(awk) script piped to head not terminating after head finishes

pan64 · 11-23-2023, 12:58 AM

I just can't reproduce it using an old centos with gawk 4.0.2 and suse with 3 different gawk (4.x 5.1). I don't really know.
From the other hand in case of a sigpipe how do you know if [g]awk | head is broken or cmd | getline is broken, which one caused the signal?

ychaouche · 11-23-2023, 03:18 AM

This is interesting,
thank you all for your input.
So since pan64 can't reproduce it in gawk4 and gawk5.1,
could it have something to do with coreutils maybe?

MadeInGermany · 11-23-2023, 05:06 AM

I think it is a bug in gawk. Happens in combination with | getline and close() and SIGPIPE on its stdout.

A workaround: avoid the SIGPIPE on stdout.

Code:

printf %s\\n {1..20} | awk '{cmd="sleep 1; echo howody?"; cmd | getline x; close(cmd); print x; if (++n == 2) exit}'; echo done

ychaouche · 11-23-2023, 05:11 AM

Thanks for the workaround @MadeInGermany,
however the goal is to have the (original) awk script unchanged,
and process its output by piping it to other utilities the unix way.

pan64 · 11-23-2023, 05:39 AM

you might collect all the output into a file and run head on that file. It is probably not that efficient, but at least should work.
Probably it depends on the version of bash too, it is the shell which must somehow manage the pipe(s) and the sigpipe signal.