Re: signature warnings on OSF/1 5.1

[Bruno Haible]

Hi Simon,

> > * test-gethostname.c:24: warning: initialization from incompatible pointer 
> > type
> > Similarly:
> >   - POSIX: int gethostname(char *, size_t);
> >   - OSF/1: int gethostname(char *, int);
> >
> > Likewise.
> >
> > * test-inet_ntop.c:24: warning: initialization from incompatible pointer 
> > type
> > Similarly:
> >   - POSIX: const char *inet_ntop(int, const void *, char *, socklen_t);
> >   - OSF/1: const char *inet_ntop(int, const void *, char *, size_t);
> >
> > Likewise.
> 
> Couldn't this lead to crashes on 64-bit systems?  Or aren't there any
> 64-bit OSF/1 systems?  I'm not familiar with OSF/1.

OSF/1 systems all have the DEC Alpha CPU, a 64-bit processor. On this processor,
all integer registers are 64 bits wide, and the first 6 arguments are passed
in integer registers (see gcc-4.5.2/gcc/config/alpha/alpha.h the definition of
FUNCTION_ARG_REGNO_P). Therefore for a value between 0 and 2^32-1 it does not
matter whether it is passed as 'unsigned int' (32 bits, zero-extended) or
'unsigned long' (64 bits). Regarding the zero-extend, I believe gcc nowadays
handles this correctly (GCC 2.x had bugs in this area).

In summary, passing 'int' / 'socklen_t' / 'size_t' by value is not critical
on Alpha. Only passing such an argument or return value through a pointer
may require a wrapper function.

Bruno