Re: IBM Acquires Rational Ada

["David Thompson" <david.thompson1@worldnet.att.net>]

Warren W. Gay VE3WWG <ve3wwg@cogeco.ca> wrote :
> Here's another one ;-)
>
> main(){
>    short i = -32768;
>    short j;
>
>    j = -i;
> }
>
> The results are implementation specific (although
> that may have changed with C99 -- I stopped
> caring after I got into Ada ;-)  I have seen
> j=0 in my travels, but don't count on it.
>
Well, in C99 your example as written is illegal, because
"implicit int" function return type, or variable type, is gone;
also, since you neither return'ed from main() or exit()'ed
in C89 the program exit status was undefined but in C99
it is now zero = successful as a special feature^Wkludge.
Not to mention that your entire body is all dead code
that any decent optimizer can completely eliminate.

But those are irrelevant to your point.

The minimum ranges of integer types are unchanged,
except for the addition of new long long types (signed
and unsigned) of at least 64 bits; signed short, and for
that matter int, still need not handle +/-32768, although
two's-complement implementations, the overwhelming majority,
IME always do support -32768 (but not +32768 in 16 bits).

Overflow in any signed integer computation was and is
Undefined Behavior, which means anything at all is
permitted, including but not limited to any value produced
or any signal.  And that initializer is really an expression:
-(unary negate) 32768(positive int or long literal).  Even
though it may well (and should) be evaluated at compile
time, its defined semantics are as for runtime.
But the initializer can't overflow because if 32768 is
representable in int then -32768 is also, and if not
it is evaluated as and must be representable in long.

For the initializer value either way, and the computation
of -i as signed int (due to the integer promotions) if it
does not overflow, the narrowing by assignment (or cast, or
at least scalar initialization) of signed integer types is still
implementation-defined but is now allowed to "raise ...
an I-D signal" instead of producing an I-D result (value).
I-D does mean that the implementation must *document*
what it does, FWTW.  And I've heard of no implementation
that raises a signal here; I'm not sure why the committee
added this new option.  The most recent Rationale draft
I've looked at says nothing on this point.

--
- David.Thompson 1 now at worldnet.att.net