A Rhino user has complained about concurrency problems in the interpreter,
and I think the attached diff fixes some problems that could be caused if
the same interpreted function or script was called simultaneously from
two different threads.
Re: Rhino1.5.R1: problems with multithreaded embedded application.
Date:
Mon, 03 Jul 2000 14:38:56 -0400
From:
Norris Boyd <nboyd@atg.com>
Organization:
Art Technology Group
To:
Fergus Gallagher <Fergus.Gallagher@orbisuk.com>
Newsgroups:
netscape.public.mozilla.jseng
References:
1
You found a bug in Rhino; I wonder if others have been running into the same thing.
The problem is with a class called LazilyLoadedCtor. I wrote this class to reduce the
time
required by initStandardObjects by only creating standard objects when the associated
constructors are first accessed. The problem is that this class was not threadsafe.
I've
made changes to that class, and to ScriptableObject as well. The design of dynamic
properties calling getters and setters (which LazilyLoadedCtor uses) didn't really
allow
any way for the getter/setter to replace itself without a thread hazard. I've now
extended
setters so that they can return a value which replaces the getter/setter to avoid this
problem.
Thanks for finding this problem. There have been a couple of other reports of problems
in
this area, so I hope this will fix them.
The patch follows.
--N
Index: LazilyLoadedCtor.java
===================================================================
RCS file: /cvsroot/mozilla/js/rhino/org/mozilla/javascript/LazilyLoadedCtor.java,v
retrieving revision 1.1
diff -u -r1.1 LazilyLoadedCtor.java
--- LazilyLoadedCtor.java 2000/02/29 21:34:37 1.1
+++ LazilyLoadedCtor.java 2000/07/03 18:31:03
@@ -58,9 +58,12 @@
}
public Object getProperty(ScriptableObject obj) {
- obj.delete(ctorName);
try {
- ScriptableObject.defineClass(obj, Class.forName(className));
+ synchronized (obj) {
+ if (!isReplaced)
+ ScriptableObject.defineClass(obj, Class.forName(className));
+ isReplaced = true;
+ }
}
catch (ClassNotFoundException e) {
throw WrappedException.wrapException(e);
@@ -83,11 +86,14 @@
return obj.get(ctorName, obj);
}
- public void setProperty(ScriptableObject obj, Object val) {
- obj.delete(ctorName);
- obj.put(ctorName, obj, val);
+ public Object setProperty(ScriptableObject obj, Object val) {
+ synchronized (obj) {
+ isReplaced = true;
+ return val;
+ }
}
private String ctorName;
private String className;
+ private boolean isReplaced;
}
Index: ScriptableObject.java
===================================================================
RCS file: /cvsroot/mozilla/js/rhino/org/mozilla/javascript/ScriptableObject.java,v
retrieving revision 1.17
diff -u -r1.17 ScriptableObject.java
--- ScriptableObject.java 2000/03/13 17:12:36 1.17
+++ ScriptableObject.java 2000/07/03 18:31:04
@@ -246,11 +246,21 @@
break;
}
}
- getterSlot.setter.invoke(start, arg);
+ Object v = getterSlot.setter.invoke(start, arg);
+ if (getterSlot.setterReturnsValue) {
+ slots[slotIndex].value = v;
+ if (!(v instanceof Method))
+ slots[slotIndex].flags = 0;
+ }
return;
}
Object[] args = { this, actualArg };
- getterSlot.setter.invoke(getterSlot.delegateTo, args);
+ Object v = getterSlot.setter.invoke(getterSlot.delegateTo, args);
+ if (getterSlot.setterReturnsValue) {
+ slots[slotIndex].value = v;
+ if (!(v instanceof Method))
+ slots[slotIndex].flags = 0;
+ }
return;
}
catch (InvocationTargetException e) {
@@ -1183,6 +1193,7 @@
slot.delegateTo = delegateTo;
slot.getter = getter;
slot.setter = setter;
+ slot.setterReturnsValue = setter != null && setter.getReturnType() !=
Void.TYPE;
slot.value = null;
slot.attributes = (short) attributes;
slot.flags = flags;
@@ -1551,6 +1562,7 @@
Object delegateTo; // OPT: merge with "value"
Method getter;
Method setter;
+ boolean setterReturnsValue;
}
Fergus Gallagher wrote:
> I am having problems getting my head around contexts and scopes and my
> multi-threaded application fall over.
>
> If I set "global=false" the following code used a per-thread
> initStandardObject() and this seems to work. But when I set
> "global=true", the global "parentScope" is used and I get some wierd
> errors.
>
> If I change "__CODE__.slice(0,5)" to
> 1. "__CODE__" - works
> 2. "__CODE__.substring(0,5)" - fails
> 3. "__CODE__.toString()" - works
>
> Any help appreciated.
>
> Fergus
>
> ===== Test.java =========================================
> import java.io.*;
> import org.mozilla.javascript.*;
>
> public class Test implements Runnable {
> private Script script;
> private Scriptable parentScope;
> private String __CODE__="ABCDEFGHIJK";
> private boolean global = true;
> private static Context globalContext = null;
> public Test() throws Exception {
> String js= "java.lang.System.out.println(__CODE__.slice(0,5));";
> globalContext.setCompileFunctionsWithDynamicScope(false);
> parentScope = globalContext.initStandardObjects(null);
> StringReader sr = new StringReader(js);
> script = globalContext.compileReader(parentScope, sr, "(compiled)",
> 1,null);
> }
>
> public void run() {
> try {
> Context context = Context.enter();
> Scriptable threadScope;
> if (global) {
> threadScope = context.newObject(parentScope);
> threadScope.setPrototype(parentScope);
> threadScope.setParentScope(null);
> } else {
> threadScope = context.initStandardObjects(null);
> }
> threadScope.put("__CODE__",threadScope,__CODE__);
> script.exec(context,threadScope);
> }
> catch (Exception e) {
> System.err.println(e.getClass().getName()+":
"+e.getMessage());
> }
> finally {
> Context.exit();
> }
> }
>
> public static void main(String args[]) throws Exception {
> globalContext = Context.enter();
> Test me = new Test();
> int count=10;
> Thread[] threads = new Thread[count];
> for (int i=0; i<count; i++) {
> Thread t = new Thread(me);
> threads[i] = t;
> t.start();
> }
> for (int i=0; i<count; i++) {
> threads[i].join();
> }
> Context.exit();
> }
> }
>
> ==== OUTPUT ===============================================
> ABCDE
> ABCDE
> org.mozilla.javascript.EcmaError: undefined is not a function.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> org.mozilla.javascript.EvaluatorException: Constructor for "String" not
> found.
> ===========================================================
>
> The number and type of exceptions is highly variable from run to run -
> anywhere from 1-9 out of 10.
> The EcmaError in particular only happens occasionally.
>
> --
> Fergus Gallagher Tel: +44 (20) 8 987 0717
> Orbis Fax: +44 (20) 8 742 2649
> The Swan Centre email: Fergus.Gallagher@orbisuk.com
> Fishers Lane Web: http://www.orbisuk.com
> London W4 1RX / UK
Commit the following contributions:
* Andi Vajda's changes to allow embedders to capture the generated bytecode (and thus control
generated class names).
* Marshall Cline's changes to allow embedders to override the default Java object wrapping
behavior
* Kurt Westerfeld's change to handle calling static methods better
Subject:
Odd behaviour on placement of .jar files?!
Date:
Mon, 05 Jun 2000 10:46:08 -0700
From:
John Raykowski <xski@xski.org>
To:
nboyd@atg.com
Hello,
I didn't want to post this directly as a rhino bug 'coz I think it may
be more of a JDK thing, but I thought I'd toss it to you as well.
The goal is to create a JavaScript object that implements a Java
interface. Straightforward enough and the example on the page using
ActionListener works without a hitch. However, when I try to do the
same with my own interface, I get an error message: error instantiating
({0}): class {1} is interface or abstract (coming from
NativeJavaClass.construct).
Here's where it gets a bit strange. Normally, I run with the jar files
in jre/lib/ext. When I remove the rhino files from jre/lib/ext and
reference them explicitly on the commandline with the -cp option, it
works as expected and my script can implement the interface just fine.
Go figure.
Anyhoo, there ya go. Like I said, I think its a JDK issue, but I
thought you'd be interested. The attached zipfile contains a set of
sample code to demonstrate this problem.
Thanks heaps,
-jmr
resulting in a NullPointerException on the following code when run on the MS VM with -opt 9:
var testcases = getTestCases();
function getTestCases() {
return new Boolean(new MyObject(true));
}
function MyObject( value ) {
this.value = value;
this.valueOf = new Function( "return this.value" );
return this;
}
nboyd%atg.com
beard%netscape.com
rogerl%netscape.com