org.mozilla.javascript.Node Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of js Show documentation
Show all versions of js Show documentation
Rhino is an open-source implementation of JavaScript written entirely in Java. It is typically embedded into Java applications to provide scripting to end users.
/* -*- Mode: java; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
*
* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0
*
* The contents of this file are subject to the Mozilla Public License Version
* 1.1 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
* http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis,
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
* for the specific language governing rights and limitations under the
* License.
*
* The Original Code is Rhino code, released
* May 6, 1999.
*
* The Initial Developer of the Original Code is
* Netscape Communications Corporation.
* Portions created by the Initial Developer are Copyright (C) 1997-1999
* the Initial Developer. All Rights Reserved.
*
* Contributor(s):
* Norris Boyd
* Roshan James
* Roger Lawrence
* Mike McCabe
*
* Alternatively, the contents of this file may be used under the terms of
* the GNU General Public License Version 2 or later (the "GPL"), in which
* case the provisions of the GPL are applicable instead of those above. If
* you wish to allow use of your version of this file only under the terms of
* the GPL and not to allow others to use your version of this file under the
* MPL, indicate your decision by deleting the provisions above and replacing
* them with the notice and other provisions required by the GPL. If you do
* not delete the provisions above, a recipient may use your version of this
* file under either the MPL or the GPL.
*
* ***** END LICENSE BLOCK ***** */
package org.mozilla.javascript;
/**
* This class implements the root of the intermediate representation.
*
* @author Norris Boyd
* @author Mike McCabe
*/
public class Node
{
public static final int
FUNCTION_PROP = 1,
LOCAL_PROP = 2,
LOCAL_BLOCK_PROP = 3,
REGEXP_PROP = 4,
CASEARRAY_PROP = 5,
/*
the following properties are defined and manipulated by the
optimizer -
TARGETBLOCK_PROP - the block referenced by a branch node
VARIABLE_PROP - the variable referenced by a BIND or NAME node
ISNUMBER_PROP - this node generates code on Number children and
delivers a Number result (as opposed to Objects)
DIRECTCALL_PROP - this call node should emit code to test the function
object against the known class and call diret if it
matches.
*/
TARGETBLOCK_PROP = 6,
VARIABLE_PROP = 7,
ISNUMBER_PROP = 8,
DIRECTCALL_PROP = 9,
SPECIALCALL_PROP = 10,
SKIP_INDEXES_PROP = 11, // array of skipped indexes of array literal
OBJECT_IDS_PROP = 12, // array of properties for object literal
INCRDECR_PROP = 13, // pre or post type of increment/decerement
CATCH_SCOPE_PROP = 14, // index of catch scope block in catch
LABEL_ID_PROP = 15, // label id: code generation uses it
MEMBER_TYPE_PROP = 16, // type of element access operation
NAME_PROP = 17, // property name
CONTROL_BLOCK_PROP = 18, // flags a control block that can drop off
PARENTHESIZED_PROP = 19, // expression is parenthesized
LAST_PROP = 19;
// values of ISNUMBER_PROP to specify
// which of the children are Number types
public static final int
BOTH = 0,
LEFT = 1,
RIGHT = 2;
public static final int // values for SPECIALCALL_PROP
NON_SPECIALCALL = 0,
SPECIALCALL_EVAL = 1,
SPECIALCALL_WITH = 2;
public static final int // flags for INCRDECR_PROP
DECR_FLAG = 0x1,
POST_FLAG = 0x2;
public static final int // flags for MEMBER_TYPE_PROP
PROPERTY_FLAG = 0x1, // property access: element is valid name
ATTRIBUTE_FLAG = 0x2, // x.@y or x..@y
DESCENDANTS_FLAG = 0x4; // x..y or x..@i
private static class NumberNode extends Node
{
NumberNode(double number)
{
super(Token.NUMBER);
this.number = number;
}
double number;
}
private static class StringNode extends Node
{
StringNode(int type, String str) {
super(type);
this.str = str;
}
String str;
}
public static class Jump extends Node
{
public Jump(int type)
{
super(type);
}
Jump(int type, int lineno)
{
super(type, lineno);
}
Jump(int type, Node child)
{
super(type, child);
}
Jump(int type, Node child, int lineno)
{
super(type, child, lineno);
}
public final Jump getJumpStatement()
{
if (!(type == Token.BREAK || type == Token.CONTINUE)) Kit.codeBug();
return jumpNode;
}
public final void setJumpStatement(Jump jumpStatement)
{
if (!(type == Token.BREAK || type == Token.CONTINUE)) Kit.codeBug();
if (jumpStatement == null) Kit.codeBug();
if (this.jumpNode != null) Kit.codeBug(); //only once
this.jumpNode = jumpStatement;
}
public final Node getDefault()
{
if (!(type == Token.SWITCH)) Kit.codeBug();
return target2;
}
public final void setDefault(Node defaultTarget)
{
if (!(type == Token.SWITCH)) Kit.codeBug();
if (defaultTarget.type != Token.TARGET) Kit.codeBug();
if (target2 != null) Kit.codeBug(); //only once
target2 = defaultTarget;
}
public final Node getFinally()
{
if (!(type == Token.TRY)) Kit.codeBug();
return target2;
}
public final void setFinally(Node finallyTarget)
{
if (!(type == Token.TRY)) Kit.codeBug();
if (finallyTarget.type != Token.TARGET) Kit.codeBug();
if (target2 != null) Kit.codeBug(); //only once
target2 = finallyTarget;
}
public final Jump getLoop()
{
if (!(type == Token.LABEL)) Kit.codeBug();
return jumpNode;
}
public final void setLoop(Jump loop)
{
if (!(type == Token.LABEL)) Kit.codeBug();
if (loop == null) Kit.codeBug();
if (jumpNode != null) Kit.codeBug(); //only once
jumpNode = loop;
}
public final Node getContinue()
{
if (type != Token.LOOP) Kit.codeBug();
return target2;
}
public final void setContinue(Node continueTarget)
{
if (type != Token.LOOP) Kit.codeBug();
if (continueTarget.type != Token.TARGET) Kit.codeBug();
if (target2 != null) Kit.codeBug(); //only once
target2 = continueTarget;
}
public Node target;
private Node target2;
private Jump jumpNode;
}
private static class PropListItem
{
PropListItem next;
int type;
int intValue;
Object objectValue;
}
public Node(int nodeType) {
type = nodeType;
}
public Node(int nodeType, Node child) {
type = nodeType;
first = last = child;
child.next = null;
}
public Node(int nodeType, Node left, Node right) {
type = nodeType;
first = left;
last = right;
left.next = right;
right.next = null;
}
public Node(int nodeType, Node left, Node mid, Node right) {
type = nodeType;
first = left;
last = right;
left.next = mid;
mid.next = right;
right.next = null;
}
public Node(int nodeType, int line) {
type = nodeType;
lineno = line;
}
public Node(int nodeType, Node child, int line) {
this(nodeType, child);
lineno = line;
}
public Node(int nodeType, Node left, Node right, int line) {
this(nodeType, left, right);
lineno = line;
}
public Node(int nodeType, Node left, Node mid, Node right, int line) {
this(nodeType, left, mid, right);
lineno = line;
}
public static Node newNumber(double number) {
return new NumberNode(number);
}
public static Node newString(String str) {
return new StringNode(Token.STRING, str);
}
public static Node newString(int type, String str) {
return new StringNode(type, str);
}
public int getType() {
return type;
}
public void setType(int type) {
this.type = type;
}
public boolean hasChildren() {
return first != null;
}
public Node getFirstChild() {
return first;
}
public Node getLastChild() {
return last;
}
public Node getNext() {
return next;
}
public Node getChildBefore(Node child) {
if (child == first)
return null;
Node n = first;
while (n.next != child) {
n = n.next;
if (n == null)
throw new RuntimeException("node is not a child");
}
return n;
}
public Node getLastSibling() {
Node n = this;
while (n.next != null) {
n = n.next;
}
return n;
}
public void addChildToFront(Node child) {
child.next = first;
first = child;
if (last == null) {
last = child;
}
}
public void addChildToBack(Node child) {
child.next = null;
if (last == null) {
first = last = child;
return;
}
last.next = child;
last = child;
}
public void addChildrenToFront(Node children) {
Node lastSib = children.getLastSibling();
lastSib.next = first;
first = children;
if (last == null) {
last = lastSib;
}
}
public void addChildrenToBack(Node children) {
if (last != null) {
last.next = children;
}
last = children.getLastSibling();
if (first == null) {
first = children;
}
}
/**
* Add 'child' before 'node'.
*/
public void addChildBefore(Node newChild, Node node) {
if (newChild.next != null)
throw new RuntimeException(
"newChild had siblings in addChildBefore");
if (first == node) {
newChild.next = first;
first = newChild;
return;
}
Node prev = getChildBefore(node);
addChildAfter(newChild, prev);
}
/**
* Add 'child' after 'node'.
*/
public void addChildAfter(Node newChild, Node node) {
if (newChild.next != null)
throw new RuntimeException(
"newChild had siblings in addChildAfter");
newChild.next = node.next;
node.next = newChild;
if (last == node)
last = newChild;
}
public void removeChild(Node child) {
Node prev = getChildBefore(child);
if (prev == null)
first = first.next;
else
prev.next = child.next;
if (child == last) last = prev;
child.next = null;
}
public void replaceChild(Node child, Node newChild) {
newChild.next = child.next;
if (child == first) {
first = newChild;
} else {
Node prev = getChildBefore(child);
prev.next = newChild;
}
if (child == last)
last = newChild;
child.next = null;
}
public void replaceChildAfter(Node prevChild, Node newChild) {
Node child = prevChild.next;
newChild.next = child.next;
prevChild.next = newChild;
if (child == last)
last = newChild;
child.next = null;
}
private static final String propToString(int propType)
{
if (Token.printTrees) {
// If Context.printTrees is false, the compiler
// can remove all these strings.
switch (propType) {
case FUNCTION_PROP: return "function";
case LOCAL_PROP: return "local";
case LOCAL_BLOCK_PROP: return "local_block";
case REGEXP_PROP: return "regexp";
case CASEARRAY_PROP: return "casearray";
case TARGETBLOCK_PROP: return "targetblock";
case VARIABLE_PROP: return "variable";
case ISNUMBER_PROP: return "isnumber";
case DIRECTCALL_PROP: return "directcall";
case SPECIALCALL_PROP: return "specialcall";
case SKIP_INDEXES_PROP: return "skip_indexes";
case OBJECT_IDS_PROP: return "object_ids_prop";
case INCRDECR_PROP: return "incrdecr_prop";
case CATCH_SCOPE_PROP: return "catch_scope_prop";
case LABEL_ID_PROP: return "label_id_prop";
case MEMBER_TYPE_PROP: return "member_type_prop";
case NAME_PROP: return "name_prop";
case CONTROL_BLOCK_PROP: return "control_block_prop";
case PARENTHESIZED_PROP: return "parenthesized_prop";
default: Kit.codeBug();
}
}
return null;
}
private PropListItem lookupProperty(int propType)
{
PropListItem x = propListHead;
while (x != null && propType != x.type) {
x = x.next;
}
return x;
}
private PropListItem ensureProperty(int propType)
{
PropListItem item = lookupProperty(propType);
if (item == null) {
item = new PropListItem();
item.type = propType;
item.next = propListHead;
propListHead = item;
}
return item;
}
public void removeProp(int propType)
{
PropListItem x = propListHead;
if (x != null) {
PropListItem prev = null;
while (x.type != propType) {
prev = x;
x = x.next;
if (x == null) { return; }
}
if (prev == null) {
propListHead = x.next;
} else {
prev.next = x.next;
}
}
}
public Object getProp(int propType)
{
PropListItem item = lookupProperty(propType);
if (item == null) { return null; }
return item.objectValue;
}
public int getIntProp(int propType, int defaultValue)
{
PropListItem item = lookupProperty(propType);
if (item == null) { return defaultValue; }
return item.intValue;
}
public int getExistingIntProp(int propType)
{
PropListItem item = lookupProperty(propType);
if (item == null) { Kit.codeBug(); }
return item.intValue;
}
public void putProp(int propType, Object prop)
{
if (prop == null) {
removeProp(propType);
} else {
PropListItem item = ensureProperty(propType);
item.objectValue = prop;
}
}
public void putIntProp(int propType, int prop)
{
PropListItem item = ensureProperty(propType);
item.intValue = prop;
}
public int getLineno() {
return lineno;
}
/** Can only be called when getType() == Token.NUMBER */
public final double getDouble() {
return ((NumberNode)this).number;
}
public final void setDouble(double number) {
((NumberNode)this).number = number;
}
/** Can only be called when node has String context. */
public final String getString() {
return ((StringNode)this).str;
}
/** Can only be called when node has String context. */
public final void setString(String s) {
if (s == null) Kit.codeBug();
((StringNode)this).str = s;
}
public static Node newTarget()
{
return new Node(Token.TARGET);
}
public final int labelId()
{
if (type != Token.TARGET) Kit.codeBug();
return getIntProp(LABEL_ID_PROP, -1);
}
public void labelId(int labelId)
{
if (type != Token.TARGET) Kit.codeBug();
putIntProp(LABEL_ID_PROP, labelId);
}
/**
* Does consistent-return analysis on the function body when strict mode is
* enabled.
*
* function (x) { return (x+1) }
* is ok, but
* function (x) { if (x < 0) return (x+1); }
* is not becuase the function can potentially return a value when the
* condition is satisfied and if not, the function does not explicitly
* return value.
*
* This extends to checking mismatches such as "return" and "return "
* used in the same function. Warnings are not emitted if inconsistent
* returns exist in code that can be statically shown to be unreachable.
* Ex.
* function (x) { while (true) { ... if (..) { return value } ... } }
* emits no warning. However if the loop had a break statement, then a
* warning would be emitted.
*
* The consistency analysis looks at control structures such as loops, ifs,
* switch, try-catch-finally blocks, examines the reachable code paths and
* warns the user about an inconsistent set of termination possibilities.
*
* Caveat: Since the parser flattens many control structures into almost
* straight-line code with gotos, it makes such analysis hard. Hence this
* analyser is written to taken advantage of patterns of code generated by
* the parser (for loops, try blocks and such) and does not do a full
* control flow analysis of the gotos and break/continue statements.
* Future changes to the parser will affect this analysis.
*/
/**
* These flags enumerate the possible ways a statement/function can
* terminate. These flags are used by endCheck() and by the Parser to
* detect inconsistent return usage.
*
* END_UNREACHED is reserved for code paths that are assumed to always be
* able to execute (example: throw, continue)
*
* END_DROPS_OFF indicates if the statement can transfer control to the
* next one. Statement such as return dont. A compound statement may have
* some branch that drops off control to the next statement.
*
* END_RETURNS indicates that the statement can return (without arguments)
* END_RETURNS_VALUE indicates that the statement can return a value.
*
* A compound statement such as
* if (condition) {
* return value;
* }
* Will be detected as (END_DROPS_OFF | END_RETURN_VALUE) by endCheck()
*/
static final int END_UNREACHED = 0;
static final int END_DROPS_OFF = 1;
static final int END_RETURNS = 2;
static final int END_RETURNS_VALUE = 4;
/**
* Checks that every return usage in a function body is consistent with the
* requirements of strict-mode.
* @return true if the function satisfies strict mode requirement.
*/
public boolean hasConsistentReturnUsage()
{
int n = endCheck();
return (n & END_RETURNS_VALUE) == 0 ||
(n & (END_DROPS_OFF|END_RETURNS)) == 0;
}
/**
* Returns in the then and else blocks must be consistent with each other.
* If there is no else block, then the return statement can fall through.
* @return logical OR of END_* flags
*/
private int endCheckIf()
{
Node th, el;
int rv = END_UNREACHED;
th = next;
el = ((Jump)this).target;
rv = th.endCheck();
if (el != null)
rv |= el.endCheck();
else
rv |= END_DROPS_OFF;
return rv;
}
/**
* Consistency of return statements is checked between the case statements.
* If there is no default, then the switch can fall through. If there is a
* default,we check to see if all code paths in the default return or if
* there is a code path that can fall through.
* @return logical OR of END_* flags
*/
private int endCheckSwitch()
{
Node n;
int rv = END_UNREACHED;
// examine the cases
for (n = first.next; n != null; n = n.next)
{
if (n.type == Token.CASE) {
rv |= ((Jump)n).target.endCheck();
} else
break;
}
// we don't care how the cases drop into each other
rv &= ~END_DROPS_OFF;
// examine the default
n = ((Jump)this).getDefault();
if (n != null)
rv |= n.endCheck();
else
rv |= END_DROPS_OFF;
// remove the switch block
rv |= getIntProp(CONTROL_BLOCK_PROP, END_UNREACHED);
return rv;
}
/**
* If the block has a finally, return consistency is checked in the
* finally block. If all code paths in the finally returns, then the
* returns in the try-catch blocks don't matter. If there is a code path
* that does not return or if there is no finally block, the returns
* of the try and catch blocks are checked for mismatch.
* @return logical OR of END_* flags
*/
private int endCheckTry()
{
Node n;
int rv = END_UNREACHED;
// check the finally if it exists
n = ((Jump)this).getFinally();
if(n != null) {
rv = n.next.first.endCheck();
} else {
rv = END_DROPS_OFF;
}
// if the finally block always returns, then none of the returns
// in the try or catch blocks matter
if ((rv & END_DROPS_OFF) != 0) {
rv &= ~END_DROPS_OFF;
// examine the try block
rv |= first.endCheck();
// check each catch block
n = ((Jump)this).target;
if (n != null)
{
// point to the first catch_scope
for (n = n.next.first; n != null; n = n.next.next)
{
// check the block of user code in the catch_scope
rv |= n.next.first.next.first.endCheck();
}
}
}
return rv;
}
/**
* Return statement in the loop body must be consistent. The default
* assumption for any kind of a loop is that it will eventually terminate.
* The only exception is a loop with a constant true condition. Code that
* follows such a loop is examined only if one can statically determine
* that there is a break out of the loop.
* for(<> ; <>; <>) {}
* for(<> in <> ) {}
* while(<>) { }
* do { } while(<>)
* @return logical OR of END_* flags
*/
private int endCheckLoop()
{
Node n;
int rv = END_UNREACHED;
// To find the loop body, we look at the second to last node of the
// loop node, which should be the predicate that the loop should
// satisfy.
// The target of the predicate is the loop-body for all 4 kinds of
// loops.
for (n = first; n.next != last; n = n.next)
/* skip */;
if (n.type != Token.IFEQ)
return END_DROPS_OFF;
// The target's next is the loop body block
rv = ((Jump)n).target.next.endCheck();
// check to see if the loop condition is true
if (n.first.type == Token.TRUE)
rv &= ~END_DROPS_OFF;
// look for effect of breaks
rv |= getIntProp(CONTROL_BLOCK_PROP, END_UNREACHED);
return rv;
}
/**
* A general block of code is examined statement by statement. If any
* statement (even compound ones) returns in all branches, then subsequent
* statements are not examined.
* @return logical OR of END_* flags
*/
private int endCheckBlock()
{
Node n;
int rv = END_DROPS_OFF;
// check each statment and if the statement can continue onto the next
// one, then check the next statement
for (n=first; ((rv & END_DROPS_OFF) != 0) && n != null; n = n.next)
{
rv &= ~END_DROPS_OFF;
rv |= n.endCheck();
}
return rv;
}
/**
* A labelled statement implies that there maybe a break to the label. The
* function processes the labelled statement and then checks the
* CONTROL_BLOCK_PROP property to see if there is ever a break to the
* particular label.
* @return logical OR of END_* flags
*/
private int endCheckLabel()
{
int rv = END_UNREACHED;
rv = next.endCheck();
rv |= getIntProp(CONTROL_BLOCK_PROP, END_UNREACHED);
return rv;
}
/**
* When a break is encountered annotate the statement being broken
* out of by setting its CONTROL_BLOCK_PROP property.
* @return logical OR of END_* flags
*/
private int endCheckBreak()
{
Node n = ((Jump) this).jumpNode;
n.putIntProp(CONTROL_BLOCK_PROP, END_DROPS_OFF);
return END_UNREACHED;
}
/**
* endCheck() examines the body of a function, doing a basic reachability
* analysis and returns a combination of flags END_* flags that indicate
* how the function execution can terminate. These constitute only the
* pessimistic set of termination conditions. It is possible that at
* runtime certain code paths will never be actually taken. Hence this
* analysis will flag errors in cases where there may not be errors.
* @return logical OR of END_* flags
*/
private int endCheck()
{
switch(type)
{
case Token.BREAK:
return endCheckBreak();
case Token.CONTINUE:
case Token.THROW:
return END_UNREACHED;
case Token.RETURN:
if (this.first != null)
return END_RETURNS_VALUE;
else
return END_RETURNS;
case Token.TARGET:
if (next != null)
return next.endCheck();
else
return END_DROPS_OFF;
case Token.LOOP:
return endCheckLoop();
case Token.LOCAL_BLOCK:
case Token.BLOCK:
// there are several special kinds of blocks
if (first == null)
return END_DROPS_OFF;
switch(first.type) {
case Token.LABEL:
return first.endCheckLabel();
case Token.IFNE:
return first.endCheckIf();
case Token.SWITCH:
return first.endCheckSwitch();
case Token.TRY:
return first.endCheckTry();
default:
return endCheckBlock();
}
default:
return END_DROPS_OFF;
}
}
public boolean hasSideEffects()
{
switch (type) {
case Token.EXPR_VOID:
case Token.COMMA:
if (last != null)
return last.hasSideEffects();
else
return true;
case Token.HOOK:
if (first == null ||
first.next == null ||
first.next.next == null)
Kit.codeBug();
return first.next.hasSideEffects() &&
first.next.next.hasSideEffects();
case Token.ERROR: // Avoid cascaded error messages
case Token.EXPR_RESULT:
case Token.ASSIGN:
case Token.ASSIGN_ADD:
case Token.ASSIGN_SUB:
case Token.ASSIGN_MUL:
case Token.ASSIGN_DIV:
case Token.ASSIGN_MOD:
case Token.ASSIGN_BITOR:
case Token.ASSIGN_BITXOR:
case Token.ASSIGN_BITAND:
case Token.ASSIGN_LSH:
case Token.ASSIGN_RSH:
case Token.ASSIGN_URSH:
case Token.ENTERWITH:
case Token.LEAVEWITH:
case Token.RETURN:
case Token.GOTO:
case Token.IFEQ:
case Token.IFNE:
case Token.NEW:
case Token.DELPROP:
case Token.SETNAME:
case Token.SETPROP:
case Token.SETELEM:
case Token.CALL:
case Token.THROW:
case Token.RETHROW:
case Token.SETVAR:
case Token.CATCH_SCOPE:
case Token.RETURN_RESULT:
case Token.SET_REF:
case Token.DEL_REF:
case Token.REF_CALL:
case Token.TRY:
case Token.SEMI:
case Token.INC:
case Token.DEC:
case Token.EXPORT:
case Token.IMPORT:
case Token.IF:
case Token.ELSE:
case Token.SWITCH:
case Token.WHILE:
case Token.DO:
case Token.FOR:
case Token.BREAK:
case Token.CONTINUE:
case Token.VAR:
case Token.CONST:
case Token.WITH:
case Token.CATCH:
case Token.FINALLY:
case Token.BLOCK:
case Token.LABEL:
case Token.TARGET:
case Token.LOOP:
case Token.JSR:
case Token.SETPROP_OP:
case Token.SETELEM_OP:
case Token.LOCAL_BLOCK:
case Token.SET_REF_OP:
return true;
default:
return false;
}
}
public String toString()
{
if (Token.printTrees) {
StringBuffer sb = new StringBuffer();
toString(new ObjToIntMap(), sb);
return sb.toString();
}
return String.valueOf(type);
}
private void toString(ObjToIntMap printIds, StringBuffer sb)
{
if (Token.printTrees) {
sb.append(Token.name(type));
if (this instanceof StringNode) {
sb.append(' ');
sb.append(getString());
} else if (this instanceof ScriptOrFnNode) {
ScriptOrFnNode sof = (ScriptOrFnNode)this;
if (this instanceof FunctionNode) {
FunctionNode fn = (FunctionNode)this;
sb.append(' ');
sb.append(fn.getFunctionName());
}
sb.append(" [source name: ");
sb.append(sof.getSourceName());
sb.append("] [encoded source length: ");
sb.append(sof.getEncodedSourceEnd()
- sof.getEncodedSourceStart());
sb.append("] [base line: ");
sb.append(sof.getBaseLineno());
sb.append("] [end line: ");
sb.append(sof.getEndLineno());
sb.append(']');
} else if (this instanceof Jump) {
Jump jump = (Jump)this;
if (type == Token.BREAK || type == Token.CONTINUE) {
sb.append(" [label: ");
appendPrintId(jump.getJumpStatement(), printIds, sb);
sb.append(']');
} else if (type == Token.TRY) {
Node catchNode = jump.target;
Node finallyTarget = jump.getFinally();
if (catchNode != null) {
sb.append(" [catch: ");
appendPrintId(catchNode, printIds, sb);
sb.append(']');
}
if (finallyTarget != null) {
sb.append(" [finally: ");
appendPrintId(finallyTarget, printIds, sb);
sb.append(']');
}
} else if (type == Token.LABEL || type == Token.LOOP
|| type == Token.SWITCH)
{
sb.append(" [break: ");
appendPrintId(jump.target, printIds, sb);
sb.append(']');
if (type == Token.LOOP) {
sb.append(" [continue: ");
appendPrintId(jump.getContinue(), printIds, sb);
sb.append(']');
}
} else {
sb.append(" [target: ");
appendPrintId(jump.target, printIds, sb);
sb.append(']');
}
} else if (type == Token.NUMBER) {
sb.append(' ');
sb.append(getDouble());
} else if (type == Token.TARGET) {
sb.append(' ');
appendPrintId(this, printIds, sb);
}
if (lineno != -1) {
sb.append(' ');
sb.append(lineno);
}
for (PropListItem x = propListHead; x != null; x = x.next) {
int type = x.type;
sb.append(" [");
sb.append(propToString(type));
sb.append(": ");
String value;
switch (type) {
case TARGETBLOCK_PROP : // can't add this as it recurses
value = "target block property";
break;
case LOCAL_BLOCK_PROP : // can't add this as it is dull
value = "last local block";
break;
case ISNUMBER_PROP:
switch (x.intValue) {
case BOTH:
value = "both";
break;
case RIGHT:
value = "right";
break;
case LEFT:
value = "left";
break;
default:
throw Kit.codeBug();
}
break;
case SPECIALCALL_PROP:
switch (x.intValue) {
case SPECIALCALL_EVAL:
value = "eval";
break;
case SPECIALCALL_WITH:
value = "with";
break;
default:
// NON_SPECIALCALL should not be stored
throw Kit.codeBug();
}
break;
default :
Object obj = x.objectValue;
if (obj != null) {
value = obj.toString();
} else {
value = String.valueOf(x.intValue);
}
break;
}
sb.append(value);
sb.append(']');
}
}
}
public String toStringTree(ScriptOrFnNode treeTop) {
if (Token.printTrees) {
StringBuffer sb = new StringBuffer();
toStringTreeHelper(treeTop, this, null, 0, sb);
return sb.toString();
}
return null;
}
private static void toStringTreeHelper(ScriptOrFnNode treeTop, Node n,
ObjToIntMap printIds,
int level, StringBuffer sb)
{
if (Token.printTrees) {
if (printIds == null) {
printIds = new ObjToIntMap();
generatePrintIds(treeTop, printIds);
}
for (int i = 0; i != level; ++i) {
sb.append(" ");
}
n.toString(printIds, sb);
sb.append('\n');
for (Node cursor = n.getFirstChild(); cursor != null;
cursor = cursor.getNext())
{
if (cursor.getType() == Token.FUNCTION) {
int fnIndex = cursor.getExistingIntProp(Node.FUNCTION_PROP);
FunctionNode fn = treeTop.getFunctionNode(fnIndex);
toStringTreeHelper(fn, fn, null, level + 1, sb);
} else {
toStringTreeHelper(treeTop, cursor, printIds, level + 1, sb);
}
}
}
}
private static void generatePrintIds(Node n, ObjToIntMap map)
{
if (Token.printTrees) {
map.put(n, map.size());
for (Node cursor = n.getFirstChild(); cursor != null;
cursor = cursor.getNext())
{
generatePrintIds(cursor, map);
}
}
}
private static void appendPrintId(Node n, ObjToIntMap printIds,
StringBuffer sb)
{
if (Token.printTrees) {
if (n != null) {
int id = printIds.get(n, -1);
sb.append('#');
if (id != -1) {
sb.append(id + 1);
} else {
sb.append("");
}
}
}
}
int type; // type of the node; Token.NAME for example
Node next; // next sibling
private Node first; // first element of a linked list of children
private Node last; // last element of a linked list of children
private int lineno = -1; // encapsulated int data; depends on type
/**
* Linked list of properties. Since vast majority of nodes would have
* no more then 2 properties, linked list saves memory and provides
* fast lookup. If this does not holds, propListHead can be replaced
* by UintMap.
*/
private PropListItem propListHead;
}