Re: AC_CHECK_FUNC is not compatible with gcc -Werror

[Dmitry V. Levin]

On Sat, Jan 23, 2010 at 09:52:20AM +0100, Paolo Bonzini wrote:
> On 01/23/2010 01:26 AM, Dmitry V. Levin wrote:
> >I'm not sure about $ac_[]_AC_LANG_ABBREV[]_werror_flag, because it is not
> >set when -Werror is included in CFLAGS, and the latter case is more common
> >than the use of AC_LANG_WERROR.
> 
> ... but still you use AC_LANG_WERROR in the testcase.

Here is another test, for gcc -Werror:

# AC_CHECK_FUNCS with gcc -Werror
# -------------------------------
# Check that it performs the correct actions:
# Must define HAVE_PRINTF, but not HAVE_AUTOCONF_FTNIRP
AT_CHECK_MACRO([AC_CHECK_FUNCS with gcc -Werror],
[AC_PROG_CC
test "$GCC" = yes || exit 77
CFLAGS="${CFLAGS-} -Werror"; export CFLAGS
AC_CHECK_FUNCS(printf autoconf_ftnirp)],
[AT_CHECK_DEFINES(
[/* #undef HAVE_AUTOCONF_FTNIRP */
#define HAVE_PRINTF 1
])])

> It's been broken forever,

Yes, but...

> I think we can afford supporting only the case when 
> AC_LANG_WERROR is used explicitly.

I'm not sure about it.  The most common case is when CFLAGS is made before
a ./configure call.  I'm aware of build systems where CFLAGS variable is
being constructed based on other build parameters including Werror.
Of course it's easy to replace
        if test "$GCC" = yes; then
with
        if test "$GCC":"$ac_[]_AC_LANG_ABBREV[]_werror_flag" = yes:yes; then
and continue telling people to avoid -Werror before ./configure, but it
means that I'll have to maintain this one-line patch that removes the
werror flag check.


-- 
ldv

pgpOcpZ_v7HVr.pgp
Description: PGP signature