08:15
Okay so SummerCoat runs this code, but RatufaCoat crashes. On x86 it runs okay until it tries to intern a string and on PowerPC is breaks after garbage collection on a return.
08:32
Okay PowerPC runs fine, so this means it is a byte swapping issue somewhere.
08:37
Crashes on 32-bit and 64-bit x86 so definitely an endianess issue.
09:38
I wonder if I write a UTF somewhere.
14:23
Okay so there is an actual write to ROM, I wonder if it happens on big endian as well?
14:24
Okay it does, just does not lead to a crash.
14:44
Okay so what is being written to ROM ends up being the vtable. So what happens is that our vtable entry is read in. Then we do a ref count on it when we are done using it??
14:48
Okay so this is an issue with INVOKEVIRTUAL, it seems that is is uncounting
volatile registers. And looking at reference pushing I exactly see why. I am
using a fixed base for values instead of the volatile get.
18:03
I am going to add a very basic debug print which just takes two characters and a number.
18:36
I believe returned values need to be ref-counted. Why? Because even if we pass
an object to a method that is this, if it returns this then we will drop a
count. And that is basically what the initial case for jvmLoadString().
19:14
So a garbage collection is happening. I do know the byte[] gets GCed but it
seems the string is being GCed too?
19:21
Okay so it seems that jvmLoadString() is returning 1.
19:25
I see no issue from the caller side, so it has to be from jvmLoadString().
19:29
Maybe the copy from return is incorrect?
19:31
Okay this sequence tells me that we do have an interned string returned but it ends up returning the value 1 instead of the value we want to return. Which is a bit strange. I know previously I used return registers as temporary values, so is such a thing still happening?
NA 00104974
NE 0010499c
NA 001049c4
Ls 0010499c
GC 00104974
Ga 00000001
ST 00000001
NA 001049f0
NE 00104974
NA 00104a34
Ls 00104974
GC 001049f0
Ga 00000001
NA 00104974
Nf 00000001
Na 00104974
19:39
I believe I have an idea of what is happening. I am copying the value to the
return register, but before I actually return from the method the code
generator generates code which can run the garbage collector. And the return
register is not valid after and invoke, it must immedietly return. So I do
know that the GC is being called to cleanup the byte[] object and then some
kind of code in there causes 1 to be set to the return register. So I need
to copy the value much later and after all the cleanup has happened.
20:13
So now I have another issue where it seems that the values passed to the object array are exactly the same index or similar. Or the access of the array is wrong? But the wrong access is unlikely maybe?