> Trying to fake two kinds of pointers when we only have one doesn't seem
> wise to me.
>
> A possible solution is to use function descriptors.
Both m32c (my local copy at least) and xstormy use 16 bit thunks to
reference the address of a function outside the first 64k of address
space.
In this case, it isn't good enough.
The particular case that *needs* a wider pointer is setting up
the reset vectors in ROM:
long rvec = (long)&start __attribute__((section(".resetvect")));
At the point where you need this address, you haven't had a chance to
copy the RAM into the lower 64k, so a thunk won't help. (yes,
assembler is one solution, but that doesn't stop the users from
complaining ;).
Note that my complaint about the lack of endian support in that code
still stands ;-)
