Re: More exit() troubles

[Akim Demaille]

>>>>> "Lars" == Lars J Aas <address@hidden> writes:

Lars> I noticed that when I moved '' first, the first exit(42) test
Lars> didn't include <stdlib.h> - is that right?  

Yes, it does, but you didn't see it because it didn't fail.  The test
goes in two steps:

do
  _AC_COMPILE_IFELSE([AC_LANG_PROGRAM(address@hidden:@include <stdlib.h>
$ac_declaration],
                                      [exit (42);])],
                     [],
                     [continue])
  _AC_COMPILE_IFELSE([AC_LANG_PROGRAM([$ac_declaration],
                                      [exit (42);])],
                     [break])
done

first with, then without.  It's the without one which failed:

configure:3313: /home/larsa/src/Coin/cfg/m4/msvccc -c -g  conftest.cc >&5
configure:3316: $? = 0
configure:3338: /home/larsa/src/Coin/cfg/m4/msvccc -c -g  conftest.cc >&5
configure(3331) : error C2065: 'exit' : undeclared identifier
configure:3341: $? = 2
configure: failed program was:
#line 3326 "configure"
#include "confdefs.h"

int
main ()
{
exit (42);
  ;
  return 0;
}




There is something which is quite annoying with the current
configures: when something turns out bad, the log is not completed.
You can guess my point:  moving the call to _AC_INIT_LOG_COMPLETE
(which dumps confdefs.h and cache vars into the log) into the trap.

Now the obvious question: would someone think we might hit some
limitations?  It means adding 1246 characters to the trap.  Alexandre?