V3 ABI

Thu Dec 7 16:24:00 GMT 2000

>> Another would be to add dummy virtual functions to Object whose
>> slots would then be abused to hold the extra info. 

Alex> Would we only change java/lang/Object.h?
We would also have to change gcj to know about the dummy virtual
functions in Object.
Alex> As I read Hans, this would be safer and easier for the GC (no
Alex> modifications required.) This would also be easier for the
Alex> run-time (no modifications required as far as the vtable
Alex> business goes.)
Me too.
>> This is relatively ugly.

Alex> It's the safest option we have at this time. I'd like to go with
Alex> it, if no one strongly opposes to this option.
As long as we document it with a lot of comments in Object.h, I have
no objection.
Alex> I'd like reuse the C++ code for our mangling purposes. If it's
Alex> possible, I'd like to create a libcxx.a or libcxxmangling.a that
Alex> we would link to jc1.
This sounds great to me. We used to have a mangling divergence or two
from the C++ compiler. If we shared the implementation this would be
less likely.
Tom