source: MondoRescue/branches/3.3/mindi-busybox/shell/hush_test/hush-misc/sigint1.tests@ 3901

# What should happen if non-interactive shell gets SIGINT?

(sleep 1; echo Sending SIGINT to main shell PID; exec kill -INT $$) &

# We create a child which exits with 0 even on SIGINT
# (The complex command is necessary only if SIGINT is generated by ^C,
# in this testcase even bare "sleep 2" would do because
# in the testcase we don't send SIGINT *to the child*...)
$THIS_SH -c 'trap "exit 0" SIGINT; sleep 2'

# In one second, we (main shell) get SIGINT here.
# The question is whether we should, or should not, exit.

# bash will not stop here. It will execute next command(s).

# The rationale for this is described here:
# http://www.cons.org/cracauer/sigint.html
#
# Basically, bash will not exit on SIGINT immediately if it waits
# for a child. It will wait for the child to exit.
# If child exits NOT by dying on SIGINT, then bash will not exit.
#
# The idea is that the following script:
# | emacs file.txt
# | more cmds
# User may use ^C to interrupt editor's ops like search. But then
# emacs exits normally. User expects that script doesn't stop.
#
# This is a nice idea, but detecting "did process really exit
# with SIGINT?" is racy. Consider:
# | bash -c 'while true; do /bin/true; done'
# When ^C is pressed while bash waits for /bin/true to exit,
# it may happen that /bin/true exits with exitcode 0 before
# ^C is delivered to it as SIGINT. bash will see SIGINT, then
# it will see that child exited with 0, and bash will NOT EXIT.

# Therefore we do not implement bash behavior.
# I'd say that emacs need to put itself into a separate pgrp
# to isolate shell from getting stray SIGINTs from ^C.

echo Next command after SIGINT was executed