Problem using an external thread with ruby

Hi,

I have an problem ....
I create an external thread with pthread_create and
create new ruby objects in this thread. it seems
to work until the GC is running.

#0 mark_locations_array (objspace=0x603d30, th=0x6039b0) at ../gc.c:1315
#1 gc_mark_locations (objspace=0x603d30, th=0x6039b0) at ../gc.c:1331
#2 mark_current_machine_context (objspace=0x603d30, th=0x6039b0) at
../gc.c:2128
#3 0x00007ffff7a2b665 in garbage_collect (objspace=0x603d30) at
../gc.c:2177
#4 0x00007ffff7a2c247 in rb_newobj_from_heap () at ../gc.c:1014
#5 rb_newobj () at ../gc.c:1097
#6 0x00007ffff7a2c306 in rb_data_typed_object_alloc (klass=6638120,
datap=0x808040, type=0x7ffff7dbf200) at ../gc.c:1132
#7 0x00007ffff7a1c5a8 in mnew (klass=6363200, obj=6319120, id=<value
optimized out>, mclass=6638120, scope=0) at ../proc.c:967
#8 0x00007ffff7b2c3ad in vm_call_cfunc (th=0x6039b0,
cfp=0x7ffff7fd7e00, num=<value optimized out>, blockptr=<value optimized

,

    flag=<value optimized out>, id=<value optimized out>, me=0x6ffdb0,
recv=6319120) at ../vm_insnhelper.c:401
#9 vm_call_method (th=0x6039b0, cfp=0x7ffff7fd7e00, num=<value
optimized out>, blockptr=<value optimized out>,
    flag=<value optimized out>, id=<value optimized out>, me=0x6ffdb0,
recv=6319120) at ../vm_insnhelper.c:523
#10 0x00007ffff7b20fc8 in vm_exec_core (th=<value optimized out>,
th=<value optimized out>) at insns.def:1006
#11 0x00007ffff7b26951 in vm_exec (th=0x6039b0) at ../vm.c:1145
#12 0x00007ffff7b27c54 in vm_call0 (th=0x6039b0, recv=6319120, id=12565,
argc=0, argv=0x0, me=<value optimized out>) at ../vm_eval.c:66
#13 0x00007ffff7a1bedb in rb_method_call (argc=0, argv=0x0,
method=<value optimized out>) at ../proc.c:1421
#14 0x00007ffff7a174cb in rb_rescue2 (b_proc=0x7ffff6565ca0
<ProcCallMethod>, data1=6319000, r_proc=0x7ffff6565cb0 <sRescueError>,

the debugger display an extremely high stack-end in the thread struct:

machine_stack_start -> 8
machine_stack_end -> 140737348020529
machine_stack_maxsize -> 7340032

- -> well it seems that a "initialization" have to run first
- -> is a hint available?

thanks for help, Andreas Otto