Running a subprocess fails when not using pytest no-capture

user_2120710 0
Viewed 387

Setup

I've simplified the problem to the core:

  1. Having pytest installed
pip install pytest==5.4.3
  1. Having a shell script
# has-stdin.sh

# Detect stdin
if [[ ! -t 0 ]]; then
    echo "yes"
else
    echo "no"
fi
  1. Having a Python test
# test.py

import subprocess
import unittest


class TestCase(unittest.TestCase):

    def test(self):
        process = subprocess.run(
            ["sh", "./has-stdin.sh"],
            stderr=subprocess.PIPE,
            stdout=subprocess.PIPE,
            check=True,
            shell=False
        )
        assert process.stdout.decode("utf-8") == "no\n"

Testing

✅ Script works in a bash shell

$ sh ./has-stdin.sh
no

$ echo '' | sh ./has-stdin.sh
yes

$ sh ./has-stdin.sh <<< ''
yes

✅ Runs successfully with -s (e.g. --capture=no)

$ pytest test.py -s
platform linux -- Python 3.7.7, pytest-5.4.3, py-1.9.0, pluggy-0.13.1
rootdir: /Users/maikel/docker/library/postgresql
collected 1 item

test.py .

=========================== 1 passed in 0.02s ===========================

❌ Runs unsuccessfully without -s

$ pytest test.py
========================== test session starts ==========================
platform linux -- Python 3.7.7, pytest-5.4.3, py-1.9.0, pluggy-0.13.1
rootdir: /Users/maikel/docker/library/postgresql
collected 1 item

test.py F                                                                                                                                                                                                                      [100%]

================================ FAILURES ===============================
_____________________________ TestCase.test _____________________________

self = <test.TestCase testMethod=test>

    def test(self):
        process = subprocess.run(
            ["sh", "./has-stdin.sh"],
            stderr=subprocess.PIPE,
            stdout=subprocess.PIPE,
            check=True,
            shell=False
        )
>       assert process.stdout.decode("utf-8") == "no\n"
E       AssertionError: assert 'yes\n' == 'no\n'
E         - no
E         + yes

test.py:23: AssertionError
======================== short test summary info ========================
FAILED test.py::TestCase::test - AssertionError: assert 'yes\n' == 'no\n'
=========================== 1 failed in 0.10s ===========================

What is making the difference in using -s? How to run this pytest successfully without -s?

I've tried Test calling of an external script run via Popen that expects no available data in stdin using pytest, but when using that method both pytest commands fail.

pytest subprocess python
2 Answers

That's a very interesting problem.

I got so intrigued so I did run the code locally and examined what's happening:

In the first case (running without -s) the shell's /proc/self/fd/0 is /dev/null but running with -s the shell's /proc/self/fd/0 is /dev/pts/7 in my case.

$ pgrep -f has-st
12727
$ ls -la  /proc/12727/fd/0 
lrwx------ 1 quoyn quoyn 64 Aug  4 00:13 /proc/12727/fd/0 -> /dev/pts/7

and

$ pgrep -f has-st
12793
$ ls -la  /proc/12793/fd/0 
lr-x------ 1 quoyn quoyn 64 Aug  4 00:17 /proc/12793/fd/0 -> /dev/null

This si because when you run with "-s" pytest does not capture the terminal and simply passes it through, that's why it is /dev/pts/7. If I run ls -l /proc/self/fd/0 in my terminal after pytest has quit I indeed get:

lrwx------ 1 quoyn quoyn 64 Aug  4 00:29 /proc/self/fd/0 -> /dev/pts/7

You can try without pytest as well, only python:

import subprocess
s = subprocess.run(["ls", "-l", "/proc/self/fd/0"], stderr=subprocess.PIPE, stdout=subprocess.PIPE, input=None, check=True, shell=False)
print(s.stdout)

And try executing as python code_above.py and echo '' | python code_above.py

EDIT:

And to answer the second question

How to run this pytest successfully without -s?

Use a pseudo terminal yourself. Run the subprocess in a pty.\

EDIT2:

The mimimal test to reproduce/demonstrate this problem, run the following code as pytest test.py and pytest -s test.py:

import sys
def test():
    assert sys.stdin.isatty()
user_140436910 
$ pip install ptyprocess==0.6.0

# test.py

import unittest

from ptyprocess import PtyProcess


class TestCase(unittest.TestCase):

    def test(self):
        process = PtyProcess.spawn(
            ["sh", "./has-stdin.sh"]
        )
        process.wait()
        output = process.read().decode("utf-8")
        if process.exitstatus:
            raise Exception(output.strip("\r\n"))
        assert output.strip("\r\n") == 'no'

✅ Runs successfully with -s:

$ pytest test.py -s
platform linux -- Python 3.7.7, pytest-5.4.3, py-1.9.0, pluggy-0.13.1
rootdir: /Users/maikel/docker/library/postgresql
collected 1 item

test.py .

=========================== 1 passed in 0.60s ===========================

✅ Runs successfully without -s:

$ pytest test.py
platform linux -- Python 3.7.7, pytest-5.4.3, py-1.9.0, pluggy-0.13.1
rootdir: /Users/maikel/docker/library/postgresql
collected 1 item

test.py .

=========================== 1 passed in 0.60s ===========================

Caveats

  • It runs quite a bit slower than subprocess (0.60s vs 0.02s in this example)
  • There is no distinction between stdout and stderr

The situation is explained by this diagram:

pty-diagram

Source: https://github.com/pexpect/ptyprocess/blob/3931cd4/docs/index.rst

user_21207100
Related