Re: Call for testing CVS Autoconf

[Jim Meyering]

Akim Demaille <address@hidden> wrote:
| >>>>> "Alexandre" == Alexandre Oliva <address@hidden> writes:
| Alexandre> On Nov 30, 2000, Akim Demaille <address@hidden> wrote:
| >> | +./configure: 24794 Alarm call
|
| >> This is why it failed: the shell complains of the signal it
| >> received.  I don't know how to address this issue.
...
| Jim, Paul, you two guys certainly know very well the mktime test.
| How come the alarm was still set?

Do you mean why did was the code still running 60 seconds
after it started?  Because that system has a rotten mktime implementation.

| And why is it
| the shell that received it (Ahem, my Unix knowledge is not too good...).

Because the program didn't handle SIGALRM itself.

| How would you fix it?

[ I haven't looked at the test suite code that tests AC_FUNC_MKTIME.
  I presume that by `it' you mean the macro itself, and not the testing
  framework. ]

One way would be to install a signal handler for SIGALRM
and then exit nonzero from the handler.

But that might be a little too involved a change to make this close
to a test release.  Especially considering that this happens only on
systems with a broken mktime function.  At worst, all it does is cause
a little confusion among the few users who actually pay attention to
the output of running configure on a losing system.