222635 - Memory leak with with activated debugger support

Status: RESOLVED FIXED

(Rhino Graveyard :: Core, defect)

Description

[Igor Bukanov]

When a debugger is attached to Rhino by Context.setDebugger, then a memory leak
occur as all local variable references are kept in memory. To test, save
Test.java and test.js from the following attachments and then compile Test.java
and run:

java Test test.js

Currently it prints ever increasing memory usage and then exit with OutOfMemory.

This is was originally reported by Dave Russo:

-------- Original Message --------
Subject: 	rhino memory leak(?) when debugger is attached


I think I've run into a memory leak in Rhino and I'd appreciate any 
help/advice you can give. 

I'm using Sun's JVM 1.4.1 on Solaris 5.8 and I'm getting an "out of 
memory" exception when I run the following script inside our application 
that embeds Rhino 1.5R4.1:

    function fread(fileName) 
    { 
        var line =
    "1234567890123456789012345678901234567890123456789012333333333"; 
        var str = "";

        for (var i = 0; i < 1000; i++) { 
            str += line + fileName; 
        }

        return (str); 
    }

    for (var i = 0; i < 2000; i++) fread("package.xml");

I've tried the latest Rhino tip and get the same results. 

We've embedded rhino in a command line application which sets a custom 
"debugger" (ShellDebugger) using:

    cx.setDebugger(new ShellDebugger(), new ShellDebugContext());

and sets optimization level to -1 using:

    cx.setOptimizationLevel(-1);

We're doing this because to get accurate line number information and 
stack trace information when errors occur.  ShellDebugger and 
ShellDebugContext do nothing other than push and pop DebugFrame's from a 
stack maintained in a ShellDebugContext.  When an exception occurs we 
can display the call stack with file and line number information.

It's possible that the leak is in our embedding and *not* in Rhino (if I 
run the script above using "java -cp js.jar 
org.mozilla.javascript.tools.Shell" there is not a problem.)

However, even if I remove *all* processing from ShellDebugger and 
ShellDebugContext in our embedding, I get the memory exception from 
Java.  In addition, if I remove the following block (via the "false &&") 
from org.mozilla.javascript.Interpreter, I don't get the exception:

        if (false && debuggerFrame != null) {
            if (argsDbl != null) {
                args = getArgsArray(args, argsDbl, argShift, argCount);
                argShift = 0;
                argsDbl = null;
            }
            if (idata.itsFunctionType != 0 && !idata.itsNeedsActivation) {
                useActivationVars = true;
                scope = ScriptRuntime.initVarObj(cx, scope, fnOrScript,
                                                 thisObj, args);
            }
            debuggerFrame.onEnter(cx, scope, thisObj, args);
        }

Is there any possibility that Interpreter has a memory leak when the 
debugger is attached?  Any help would be appreciated.

Thanks,

dave

Comment 1

[Igor Bukanov]

Attachment: Java source for a simple Rhino embedding with debugger attached

Comment 2

[Igor Bukanov]

Attachment: test script

Comment 3

[Igor Bukanov]

Attachment: Fix for omg/Interpreter.java: always pop activation record on function exit if debugFrame is not null.

The bug is regression since Rhino 1.5R3 and caused by my code to support
debugging of scripts that were loaded prior the debugger was attached. That
code forces creation of the activation object even for functions that otherwise
optimize it away so the debugger would see local variable. But I forgot to
update the part that releases the activation record and the patch addressed
that.

Comment 4

[Igor Bukanov]

I committed the patch: Dave Russo confirmed that his problem went away.

Comment 5

[Igor Bukanov]

Trageting as resolved against 1.5R5