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

[jay]

On Fri, Dec 10, 2004 at 08:34:55PM +0100, Bruno Haible wrote:

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

SIG_IGN is inherited across an execve(), except for SIGCHLD.

>   - If the SIG_IGN is not installed by the program but by its parent process?

Of course, here you're also possibly talking about the same process,
not its parent (execve() is of course called in the child).  

I suppose it's also concievable that we'd want to use gnulib in a
library package which sets up temporary signal handlers while it's in
control.  Hence some singals could be set to SIG_IGN by the parent
program, which merely uses the library and hence won't know what
signal handling arrangements are made by the library.  As a general
rule of course I'd suggest that people should think three times before
installing a signal handler in a library function.  Normally this
would be surprising and unwelcome.

Regards, 
James.