TITLE: Parenthesis and declarations


PROBLEM: sha@krym.equinox.gen.nz (Alex & Natasha Maclinovsky)

#include <iostream.h>
class C
{
public:
   long l;
   C(long ll) { l = ll; }
}

static char ch = 33;

int main()
{
   C c1((long(ch)));
   C c2(long(ch));      // Problem here

   cout << c1.l << ' ';
   cout << c2.l;        // Compilers give error here
}

last line of main() complaining something like "c2 - not a structure".
It took us few hours to realise where the problem was: the stupid things
instead of definition of <c2> saw a PROTOTYPE  of a function called
c2, which takes one argument of type long and returns an object of class C
by value.

Is it just a coincidence that 3 such a different compilers (which I'm sure
have no common code) fail in exactly the same way?


RESPONSE: clamage@Eng.Sun.COM (Steve Clamage), 11 Mar 95

Not exactly a coincidence, but a long-standing language rule. If a
statment *can be* interpreted as a simple declaration, it *is* interpreted
as a simple declaration. (For more details, see the ARM.)

>   C c2(long(ch));      // Problem here

In a function declaration, redundant parens around dummy argument names
are permitted, so this is equivalent to
	C c2(long ch);
or just
	C c2(long);
Thus, "c2" is a function , not an initialized object.

>   C c1((long(ch)));

This cannot be a declaration of a function "C c1(long)", since the
extra parens around "long(ch)" would not be allowed. Thus it must be a
definition of object "c1" initialized with "ch" converted to long.

You can avoid the problem by writing
	C c2( (long)ch ); // instead of "long(ch)"
And of course in the particular example no cast is needed:
	C c2(ch); // implicit conversion of char to long
But in a different example a cast might be necessary to avoid an ambiguous
call to overloaded constructors.