Re: [PATCH] Replacement for the sigs_to_ignore hack in timeout.c

[Jim Meyering]

Pádraig Brady <address@hidden> wrote:
> Eric Blake wrote:
>> On the other hand, POSIX is explicit that mixing signal and sigaction is
>> not portable.  For that matter, now that gnulib provides a guaranteed
>> sigaction, why don't we just change all of coreutils to use it?  Affected
>> are: csplit, dd, install, ls, nohup, sort, tee, and timeout.
>
> Good idea.
>
> Perhaps we should just define our own signal()
> or bsd_signal() that calls sigaction() appropriately.
> I find signal() a much simpler interface than sigaction().

Hi Pádraig,

So do I.
I suppose it's worth considering, but "signal" is old and
unportable enough that I don't want to use that name.
Even bsd_signal carries baggage that we'd best avoid.

However, we have to weigh the cost of using a nonstandard
wrapper (readers would not recognize it right away) against
using POSIX's sigaction.

For now, let's look at the result after Eric's proposed signal-removing
clean-up/modernization is done.  Removing all of those in-function #ifdefs
is already a big improvement.  Maybe that'll be enough.  If not, you can
show us how much cleaner it can be using a simpler interface ;-)