GCJ4 on PPC64: ICE in mark_reference_fields, at java/boehm.c:105

Sun Jan 29 19:46:00 GMT 2006

>>>>> ">" == Wu Zhou <woodzltc@cn.ibm.com> writes:

>> I encontered one problem while using the following script to build 
>> a testcase in 32-bits mode on PPC64, it reports "ICE in 
>> mark_reference_fields, at java/boehm.c:105.". 

Please file this in bugzilla, if you didn't already.
The failing line is this abort:
	 /* If this reference slot appears to overlay a slot we think
	 we already covered, then we are doomed. */
	 if (offset <= *last_view_index)
	 abort ();
This means that this compilation is somehow triggering a "shouldn't
happen" case. It could be a bug somewhere in the compiler, or it
could be that this check is incorrect.
It would help a lot if you could debug jc1 with gdb and show some of
the local state.
>> If building in 64-bits mode, 
>> it only reports an undefined reference: 
>> /tmp/ccggTln9.o(.toc+0x10): undefined reference to `cniinner::child::class$'
>> collect2: ld returned 1 exit status

>> $GCJ -g -I. --main=cniinner.child -o cniinner/a.out \
>> Parent.java Child.java cniinner/native.o

You are passing 'cniinner.child' to --main, but the class is really
called 'cniinner.Child'; i.e., case matters. This is why the linker
is complaining.
Tom