[Patch] libgfortran: do not assume libm

[Tristan Gingold]

On Mar 20, 2012, at 10:56 AM, Ralf Corsepius wrote:

> On 03/20/2012 10:01 AM, Tristan Gingold wrote:
>> Hi,
>> 
>> I am starting to build fortran for VMS.
>> 
>> The first serious issue was with libgfortran/configure.  It checks for several math functions, but directly in libm using AC_CHECK_LIB.
>> But there is on such things as libm on VMS systems (thus requiring to define MATH_LIBRARY as "").  Therefore all these tests fail,
>> resulting in a failure during libgfortran build.
>> 
>> I think the best way to test for the math function is to use AC_CHECK_DECLS as hinted by the autoconf manual
>> (cf the AC_CHECK_DECLS example in http://www.gnu.org/software/autoconf/manual/autoconf.html
>>          AC_CHECK_DECLS([j0], [], [], [[#include<math.h>]])
>>          AC_CHECK_DECLS([[basename(char *)], [dirname(char *)]])
>> 
>> )
>> but I am not an autoconf expert.
>> 
>> I think that this change makes the AIX specific check for __clog obsolete, but I haven't removed it.
>> 
>> With this change I was able to cross build libgfortran for VMS (ia64 and alpha).
>> I have also bootstrapped gcc for x86_64-darwin, without fortran regressions.
>> 
>> Ok for trunk ?
> 
> IMO, no.
> 
> What you are doing here is to replace checks, which check for presence of a symbol in a library (e.g. cos in libm.*) by checks for presence of a symbol's declaration inside of a header.
> 
> This is something entirely different (This also reflects in AC_CHECK_DECL by default providing HAVE_DECL_XXX defines).

Ah, indeed.  I missed that.

> Also note that it's not uncommon to have declarations inside of headers, which are not present inside of a library.

So, I should use AC_LINK_IFELSE instead, shouldn't I ?

Tristan.