Re: Tru64 C V6 and _AC_PROG_CC_STDC

[Norman Gray]

Paul,

On 2005 Feb 11 , at 18.23, Paul Eggert wrote:

> Norman Gray <address@hidden> writes:
>
> > Having __STDC__ equal to 0 violates the standard, thus this test is
> > giving the wrong result.  I've no idea what these `useful extensions'
> > are, and if I did, I'd certainly be sure to avoid them,
>
> Some of these "useful extensions" are <stdio.h> functions like fileno,
> which are required by POSIX.
[...]
> However, as a practical matter, I've found that it's more hassle to
> disable functions like fileno than it is to enable them.

That's wise.

Myself, I like --std-with-bondage-and-discipline flags, as they keep me 
honest.  I feel they're more useful with C++ than with C, because there 
are more ways to stuff up with C++; and the fact that this is so I feel 
is one of the arguments against C++ as a generally useful language.

It might be worth a footnote in the AC_PROG_CC documentation, though, 
saying that `ANSI C' actually means `ANSI/ISO C without the surprising 
bits', and...

> That being said, it might be reasonable to define a new Autoconf macro
> AC_PROG_CC_STDC_PEDANTIC, or something like that, which tests for 100%
> conformance to the C Standard in matters like this.

...gets you `ISO C, and we really mean it'.

The following patch to autoconf 2.59's lib/autoconf/c.m4 might be a 
start.


497a498,507
> # AC_PROG_CC_STRICT([COMPILER...])
> # --------------------------------
> # As with AC_PROG_CC, except that the compiler is checked for strict
> # compliance with the ISO C standard.
> m4_define([AC_PROG_CC_STRICT],
> [m4_define([_AC_PROG_CC_STRICT_FLAG], [yes])
> AC_PROG_CC($1)
> m4_undefine([_AC_PROG_CC_STRICT_FLAG])
> ])# AC_PROG_CC_STRICT
>
794a805,807
> #
> # If the macro _AC_PROG_CC_STRICT_FLAG is set to "yes", then check
> # for strict compliance with the ISO C standard.
804a818,824
> ]m4_if(_AC_PROG_CC_STRICT_FLAG, [yes], [dnl
>
> /* Check that the __STDC__ flag is defined to 1, not 0 */
> #if !__STDC__
>   choke me
> #endif
> ])[


...but only a start.  The `strict' test in there is pretty primitive, 
and doesn't, for example, end up requiring gcc's -ansi or -pedantic 
flags.  I'll leave for a separate discussion precisely what tests would 
be appropriate there, and which of those flags would be useful.  Pace 
my remarks above, aiming for -ansi or equivalent would probably satisfy 
_me_ at least, since that ends up with the compiler behaving exactly as 
ISO-9899 or ISO-14882 says it should.

[And just by the way, an -iso flag would probably be less parochial 
than -ansi, unless we can have a -bsi, a -din, ... as well (hrumph).  
But that really is a separate issue!]

All the best,

Norman


--
----------------------------------------------------------------------
Norman Gray  :  Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/  :  www.starlink.ac.uk