+// Copyright 2003 Adam Megacz, see the COPYING file for licensing [GPL]
package org.xwt.js;
+
import org.xwt.util.*;
import java.io.*;
-// FIXME: for..in
-public class Parser extends Lexer {
+/**
+ * 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 {
- public static void main(String[] s) throws Exception { System.out.println(new Parser(new InputStreamReader(System.in)).parseExpr()); }
- public Parser(Reader r) throws IOException { super(r); }
- private Parser skipToken() throws IOException { getToken(); return this; }
-
- /** sorta like gcc trees */
- public static class Expr {
- int code = -1;
-
- Expr left = null;
- Expr right = null;
- Expr extra = null;
-
- Expr next = null; // if this expr is part of a list
-
- String string = null;
- Number number = null;
-
- public String toString() { return toString(0); }
-
- public String toString(int indent) {
- String ret = "";
- for(int i=0; i<indent; i++) ret += " ";
- ret += codeToString[code];
- if (code == STRING) ret += " \"" + string + "\"";
- else if (code == NUMBER) ret += " " + number;
- ret += "\n";
- if (left != null) ret += left.toString(indent + 2);
- if (right != null) ret += left.toString(indent + 2);
- if (extra != null) ret += left.toString(indent + 2);
- // FIXME: next
- return ret;
- }
-
- public Expr(String s) { code = STRING; this.string = s; } // an identifier or label
- public Expr(Number n) { code = NUMBER; this.number = n; } // an identifier or label
- public Expr(int code) { this(code, null, null, null); }
- public Expr(int code, Expr left) { this(code, left, null, null); }
- public Expr(int code, Expr left, Expr right) { this(code, left, right, null); }
- public Expr(int code, Expr left, Expr right, Expr extra) { this.left = left; this.right = right; this.extra = extra; this.code = code; }
- }
-
- /** parses a single statement */
- public Expr parseStatement() throws IOException {
- int tok;
- Expr ret;
- switch(tok = peekToken()) {
-
- case LC:
- ret = parseBlock(true);
-
- case THROW: case RETURN: case ASSERT:
- ret = new Expr(ASSERT, skipToken().parseExpr());
-
- case GOTO: case BREAK: case CONTINUE:
- skipToken();
- if (getToken() == NAME)
- ret = new Expr(tok, new Expr(string));
- else if (tok == GOTO)
- throw new Error("goto must be followed by a label");
- else
- ret = new Expr(tok);
-
- default:
- ret = parseExpr();
- }
-
- if (getToken() != SEMI) throw new Error("expected ;");
- return ret;
- }
+ // Constructors //////////////////////////////////////////////////////
+
+ public Parser(Reader r, String sourceName, int line) throws IOException { super(r, sourceName, line); }
- /** a block is either a single statement or a list of statements surrounded by curly braces; all expressions are also statements */
- public Expr parseBlock(boolean requireBraces) throws IOException {
- int tok = peekToken();
- if (requireBraces && tok != LC) throw new Error("expected {");
- if (tok != LC) return parseStatement();
- skipToken();
- Expr head = null;
- Expr tail = null;
- while(peekToken() != RC)
- if (head == null) head = tail = parseStatement(); else tail = tail.next = parseStatement();
- skipToken();
- return new Expr(LC, head);
+ /** for debugging */
+ public static void main(String[] s) throws Exception {
+ JSFunction block = JSFunction.fromReader("stdin", 0, new InputStreamReader(System.in));
+ if (block == null) return;
+ System.out.println(block);
}
+
+ // Statics ////////////////////////////////////////////////////////////
+
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 {
- precedence[COMMA] = 1;
- precedence[ASSIGN] = 2;
- precedence[GT] = precedence[GE] = 3;
- precedence[OR] = precedence[AND] = 4;
- precedence[BITOR] = 5;
- precedence[BITXOR] = 6;
- precedence[BITAND] = 7;
- precedence[EQ] = precedence[NE] = 8;
- precedence[LT] = precedence[LE] = 9;
- precedence[SHEQ] = precedence[SHNE] = 10;
- precedence[LSH] = precedence[RSH] = precedence[URSH] = 11;
- precedence[ADD] = precedence[SUB] = 12;
- precedence[MUL] = precedence[DIV] = precedence[MOD] = 13;
- precedence[BITNOT] = precedence[INSTANCEOF] = 14;
- precedence[INC] = precedence[DEC] = 15;
+ 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 /////////////////////////////////////////////////////////
+
+ /** gets a token and throws an exception if it is not <tt>code</tt> */
+ 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]));
+ }
+ }
+
+ /**
+ * Parse the largest possible expression containing no operators
+ * of precedence below <tt>minPrecedence</tt> and append the
+ * bytecodes for that expression to <tt>appendTo</tt>; 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;
+ }
+ 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);
+ }
+
+
+ /**
+ * Assuming that a complete assignable (lvalue) has just been
+ * parsed and the object and key are on the stack,
+ * <tt>continueExprAfterAssignable</tt> 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_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,
+ * <tt>continueExpr</tt> will attempt to extend this expression by
+ * parsing additional tokens and appending additional bytecodes.
+ *
+ * No operators with precedence less than <tt>minPrecedence</tt>
+ * 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 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;
+ }
+ }
- // called after each parseExpr(); returns null if we can't make the expression any bigger
- public Expr parseExpr() throws IOException { return parseExpr(null, 0); }
- public Expr parseExpr(Expr prefix, int minPrecedence) throws IOException {
- Expr e1 = null;
- Expr e2 = null;
- Expr e3 = null;
- Expr head = null;
- Expr tail = null;
- Expr ret = null;
- int tok = getToken();
-
- if (minPrecedence != 0 && tok < precedence.length && precedence[tok] != 0 && precedence[tok] < minPrecedence)
- return null;
-
- // these case arms match the precedence of operators; each arm is a precedence level.
- switch (tok) {
-
- case COMMA: case ASSIGN: case GT: case GE: case OR: case AND:
- case BITOR: case BITXOR: case BITAND: case EQ: case NE: case LT:
- case LE: case SHEQ: case SHNE: case LSH: case RSH: case URSH:
- case ADD: case SUB: case MUL: case DIV: case MOD:
- return new Expr(tok, prefix, parseExpr(null, precedence[tok]));
-
- case BITNOT: case INSTANCEOF:
- return new Expr(tok, parseExpr(null, precedence[tok]));
-
- // FIXME: this isn't 100% right
- case INC: case DEC:
- return new Expr(tok, prefix, (tok == INC || tok == DEC) ? null : parseExpr());
-
- case LP:
- while(peekToken() != RP) {
- if (head == null) head = tail = parseExpr(); else tail = tail.next = parseExpr();
- tok = getToken();
- if (tok == RP) break;
- if (tok != COMMA) throw new Error("expected comma or right paren");
- }
- return new Expr(LP, prefix, head);
-
- case LB:
- if (prefix != null) {
- // subscripting
- e1 = parseExpr();
- if (getToken() != RB) throw new Error("expected a right brace");
- return new Expr(LB, prefix, e1);
- } else {
- // array ctor
- tok = getToken();
- while(true) {
- if (tok == RB) return new Expr(LB, prefix, head);
- if (head == null) head = tail = parseExpr(); else tail = tail.next = parseExpr();
- tok = getToken();
- if (tok != COMMA && tok != RP) throw new Error("expected right bracket or comma");
- }
- }
-
- case LC:
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- tok = getToken();
- while(true) {
- if (tok == RP) return new Expr(LC, head);
- if (tok != NAME) throw new Error("expecting name");
- Expr name = parseExpr();
- if (tok != COLON) throw new Error("expecting colon");
- e1 = new Expr(COLON, name, parseExpr());
- if (head == null) head = tail = e1; else tail = tail.next = e1;
- tok = getToken();
- if (tok != COMMA && tok != RP) throw new Error("expected right curly or comma");
- }
-
- case HOOK:
- e2 = parseExpr();
- if (getToken() != COLON) throw new Error("expected colon to close ?: expression");
- e3 = parseExpr();
- return new Expr(HOOK, prefix, e2, e3);
-
- case SWITCH: {
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- if (getToken() != LP) throw new Error("expected left paren");
- Expr switchExpr = parseExpr();
- if (getToken() != RP) throw new Error("expected left paren");
- if (getToken() != LC) throw new Error("expected left brace");
- Expr firstExpr = null;
- Expr lastExpr = null;
- while(true) {
- if (getToken() != CASE) throw new Error("expected CASE");
- Expr caseExpr = parseExpr();
- if (getToken() != COLON) throw new Error("expected COLON");
- Expr e = new Expr(CASE, caseExpr, parseBlock(false));
- if (lastExpr == null) firstExpr = e;
- else lastExpr.next = e;
- lastExpr = e;
- if (getToken() == RC) return new Expr(SWITCH, switchExpr, firstExpr);
- }
- }
-
- case FUNCTION: {
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- if (getToken() != LP) throw new Error("function keyword must be followed by a left paren");
- Expr formalArgs = null, cur = null;
- tok = getToken();
- while(tok != RP) {
- if (tok != NAME) throw new Error("expected a variable name");
- if (cur == null) { formalArgs = cur = new Expr(string); }
- else { cur.next = new Expr(string); cur = cur.next; }
- tok = getToken();
- if (tok == RP) break;
- if (tok != COMMA) throw new Error("function argument list must consist of alternating NAMEs and COMMAs");
- tok = getToken();
- }
- return new Expr(tok, formalArgs, parseBlock(true));
- }
-
- case STRING:
- return new Expr(string);
-
- case NUMBER:
- return new Expr(number);
-
- case VAR:
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- while(true) {
- if (getToken() != NAME) throw new Error("variable declarations must start with a variable name");
- Expr name = new Expr(NAME, new Expr(string));
- Expr initVal = null;
- tok = peekToken();
- if (tok == ASSIGN) {
- skipToken();
- initVal = parseExpr();
- tok = peekToken();
- }
- Expr e = new Expr(VAR, name, initVal);
- if (head == null) head = tail = e; else tail = tail.next = e;
- if (tok != COMMA) break;
- skipToken();
- }
- return new Expr(VAR, head);
-
- case NAME:
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- return new Expr(string);
+ continueExpr(b, minPrecedence); // try to continue the expression
+ }
- case TRUE: case FALSE: case NOP:
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- return new Expr(tok);
-
- case TRY: {
- // FIXME: we deliberately allow you to omit braces in catch{}/finally{} if they are single statements...
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- Expr tryBlock = parseBlock(true);
- while ((tok = peekToken()) == CATCH)
- if (head == null) head = tail = parseBlock(false); else tail = tail.next = parseBlock(false);
- if (head == null) throw new Error("try without catch");
- return new Expr(TRY, tryBlock, head, tok == FINALLY ? skipToken().parseBlock(false) : null);
- }
-
- case IF: case WHILE: {
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- if (getToken() != LP) throw new Error("expected left paren");
- Expr parenExpr = parseExpr();
- if (getToken() != RP) throw new Error("expected right paren");
- Expr firstBlock = parseBlock(false);
- if (tok == IF && peekToken() == ELSE) return new Expr(tok, parenExpr, firstBlock, skipToken().parseBlock(false));
- return new Expr(tok, parenExpr, firstBlock);
- }
-
- case FOR:
- // FIXME: for..in
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- if (getToken() != LP) throw new Error("expected left paren");
- e1 = parseStatement();
- e2 = parseStatement();
- e3 = parseStatement(); // FIXME: this guy has to be okay with ending via a )
- if (getToken() != RP) throw new Error("expected right paren");
- throw new Error("not yet implemented");
- //return new Expr(FOR, e1, e2, e3, parseBlock(false));
-
- case DO: {
- if (prefix != null) throw new Error("didn't expect non-null prefix");
- Expr firstBlock = parseBlock(false);
- if (getToken() != WHILE) throw new Error("expecting WHILE");
- if (getToken() != LP) throw new Error("expected left paren");
- Expr whileExpr = parseExpr();
- if (getToken() != RP) throw new Error("expected right paren");
- if (getToken() != SEMI) throw new Error("semicolon");
- return new Expr(DO, firstBlock, whileExpr);
- }
-
- case VOID: case RESERVED:
- throw new Error("reserved word that you shouldn't be using");
-
- case WITH:
- throw new Error("WITH not yet implemented"); // FIXME
-
- default:
- pushBackToken();
- return null;
- }
+ // 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;
}
+ /** 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;
+ }
+ 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);
+ }
+ }
+
+ /** 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); }
+
}
projects / org.ibex.core.git / blobdiff