) b.add(parserLine, LABEL, label); + b.add(parserLine, LOOP); + int size0 = b.size; + startExpr(b, -1); + consume(RP); + consume(LC); + while(true) + if (peekToken() == CASE) { // we compile CASE statements like a bunch of if..else's + consume(CASE); + b.add(parserLine, DUP); // duplicate the switch() value; we'll consume one copy + startExpr(b, -1); + consume(COLON); + b.add(parserLine, EQ); // check if we should do this case-block + b.add(parserLine, JF, JS.ZERO); // if not, jump to the next one + int size = b.size; + while(peekToken() != CASE && peekToken() != DEFAULT && peekToken() != RC) parseStatement(b, null); + b.set(size - 1, JS.N(1 + b.size - size)); + } else if (peekToken() == DEFAULT) { + consume(DEFAULT); + consume(COLON); + while(peekToken() != CASE && peekToken() != DEFAULT && peekToken() != RC) parseStatement(b, null); + } else if (peekToken() == RC) { + consume(RC); + b.add(parserLine, BREAK); // break out of the loop if we 'fall through' + break; + } else { + throw pe("expected CASE, DEFAULT, or RC; got " + codeToString[peekToken()]); + } + b.set(size0 - 1, JS.N(b.size - size0 + 1)); // end of the loop + break; + } + + case DO: { + if (label != null) b.add(parserLine, LABEL, label); + b.add(parserLine, LOOP); + int size = b.size; + parseStatement(b, null); + consume(WHILE); + consume(LP); + startExpr(b, -1); + b.add(parserLine, JT, JS.N(2)); // check the while() clause; jump over the BREAK if true + b.add(parserLine, BREAK); + b.add(parserLine, CONTINUE); + consume(RP); + consume(SEMI); + b.set(size - 1, JS.N(b.size - size + 1)); // end of the loop; write this location to the LOOP instruction + break; + } + + case TRY: { + b.add(parserLine, TRY); // try bytecode causes a TryMarker to be pushed + int tryInsn = b.size - 1; + // parse the expression to be TRYed + parseStatement(b, null); + // pop the try marker. this is pushed when the TRY bytecode is executed + b.add(parserLine, POP); + // jump forward to the end of the catch block, start of the finally block + b.add(parserLine, JMP); + int successJMPInsn = b.size - 1; + + if (peekToken() != CATCH && peekToken() != FINALLY) + throw pe("try without catch or finally"); + + int catchJMPDistance = -1; + if (peekToken() == CATCH) { + catchJMPDistance = b.size - tryInsn; + String exceptionVar; + getToken(); + consume(LP); + consume(NAME); + exceptionVar = string; + consume(RP); + b.add(parserLine, TOPSCOPE); // the exception is on top of the stack; put it to the chosen name + b.add(parserLine, SWAP); + b.add(parserLine, LITERAL,exceptionVar); + b.add(parserLine, SWAP); + b.add(parserLine, PUT); + b.add(parserLine, POP); + b.add(parserLine, POP); + parseStatement(b, null); + // pop the try and catch markers + b.add(parserLine,POP); + b.add(parserLine,POP); + } + + // jump here if no exception was thrown + b.set(successJMPInsn, JS.N(b.size - successJMPInsn)); + + int finallyJMPDistance = -1; + if (peekToken() == FINALLY) { + b.add(parserLine, LITERAL, null); // null FinallyData + finallyJMPDistance = b.size - tryInsn; + consume(FINALLY); + parseStatement(b, null); + b.add(parserLine,FINALLY_DONE); + } + + // setup the TRY arguments + b.set(tryInsn, new int[] { catchJMPDistance, finallyJMPDistance }); + + break; + } + + case FOR: { + consume(LP); + + tok = getToken(); + boolean hadVar = false; // if it's a for..in, we ignore the VAR + if (tok == VAR) { hadVar = true; tok = getToken(); } + String varName = string; + boolean forIn = peekToken() == IN; // determine if this is a for..in loop or not + pushBackToken(tok, varName); + + if (forIn) { + b.add(parserLine, NEWSCOPE); // for-loops always create new scopes + b.add(parserLine, LITERAL, varName); // declare the new variable + b.add(parserLine, DECLARE); + + b.add(parserLine, LOOP); // we actually only add this to ensure that BREAK works + b.add(parserLine, POP); // discard the first-iteration indicator + int size = b.size; + consume(NAME); + consume(IN); + startExpr(b, -1); + b.add(parserLine, PUSHKEYS); // push the keys as an array; check the length + b.add(parserLine, LITERAL, "length"); + b.add(parserLine, GET); + consume(RP); + + b.add(parserLine, LITERAL, JS.N(1)); // decrement the length + b.add(parserLine, SUB); + b.add(parserLine, DUP); + b.add(parserLine, LITERAL, JS.ZERO); // see if we've exhausted all the elements + b.add(parserLine, LT); + b.add(parserLine, JF, JS.N(2)); + b.add(parserLine, BREAK); // if we have, then BREAK + b.add(parserLine, GET_PRESERVE); // get the key out of the keys array + b.add(parserLine, LITERAL, varName); + b.add(parserLine, PUT); // write it to this[varName] + parseStatement(b, null); // do some stuff + b.add(parserLine, CONTINUE); // continue if we fall out the bottom + + b.set(size - 1, JS.N(b.size - size + 1)); // BREAK to here + b.add(parserLine, OLDSCOPE); // restore the scope + + } else { + if (hadVar) pushBackToken(VAR, null); // yeah, this actually matters + b.add(parserLine, NEWSCOPE); // grab a fresh scope + + parseStatement(b, null); // initializer + JSFunction e2 = // we need to put the incrementor before the test + new JSFunction(sourceName, parserLine, null); // so we save the test here + if (peekToken() != SEMI) + startExpr(e2, -1); + else + e2.add(parserLine, b.LITERAL, Boolean.TRUE); // handle the for(foo;;foo) case + consume(SEMI); + if (label != null) b.add(parserLine, LABEL, label); + b.add(parserLine, LOOP); + int size2 = b.size; + + b.add(parserLine, JT, JS.ZERO); // if we're on the first iteration, jump over the incrementor + int size = b.size; + if (peekToken() != RP) { // do the increment thing + startExpr(b, -1); + b.add(parserLine, POP); + } + b.set(size - 1, JS.N(b.size - size + 1)); + consume(RP); + + b.paste(e2); // ok, *now* test if we're done yet + b.add(parserLine, JT, JS.N(2)); // break out if we don't meet the test + b.add(parserLine, BREAK); + parseStatement(b, null); + b.add(parserLine, CONTINUE); // if we fall out the bottom, CONTINUE + b.set(size2 - 1, JS.N(b.size - size2 + 1)); // end of the loop + + b.add(parserLine, OLDSCOPE); // get our scope back + } + break; + } + + case NAME: { // either a label or an identifier; this is the one place we're not LL(1) + String possiblyTheLabel = string; + if (peekToken() == COLON) { // label + consume(COLON); + parseStatement(b, possiblyTheLabel); + break; + } else { // expression + pushBackToken(NAME, possiblyTheLabel); + startExpr(b, -1); + b.add(parserLine, POP); + if ((mostRecentlyReadToken != RC || peekToken() == SEMI) && peekToken() != -1 && mostRecentlyReadToken != SEMI) consume(SEMI); + break; + } + } + + case SEMI: return; // yep, the null statement is valid + + case LC: { // blocks are statements too + pushBackToken(); + b.add(parserLine, NEWSCOPE); + parseBlock(b, label); + b.add(parserLine, OLDSCOPE); + break; + } + + default: { // hope that it's an expression + pushBackToken(); + startExpr(b, -1); + b.add(parserLine, POP); + if ((mostRecentlyReadToken != RC || peekToken() == SEMI) && peekToken() != -1 && mostRecentlyReadToken != SEMI) consume(SEMI); + break; + } + } } + + + // ParserException ////////////////////////////////////////////////////////////////////// + private IOException pe(String s) { return new IOException(sourceName + ":" + parserLine + " " + s); } }