Using capfd in fixture breaks output capturing

[Nikratio]

Consider the following code:

import sys
import subprocess
import pytest

#@pytest.yield_fixture(autouse=True)
#def does_nothing(request, capfd):
#   yield

@pytest.fixture(scope='function')
def emits_stdout():
    sub = subprocess.Popen([
        sys.executable,
        '-c', ';'.join((
            'import time',
            'print("hello world")',
            'time.sleep(0.5)',
            'print("hello whirled")',
        ))
    ])
    yield 'foo'
    sub.wait()

def test_something(emits_stdout):
    import time
    time.sleep(0.7)
    assert False

The output is as expected:

$ python3 -m pytest test_it.py 
==================================== test session starts =====================================
platform linux -- Python 3.5.3, pytest-3.8.0, py-1.6.0, pluggy-0.7.1
rootdir: /home/nikratio/tmp/tests, inifile:
plugins: trio-0.5.0
collected 1 item                                                                             

test_it.py F                                                                           [100%]

========================================== FAILURES ==========================================
_______________________________________ test_something _______________________________________

emits_stdout = 'foo'

    def test_something(emits_stdout):
        import time
        time.sleep(0.7)
>       assert False
E       assert False

test_it.py:27: AssertionError
------------------------------------ Captured stdout call ------------------------------------
hello world
hello whirled
================================== 1 failed in 0.73 seconds ==================================

However, if I uncomment the "noop" fixture, no output is captured anymore:

$ python3 -m pytest test_it.py 
==================================== test session starts =====================================
platform linux -- Python 3.5.3, pytest-3.8.0, py-1.6.0, pluggy-0.7.1
rootdir: /home/nikratio/tmp/tests, inifile:
plugins: trio-0.5.0
collected 1 item                                                                             

test_it.py F                                                                           [100%]

========================================== FAILURES ==========================================
_______________________________________ test_something _______________________________________

emits_stdout = 'foo'

    def test_something(emits_stdout):
        import time
        time.sleep(0.7)
>       assert False
E       assert False

test_it.py:27: AssertionError
================================== 1 failed in 0.75 seconds ==================================

[RonnyPfannschmidt]

using the capfd fixture i not a noop, i beleive a integration point with it is problematic, but currently i dont recall the details

[Nikratio]

You are right! Adjusted the bug title.

[RonnyPfannschmidt]

based on https://docs.pytest.org/en/latest/capture.html#accessing-captured-output-from-a-test-function this is expected behaviour

[RonnyPfannschmidt]

@nicoddemus i believe this is yet another indication that a completely different structure for capture is needed

[Nikratio]

@RonnyPfannschmidt Could you elaborate on how this is expected behavior? I read the link several times now, but I don't understand how it explains what I am seeing.

[RonnyPfannschmidt]

@Nikratio the fixture captures the actual output, so it can be used in the test, its not passed trough to the outer pytest

[Nikratio]

Is there any workaround?

[RonnyPfannschmidt]

for example the capture fixtures have a method contextmanager to disable them for a portion of code

[Nikratio]

Hmm. Not sure I follow. I do not want to disable them, I want them to work..

[xmo-odoo]

the fixture captures the actual output, so it can be used in the test, its not passed trough to the outer pytest

So capsys / capfd doesn't rely on (and is not compatible with) the "builtin" capturing of stdout/stderr?

That would explain my problem, I wasn't sure whether it was worth opening a separate issue so maybe it's not if this is a known issue: my tests are interacting with a subprocess which is created in a fixture and inherits the test runner's stdout/stderr (just a subprocess.Popen(args…), no kwargs).

The output of that subprocess is correctly captured by pytest itself (as part of capturing the overall stdout/stderr), however neither capsys nor capfd can see it. Which makes sense if they just swap out the fds/streams locally during the test, they're too late for the subprocess to be impacted.

Am I correct that capfd-ing the subprocess's fixture would work, but then the subprocess's output wouldn't be displayed by pypy anymore (on test failure)? And even if I tried to manually inject it e.g.

out, err = capfd.readouterr()
with capfd.disable():
    print(out)
    print(err, file=sys.stderr)

I would still be missing bits if the test used readouterr internally?

[xmo-odoo]

Also somewhat of a duplicate of #3448 ?