Hi Kaz,
On Sat, Jun 5, 2021 at 11:41 PM Kaz Kylheku (gmake)
<729-670-0...@kylheku.com> wrote:
>
> On 2021-06-05 07:00, Masahiro Yamada wrote:
> > make 2>&1 | tee build.log
> >
> > is a very common use case, and people often
> > miss to add (trap "" 2; ...), I think.
>
> Hi Masahiro,
>
> tee has a -i option to ignore the SIGINT signal.
> This is even specified by POSIX!
>
> Might that help in this specific scenario?
>
> (If tee ignores the SIGINT signal, how does it then terminate?
> make will quit first, and then tee reacts to the EOF on its standard
> input.)
Thanks!
I did not know this option.
It is true that this option is very useful,
but a problem is that not all people in the Linux kernel
community know the GNU Make internal
or tee having -i option.
I was considering two options.
[1] Add documentation and ask users to ensure that
when stderr of make is piped to another command,
that command should not die. Use -i option for 'tee'
[2] Modify Makefiles to do cleanups (for example by using 'trap')
without relying on Make.
Maybe I want to try [2] because
not all users read the documentation.
Anyway, thank you for the useful input!
>
> If that helps, consider that the filename arguments in tee are optional;
> tee copies stdin to stdout while duplicating the stream to
> *zero* or more files.
>
> Thus:
>
> make | tee -i | grep ....
>
> Whoever added this -i option might have intended it
> for exactly this scenario: as a pipe element that protects the
> previous element from SIGPIPE.
>
> The Coreutils tee's documentation says that -i ignores interrupt
> signals (suggesting more than one) but strace reveals only
> this:
>
> rt_sigaction(SIGINT, {sa_handler=SIG_IGN, sa_mask=[INT],
> sa_flags=SA_RESTART}, {sa_handler=SIG_DFL, sa_mask=[], sa_flags=0}, 8) =
> 0
>
>
--
Best Regards
Masahiro Yamada
