test.pl: watchdog: Simplify, clarify, recover from failure

[khwilliamson]

@tonycoz noticed a simplification this makes.

If something goes wrong on WIndows and VMS, previously it gave up; now it tries an alarm.

It now doesn't try to use the thread method unless safe signals are in effect.

Some comments are clarified

• This set of changes does not require a perldelta entry.

[tonycoz]

How is:

watchdog(5);
sleep 4;
watchdog(10);
sleep 8;
# eof

meant to behave?

Right now with the changes from this PR a simple test like:

#!perl
    BEGIN {
        chdir 't' if -d 't';
        require './test.pl';
        set_up_inc(qw '../lib ...');
    }

watchdog(5);
sleep 4;
watchdog(10);
sleep 8;
ok(1);
done_testing();

I get:

$ ./perl t/try.t
# Test process timed out - terminating
Killed

One option here would be for all entries to the watchdog() function to clear existing watchdogs and then return immediately after if the timeout is zero

With that the above produces:

$ ./perl t/try.t
ok 1 - [at t/try.t line 12]
1..1

I tested with:

@@ -1849,7 +1849,6 @@ sub watchdog ($;$)
 
     # If cancelling, use the state variables to know which method was used to
     # create the watchdog.
-    if ($timeout == 0) {
 
         # Is a no-op if no watchdog is currently set
         if ($watchdog_thread) {
@@ -1865,8 +1864,8 @@ sub watchdog ($;$)
         undef $watchdog_thread;
         undef $watchdog_process;
         undef $watchdog_alarm;
-        return;
-    }
+        return if $timeout == 0;
+

[khwilliamson]

fixed