[Kde-java] Possible KProgressTest memory leak using GCJ

[Richard Dale]

On Monday 08 March 2004 13:27, Richard Dale wrote:
> On Tuesday 02 March 2004 11:47, Daniel Hanlon wrote:
> > Then, for example, the KProgressTest.java example compiles with:
> >
> > gcj KProgressTest.java --main=KProgressTest -o kprogresstest -lkoalabind
> > -I<path>/qtjava/javalib -I <path>/kdejava/koala
> >
> > Whilst the resulting executable worked fine, there was clearly a problem
> > somewhere (probably in libgcj?) as when I left it running for a day or
> > so, it ended up using 300 odd MB of RAM....
>
> I've just been experimenting with running the KProgressTest example with
> both the jdk 1.3 interpreter, and compiled with gcj:
>
>   PID USER     PRI  NI  SIZE  RSS SHARE STAT %CPU %MEM   TIME COMMAND
> 17796 duke       9   0 49928  48M 34588 S     0.0 12.8   0:03 kprogress
>  1250 duke      10   0 43584  34M 31476 S     0.1  9.1  12:13 kdeinit
> 17799 duke       9   0 26752  26M 21408 S     0.0  6.9   0:02 java
>
> But neither of them seem to leak for me - I'm using gcj 3.2.2 on PowerPC:
>
> baldhead duke 1013% gcj --version
> gcj (GCC) 3.2.2 20030217 (Yellow Dog Linux 3.0 3.2.2-2a)
>
> The compiled version seems to gain a couple of Mb in the 5 mins when it
> first starts, and then settles down and the memory usage stays constant.
Thats wrong!

17796 duke       9   0 51128  49M 34596 S     0.0 13.1   0:37 kprogress

After 5 hours or so, kprogress has grown from 49.928 Mb to 51.128 Mb, which 
isn't the same rate of memory leak that Daniel had - 300 Mb in 3 days is 
about 4 Mb per hour. But it is a bug somewhere - I think it's something in 
the gcj JNI method invocation, rather than anything to do with this koala 
java code:

      public void timerEvent(QTimerEvent timer) {

         if (direction == fwd)  {
            if (Progress.progress() == Progress.totalSteps())
                direction = back;
            else
                Progress.advance(1);
         }
         else {
            if (Progress.progress() == 0)  { /*Progress.minValue()*/
               direction = fwd;
            }
            else
               Progress.advance(-1);
        }
      }
   }

As long as the QTimerEvent instance 'timer' is garbage collected after the 
event handler, then no other java instances are being created to add to the 
memory occupancy.

-- Richard