(This was originally spotted in puma/puma#1558, thanks @dekellum for great help there.)

With JRuby 9.1.16.0 on Java 8 (on latest macOS), I get this:

$ jruby -e 'raise SignalException, "SIGTERM"' ; echo $?
SignalException: SIGTERM
  <main> at -e:1
1

Same with JRuby 1.7 actually:

$ rvm use jruby-1.7.27
[09:59:19] plundberg@ecvaawplun6:~/git/tmp (master)
$ jruby -e 'raise SignalException, "SIGTERM"' ; echo $?
SignalException: SIGTERM
  (root) at -e:1
1

However, with MRI 2.5.1 I get this:

$ ruby -e 'raise SignalException, "SIGTERM"' ; echo $?
Terminated: 15
143

...which is indeed the correct value for a SIGTERM-triggered exit of a process.

As inspired by #5049, I tried another approach also which worked slightly differently:

$ rvm use 2.5.1
Using /Users/plundberg/.rvm/gems/ruby-2.5.1
$ ruby -e "Process.kill('TERM', Process.pid)" ; echo $?
Terminated: 15
143

Here be the surprise: JRuby behaves differently, and even differently than the raise SignalException, "SIGTERM" scenario.

$ rvm use jruby
Using /Users/plundberg/.rvm/gems/jruby-9.1.16.0
$ ruby -e "Process.kill('TERM', Process.pid)" ; echo $?
0

Exit code 0 is not entirely correct, but it's at least much better than exit code 1 which indicates failure. 😄

Inspired by this Oracle article linked to from the JRuby wiki, I tested with the -Xrs parameter:

$ rvm use jruby
Using /Users/plundberg/.rvm/gems/jruby-9.1.16.0
$ export JAVA_OPTS=-Xrs
$ ruby -e "Process.kill('TERM', Process.pid)" ; echo $?
Terminated: 15
143
$ ruby -e 'raise SignalException, "SIGTERM"' ; echo $?
SignalException: SIGTERM
  <main> at -e:1
1

So that makes the Process.kill approach work just like MRI, which is good (so we have a decent enough workaround for that.) However, the raise SignalException, 'SIGTERM' still gives the seemingly incorrect exit code.

Is this by design or is it a bug?