On Tue, 9 Sep 2003 21:57:23 +0200, Matthieu Herrb wrote:
>Thomas Dickey wrote (in a message from Tuesday 9)
> > On Tue, 9 Sep 2003, Warren Turkal wrote:
> > > -#if NeedFunctionPrototypes
> > > -extern void scale_init(void);
> > > -extern char *nscale(int, int, int, char *);
> >
> > while I'm perfectly aware that "extern" is redundant, there are two
> > things to be said in favor of keeping it:
> >
> > a) it's easy to grep for
> >
> > b) some compilers silently ignore conflicts with a "static" definition of
> > the prototype, but can be persuaded to warn if the extern is explicit.
> > (gcc does this, making it unsuitable as the only compiler to use for
> > testing).
>
>... and a 3rd reason is that 'extern' is not optional for
>variables.
Wrong.
>Most traditional Unix linkers will allow the same variable
>to be declared without 'extern' in multiple object files and merge
>them into only one, but this behaviour is not a feature one should
>rely on. And it fact, at least the Darwin linker treats this as an
>error.
If so, then the Darwin linker is defective. ISO Standard C variables do NOT
require the "extern" modifier in order to have external linkage. A variable
at file scope without an initializer is automatically "extern".
<<<file_1.c>>>
#include <stdio.h>
int xxx;
int main()
{
printf( "xxx is %d\n", xxx );
return 0;
}
<<<file_2.c>>>
int xxx = 8;
Those two files comprise an ANSI/ISO compliant C program which produces a
well-defined result. A compiler which fails to compile this is not ISO
compliant.
--
- Tim Roberts, [EMAIL PROTECTED]
Providenza & Boekelheide, Inc.
_______________________________________________
Devel mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/devel
