Re: [xmlvm-users] memory bug found

[Joshua M. <xm...@me...>]

Hi Panayotis,

I am aware of this issue and have been working on a fix for it.
Sadly, the fix is non trivial though I am making (slow) progress on
it.

For now, the best workaround is to comment out
// new ExcessRetainsOptimization().Process(curRun.allCodePaths, curRun.beenTo,
//		codeElement);

in ReferenceCounting.java.

Since this bug is affecting people, I will talk to Arno about checking
in that workaround until I can actually implement a correct fix.

Sorry you encountered this...

-- Joshua Melcon



On Mon, May 24, 2010 at 6:28 AM, Panayotis Katsaloulis
<pan...@pa...> wrote:
> I think I found a serious memory bug with current implementation of xmlvm.
>
> Here is the minimum java code to reproduce this bug:
>
> package xmlvm;
> import org.xmlvm.iphone.UIApplication;
> import org.xmlvm.iphone.UIView;
>
> public class Main extends UIApplication {
>    public void applicationDidFinishLaunching(UIApplication app) {
>        testView view = new testView();
>    }
>    public static void main(String[] args) {
>        UIApplication.main(args, Main.class);
>    }
>    public class testView extends UIView {
>        public testView() {
>            System.out.println(getFrame().size);
>        }
>    }
> }
>
>
> the initialization code of testView produces this code:
>
> - (void) __init_xmlvm_Main_testView___xmlvm_Main :(xmlvm_Main*)n1
> {
>    XMLVMElem _rtmp;
>    XMLVMElem _r0;
>    XMLVMElem _r1;
>    XMLVMElem _r2;
>    XMLVMElem _r3;
>    _r2.o = self;
>    _r3.o = n1;
>    _r0.o = JAVA_NULL;
>    _r1.o = JAVA_NULL;
>    [_r3.o retain];
>    [((xmlvm_Main_testView*) _r2.o)->this_0_xmlvm_Main release];
>    ((xmlvm_Main_testView*) _r2.o)->this_0_xmlvm_Main = _r3.o;
>    [((org_xmlvm_iphone_UIView*) _r2.o) __init_org_xmlvm_iphone_UIView__];
>    _r0.o = [java_lang_System _GET_out];
>    _r1.o = [((xmlvm_Main_testView*) _r2.o) getFrame__];
>    _rtmp.o =  _r1.o;
>    _r1.o = ((org_xmlvm_iphone_CGRect*) _r1.o)->size_org_xmlvm_iphone_CGSize;
>    [_rtmp.o release];
>    _rtmp.o = JAVA_NULL;
>    [((java_io_PrintStream*) _r0.o) println___java_lang_Object:_r1.o];
>    return;
> }
>
>
>
> Now, let me copy & paste the code and add some comments, to describe it:
> // perform getFrame()
>    _r1.o = [((xmlvm_Main_testView*) _r2.o) getFrame__];
>
> // keep result of getFrame() into temporary variable _rtmp
>    _rtmp.o =  _r1.o;
>
> // frame's size is received from the structure
>    _r1.o = ((org_xmlvm_iphone_CGRect*) _r1.o)->size_org_xmlvm_iphone_CGSize;
>
> // **release frame ** - here is the bug
>    [_rtmp.o release];
>    _rtmp.o = JAVA_NULL;
>
> // Use CGSize data, although the structure that was encapsulating it (the frame) is already freed!
>    [((java_io_PrintStream*) _r0.o) println___java_lang_Object:_r1.o];
>
>
>
> It seems that objects whose member items are used later on, should also be tracked and released ONLY after the inside items are not used any more!
> Or, even better, every time a reference with the -> operator is made, also retain this object and properly release it afterwards.
>
> What others do say?
> ------------------------------------------------------------------------------
>
> _______________________________________________
> xmlvm-users mailing list
> xml...@li...
> https://lists.sourceforge.net/lists/listinfo/xmlvm-users
>