Alpha (yet again, but it works now [?])

Godmar Back gback at
Fri Oct 23 11:42:08 PDT 1998

 Hi Dirk,

> 2) GCTest fails occasionally. 
> 	[rushmore] ./Kaffe GCTest
> 	[22]:  Failure: testObjArray: wanted 4
> 	[21]:  Failure: testObjChain at 85(0x1079229360): id is 85 (should be 85); name is 16 (should be '85').
>    .. etc. Usually, it will run and just work, but these occasional
>    flaws bother me. Anyone have any suggestions on how to debug? Are
>    there module tests for the GC? Was it designed to work on 64-bit
>    systems?

It's possible, but less likely that the 64-bittiness is a problem.
Try to turn preemption off by either using unix-internal threads
or compiling with "EXTRA_CFLAGS=-DDEBUG" and then using -vmdebug
JTHREADNOPREEMPT and see whether the errors still occur sporadically.
If that turns it off, we have a hidden race condition somewhere.
This is the behavior I saw on the arm.  I haven't had the time to
track it down.  One possibility would be to implement a "break"
native method that is called like a panic: then you could use gdb
to find out what the matter is.

> 5) There are a lot of alpha-related bits with the word "jit" in them, but
>    based on the experience of getting the interpreter running, I don't know
>    how much to trust them. Who did the Alpha port and what was the state
>    when it was last used? Was only the JIT being used?

Check the CVS log and the ChangeLog files for who did what.
I think the alpha JIT is out-of-date.  Some patches to get the alpha
intrp working were submitted by Rick Greer this summer.  Check the
mailing list archives.

	- Godmar

More information about the kaffe mailing list