Re: V3 ABI

[Alexandre Petit-Bianco <apbianco at cygnus dot com>]

Tom Tromey writes:

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

Would we only change java/lang/Object.h?

As I read Hans, this would be safer and easier for the GC (no
modifications required.) This would also be easier for the run-time
(no modifications required as far as the vtable business goes.)

> This is relatively ugly.

It's the safest option we have at this time. I'd like to go with it,
if no one strongly opposes to this option.

Now this is not strictly an ABI issue, but the new C++ compiler forces
us to change our name mangling. Currently, we have code
({class.c,mangle.c}:append_gpp_mangled_*) that mimics what the C++
compiler used to do. The new C++ mangling schemes are different. A few
changes in cp/init.c and cp/mangle.c let me build the run-time with
the new ABI, but of course I get link problems.

I'd like reuse the C++ code for our mangling purposes. If it's
possible, I'd like to create a libcxx.a or libcxxmangling.a that we
would link to jc1.

I'm going to meet Mark Mitchell and Alex Samuel (the author of the
mangling code) tomorrow.  I'm going to suggest that I work with their
help at turning the C++ mangling code into a library. Hopefully it's
feasible.

./A