wait isn't interruptable

[terinjokes]

I've noticed that wait isn't interruptable like it is in bash, which cased some issues with scripts relying on background sleep and trap behavior.

---

This is a trivial reproduction case trimmed from a project with lots of similar scripts.

action() {
	echo "from signal" | ts
}

status() {
	# do some background status report
	date
}

trap "action" USR1

echo $$ | ts
while true; do
        status
	sleep 10 &
	wait $!
done

---

When running in bash, sending USR1 to the process will immediately run the trap handler.

$ bash ./test.sh
Nov 05 01:07:01 7362
Thu 05 Nov 2020 01:07:01 AM PST
Thu 05 Nov 2020 01:07:11 AM PST
Nov 05 01:07:13 from signal
Thu 05 Nov 2020 01:07:13 AM PST
Nov 05 01:07:14 from signal
Thu 05 Nov 2020 01:07:14 AM PST
Nov 05 01:07:15 from signal
Thu 05 Nov 2020 01:07:15 AM PST
Thu 05 Nov 2020 01:07:25 AM PST
^C

$ kill -USR1 7362 && date
Thu 05 Nov 2020 01:07:13 AM PST
$ kill -USR1 7362 && date
Thu 05 Nov 2020 01:07:14 AM PST
$ kill -USR1 7362 && date
Thu 05 Nov 2020 01:07:15 AM PST

---

Running the same script in osh requires us to wait for the full 10 second sleep before the handler is executed.

$ osh ./test.sh
Nov 05 01:09:31 8770
Thu 05 Nov 2020 01:09:31 AM PST
[%1] Started PID 8774
Thu 05 Nov 2020 01:09:41 AM PST
[%2] Started PID 8857
Nov 05 01:09:51 from signal
Thu 05 Nov 2020 01:09:51 AM PST
[%3] Started PID 8933
Nov 05 01:10:01 from signal
Thu 05 Nov 2020 01:10:01 AM PST
[%4] Started PID 9001
Nov 05 01:10:11 from signal
Nov 05 01:10:11 from signal
Thu 05 Nov 2020 01:10:11 AM PST
[%5] Started PID 9094
Thu 05 Nov 2020 01:10:21 AM PST
[%6] Started PID 9178
^C

$ kill -USR1 8770 && date
Thu 05 Nov 2020 01:09:44 AM PST
$ kill -USR1 8770 && date
Thu 05 Nov 2020 01:09:55 AM PST
$ kill -USR1 8770 && date
Thu 05 Nov 2020 01:10:03 AM PST
$ kill -USR1 8770 && date
Thu 05 Nov 2020 01:10:05 AM PST

[andychu]

Thanks for the report! I reproduced this bug.

I think the bug is that signal handlers run in the main interpreter loop, and wait $pid runs in its own loop until the process stops.

Originally I thought this would be better fixed in the C++ translation, because signals are a bit different there, but it's possible it is an algorithmic problem. I'll look into it, thanks!

[andychu]

I just fixed this -- thanks for this great bug report!

It looks like bash/zsh/mksh all have a special flag to make the wait builtin interruptable, so now OSH does the same. (dash doesn't have it, so it has the same bug!)

The fix will be out with the next release.

---

Sort of related: I'm looking for feedback on Oil's enhanced shell tracing:

http://www.oilshell.org/preview/doc/xtrace.html

https://old.reddit.com/r/oilshell/comments/l0sshn/tracing_execution_in_oil_set_x_xtrace/

[andychu]

Fixed https://www.oilshell.org/release/0.8.7/

[terinjokes]

Oh wow, xtrace looks awesome for debugging.

[andychu]

Yup I just used it to debug some really poorly formatted code in #881 -- it was a big improvement. It's still early and up for discussion. More ideas on #891 !

It's out with 0.8.7 so keep it in mind when you need to debug :)