Eric Blake wrote: > > Could it be that the process was started with SIGALRM inherited as ignored?
Thanks for the fix. > The corresponding tests/test-strstr.c and friends can continue to > use alarm() in an environment with SIGALRM ignored, rather than > also doing the signal reset, because a) we should already be using > the linear gnulib replacement which should be fast enough to never > trip the alarm, and b) if we aren't, the user will hopefully notice > that 'make check' is taking forever and file a bug report. I disagree here. When a user does "make check" in gettext, I don't want to clog his machine for longer than necessary. Also, he is more likely to report a FAILed test rather than a 3-hours test, because he doesn't know a priori which tests are supposed to take a while (e.g. test-getaddrinfo) and which are not. May I add the 'signal (SIGALRM, SIG_DFL);' also in the 3 tests? Bruno
