spawn's .before might contain input of sendline #821
Description
I have noticed that on one of the systems I am working one command "breaks" the spawn, in the sense that after it is run, the .before member contains content that actually comes from the last call to sendline().
This actually breaks the jupyter bash kernel on my system (see takluyver/bash_kernel#151).
import pexpect
import os
bashrc= os.path.join(os.path.dirname(pexpect.__file__), 'bashrc.sh')
child = pexpect.spawn("bash",
['--rcfile',bashrc],
echo=False,
encoding = 'utf-8',
codec_errors='replace')
child.expect_exact(["$"], timeout=2)
# This mimics, in a simplified way, what happens in REPLWrapper.run_command()
def run_command(command):
print("-----------------------------------------")
print(f"Command: {command}")
print("-----------------------------------------")
child.sendline(command)
child.expect_exact(["$"], timeout=2000)
out= u''.join([child.before])
print(f"Output:\n{out}")
# this works fine
run_command("pwd")
# this works fine too
run_command("ls")
breaking_command = "enroot import docker://nvcr.io\#nvidia/pytorch:21.04-py3 &> enroot-import-output.txt"
# Now this breaks the spawn .before member, even if **there is no output**
run_command(breaking_command)
# The output of the subsequent commands shows the problem
run_command("ls")
# here too
run_command(" {echo $?; } > /dev/null")Output:
-----------------------------------------
Command: pwd
-----------------------------------------
Output:
/home/hk-project-scs/hs2454/jupyter
-----------------------------------------
Command: ls
-----------------------------------------
Output:
alpine.sqsh error.txt reslog.txt
beforetype.txt logfile.txt send_logfile.txt
child_output.txt output start-jupyter.sh
config pty_reproducer_attempt.py Untitled.ipynb
enroot-import-output.txt read_logfile.txt venv
ENV reproducer.py write_logfile.txt
error_codes.txt requirements.txt
-----------------------------------------
Command: enroot import docker://nvcr.io\#nvidia/pytorch:21.04-py3 &> enroot-import-output.txt
-----------------------------------------
Output:
-----------------------------------------
Command: ls
-----------------------------------------
Output:
ls # <<<<<<<<< THIS SHOULD NOT BE HERE
alpine.sqsh error.txt requirements.txt
beforetype.txt logfile.txt reslog.txt
child_output.txt nvidia+pytorch+21.04-py3.sqsh send_logfile.txt
config output start-jupyter.sh
enroot-import-output.txt pty_reproducer_attempt.py Untitled.ipynb
ENV read_logfile.txt venv
error_codes.txt reproducer.py write_logfile.txt
-----------------------------------------
Command: {echo $?; } > /dev/null
-----------------------------------------
Output:
{echo # <<<<<<< THIS SHOULD NOT BE HERE (truncated because there's a $ which is the prompt...)
I will do some additional, lower-level tests and check if I manage to understand more.
Notes:
- the "breaking command" causes the issue every time I try it on that particular machine. On a different machine it does not.
- replacing that "breaking command" with a
sleep 240(which lasts a similar amount of time) makes the problem disappear. - Shorter commands invocation of the same tool as in the "breaking command" do not trigger the issue either.