getting the adjusted object pointer

[Jason Merrill]

>>>>> "Richard" == Richard Henderson <rth at redhat.com> writes:

> It is clear that the handler should match.  It is also clear that a
> pointer adjustment is required in initializing the user variable,
> and that this adjusted pointer can be computed by the personality
> routine.

Yes.

> What is not clear is how to get at this adjusted pointer.  If we
> install this adjusted value in gr15, then we pass a effectively 
> random value to __cxa_begin_catch, and it becomes very difficult
> to manage the caughtExceptions stack.

Yes, I think we definitely want to pass the unadjusted pointer.

> The most obvious solution is to have __cxa_begin_catch receive the
> unadjusted pointer (at __cxa_exception+1) and return the cached
> adjustedPtr.  But in the current documents, __cxa_begin_catch 
> returns void.

Your idea would make sense to me.  And looking through the discussions, I
see a note from Christophe which contains

LP1:
        // Pushes the X on the caught stack
        X* xptr = __begin_catch(exc)

so perhaps the documents are just wrong.  Christophe?

Jason