So I think this is the construct that is broken right now:
public Object clone() {
try {
return super.clone();
} catch (CloneNotSupportException cnse) {
throw new RuntimeException(cnse);
}
}
Correct me if I'm wrong (I haven't studied the bytecode for this
scenario), but I thought the problem was that the exception handler we
introduce has the same range as the existing one, and never ends up
getting called -- thus never performing a MONITOREXIT on the resolve lock,
and thus tripping the method level accounting for monitors?
It's hard to precisely express with source code, but I imagine our current
transformation would end up looking something like this:
public Object clone() {
TCObject tco;
try {
<T> returnValue;
<T> toClone = this;
tco = (toClone instanceof Manageable) ?
((Manageable)toClone).__tc_managed() :
null;
if (tco != null) {
MONITORENTER(tco.getResolveLock);
tco.resolveAllRefs();
returnValue = Util.fixTC(..);
MONITOREXIT(tco.getResolveLock());
} else {
returnValue = toClone.clone();
}
return returnValue;
} catch (CloneNotSupportException cnse) {
throw new RuntimeException(cnse);
} catch (Throwable t) {
MONITOREXIT(tco.getResolveLock());
throw t;
}
}
That code is all kinds of wrong (minimally in how it handles the resolve
lock monitor)
I'm probably missing something, but why can't we just put our processing
in a separate method and avoid any complications with existing exception
handlers? The code end up looking something like this then:
public Object clone() {
try {
return __tc_clone();
} catch (CloneNotSupportedException cnse) {
Throw new RuntimeException(cnse);
}
}
private Object __tc_clone() throws CloneNotSupportedException {
TCObject tco = (this instanceof Manageable) ?
((Manageable) this).__tc_managed() :
null;
if (tco != null) {
synchronized (tco.getResolveLock()) {
tco.resolveAllRefs();
return Util.fixTC(this, super.clone());
}
}
return super.clone();
}
The only real difference is that our special logic is self contained.
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:tc-dev-
> [EMAIL PROTECTED] On Behalf Of Alex Miller
> Sent: Thursday, August 23, 2007 3:40 PM
> To: tc-dev
> Subject: Re: [tc-dev] [JIRA] Commented: (DEV-865)
> CloneNotSupportedException got swallowed and replaced by
> IllegalMonitorStateException
>
> Anyone have any thoughts on this?
>
>
_______________________________________________
tc-dev mailing list
[email protected]
http://lists.terracotta.org/mailman/listinfo/tc-dev
