16:33
Okay, so I definitely have a reference counting issue.
16:43
Okay so I added a hit to detect negative uncounts and it definitely looks like
there is a big in jvmGarbageCollect(). This is in the garbage collection of
array elements. Very much appears that for larger arrays I am just uncounting
the first.
16:53
Actually I just realized I need a monitor count.
16:55
Actually no, I can just hide that in the monitor code since it is JVM handled. I think what I can do is make a temporary structure of sorts which stores the information. Just that when something is locked it WILL have thread and count information in it. However problem with this is that it requires two atomic accesses that could differ. It has to have just one.
18:13
I need like SquirrelGrind to detect memory corruption within the VM or similar. I know I have memory corruption, just not sure where.
18:14
It does not happen if GC does nothing. And it does not happen if free is ever called. So it has to be in the free area.
19:19
Okay so the setting the memory block stuff causing free to fail, if free does nothing then it passes. So I think it is corruption after free?
19:29
Okay so, garbage collection works, but free is messed up.
19:30
Okay, so hypothetically I can get very tiny blocks during splits so those have to not be possible.
19:33
Actually there is memory loss, because the chunk wanted size drops and such so this definitely needs a different case!