Re: Ada Core Technologies announces GNATCOM

[Robert Dewar <robert_dewar@my-deja.com>]

In article <d41J4.4674$q8.993149@news-east.usenetserver.com>,
  "David Botton" <David@Botton.com> wrote:
> Turbo C did that and many compilers still do.
>
> David Botton

Since I see David sharing the confusion, perhaps an example
is appropriate

  int clobber (struct q m) {
      m.a++;
      return m.a;
  }

called in c passing a value of the struct type does NOT change
the value of the passed variable.

But the RM requires (well its AI, but in practice we have
treated it as a requirement for the purposes of
interoperability) the corresponding Ada function

   function clobber (m : q) return int;

to be interfaced by using call by reference, as a result of
which clobber will destroy the passed parameter.

But you can't unconditionally create a copy, since you may
want the reference semantics. Indeed the thought behind the
bad advice in the RM is that usually the C function will
look like

   int clobber (struct q *m) ....

in which case the call by reference is appropriate.

The effect of C_Pass_By_Copy is to force the appropriate C
calling sequence for a pass by value, i.e. assuming that the
value is reasonably large, create a copy on the calling side,
and then pass a pointer to the copy.

I hope that makes this clear, I am surprised how many people
seem incredibly confused on this issue.

Part of the trouble historically is that C did not allow
structs to be passed by value till later on, so it is not
typical C style to use this feature.

This means that indeed the most common convention for passing
records in C is to pass them by pointer, but although this is
indeed the most common case, it was a goof in the RM to leave
Ada programs with no way of passing records by value.

Remember, call-by-address and passing-in-registers are
implementation techniques.

call-by-reference and call-by-value are high level semantic
concepts.

Don't get the two confused.



Sent via Deja.com http://www.deja.com/
Before you buy.