+++ This bug was initially created as a clone of Bug #584007 +++

Description of problem:
Return of a command substitution from within another command substitution causes stdin of coprocess to be closed in a ksh script.

Version-Release number of selected component (if applicable):
ksh-20100309-3.fc14.x86_64

How reproducible:
Always

Steps to Reproduce:
1. Execute the following script in ksh:

#######################################
#!/bin/ksh

set -e

co_process()
{
    while read line
    do
    printf "\n%s(): Read line from 'stdin'\n" "${.sh.fun}" >/dev/tty
    done
    printf "\n%s(): Exiting -- EOF on read of 'stdin'.\n" "${.sh.fun}" >/dev/tty
    # kill the main shell process when the end-of-file is detected on 'stdin'.
    kill $$
}

co_process |&

loop_count=0

while :
do
    (( loop_count++ ))
    printf "%05d: %s\r" "${loop_count}" "$( echo $( /bin/ls ) )"
done 
#######################################

Actual results:

Script ends with the following output:

test_script.ksh: Exiting -- EOF on read of 'stdin'.\n

Expected results:

Script should run in an infinite loop.

Additional info:

--- Additional comment from spoyarek on 2010-04-20 09:50:50 EDT ---

Created an attachment (id=407821)
Close coprocess file descriptors only if the coprocess pid is nonzero

The root cause for this is the race condition between the job_reap called within the signal handler and job_wait in the parent process. job_reap sets the lastpid to the pid of the process that exited last, so if job_wait is called after it, it correctly compares a valid pid against the coprocess pid to see if it is the coprocess that has ended.

But in case of a command substitution, the job env (including the coprocess pid) is backed up and reinitialized, hence making the coprocess pid 0. In this case, if job_wait happens to execute before job_reap, it compares the 0 coprocess id with the 0 lastpid (since it has not been set through the signal handler yet) and goes on to close the coprocess assuming that it has quit.

Attached fix ensures that the coprocess pid is actually set before trying to close its handlers.

--- Additional comment from mhlavink on 2010-04-20 10:38:50 EDT ---

Hi,

thanks for the report and especially for the patch. I can reproduce this bug and confirm attached patch fixes it. I've emailed all information to upstream and I'm waiting for their response.