Ada in gcc-3.4.3 on x86_64

[Florian Weimer]

* Dean Kolosiek:

>>That would be unnecessary overhead to achieve something that is not supported
>>in any case.
>
> Speed does not have precedence over correctness.

As explained, GCC is already correct.  Most binding authors already
write wrappers around variadic functions, I think.  This change would
have very little effect as a result.

Furthermore, as Robert said, to implement this correctly for all
targets is non-trivial.  I really don't see why AMD64 should get a
special treatment here.  Of course, for many programmers it's the
first time in years they see a non-x86 architecture, but this isn't
GCC's fault, really.

>>> So, the LRM is saying that the proper approach is to use overloaded
>>> Ada subprograms to call variadic C routines.
>>
>>It is not saying that at all!
>
> No. Get a dictionary. "Can" denotes ability. The word you defined is
> "may".

Dictionary definitions are not relevant.  The ARM is written in
standardese, not English.

> The note is not a suggestion, it is a statement of fact.

But we disagree on the fact that it states.

You can't read this note in isolation.  It's clear from the rest of
appendix B that an implementation may impose arbitrary restrictions on
pragma Import.  GCC does not support importing variadic C functions.