Re: [bug-gnulib] fatal-signal.c's install_handlers catches ignored signals?

[Paul Eggert]

Bruno Haible <address@hidden> writes:

> What's precisely the case when this would matter?

nohup is the classic example.  There are a few other cases where
programs ignore SIGINT, SIGPIPE, etc. for one reason or another.

>   - What sense would it make for a program to install SIG_IGN handlers for
>     a signal and also call at_fatal_signal for it ?

Normally a program would like to catch SIGHUP and clean up before
exiting.  But when it's invoked by nohup, the program should simply
ignore SIGHUP.  Similarly for the other signals.

>     This is dubious practice, as written in
>       http://www.opengroup.org/onlinepubs/009695399/functions/exec.html

That same page says that the usual behavior "permits some useful
functionality, such as the nohup command".  It is this sort of
functionality that I'd like to support.  That page mostly disparages
blocking signals across an exec, not ignoring them.