Signal hang

This is the mail archive of the cygwin@sourceware.cygnus.com mailing list for the Cygwin project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]

I am attempting to build libstdc++ 2.90.6 with the following configuration:

- Windows NT4 sp3

- cygwin full b20.1

- Mumit's gcc 2.95.2 installed according to instructions

The symptom is that when mknumeric_limits runs, it hangs in a loop (CPU is chewed-up and nothing useful happens). When I dig into things, I find that gen-num-limits.exe is where the loop is happening, and, when I dig in farther, I find that the loop is in exception handling code. gen-num-limits is trying to generate divide-by-zero and overflow errors for various intrinsic types; this is where the problem is occurring.

If I run gen-num-limits interactively (so it spews to the terminal), I see that the behavior is not consistent; sometimes it chokes on divide-by-zero for one type and sometimes for another.

Here is a small program with which I can reproduce the problem simply (built with no special options):

------------------------------ cut here ----------------------------

#include <signal.h>
#include <setjmp.h>
#include <stdio.h>

jmp_buf env;

void signal_handler(int sig)
{
fprintf(stderr, "caught a signal\n");
longjmp(env, sig);
}

typedef char tested_type;

int main(int argc, char **argv)
{
   signal(SIGFPE, signal_handler);
   if (setjmp(env) == 0)
   {
      fprintf(stderr, "set the jump return location\n");

      tested_type     zero = tested_type();
      tested_type     one = tested_type(1);
      tested_type     junk = one / zero;
   }

fprintf(stderr, "Done!\n");
return 0;
}

------------------------------ cut here ----------------------------

This is essentially the flow of processing that occurrs in gen-num-limits. With tested_type set to an integral type (including bool), sometimes it works, sometimes it loops. Sometimes it just exits.

Does anyone know of an issue here?

Vadim Egorov posted a message about this a few months ago, but there are no replies on this list.

Thanks for any help.

-- Matthew Brown