capfd not working in 3.3.1

[trrichard]

Fixtures that use capfd don't seem to work in version 3.3.1 running on Ubuntu 14.04 LTS with anaconda 3.6.3

import sys

@pytest.fixture()
def fix(capfd):
    yield None
    out, err = capfd.readouterr()
    stdout_filename = "/tmp/stdout"
    stderr_filename = "/tmp/stderr"
    mode = 'w'
    with open(stdout_filename, mode) as stdout:
        stdout.write(out)
    with open(stderr_filename, mode) as stderr:
        stderr.write(err)

def test_a(fix):
    print("lolcatz")
    print("this is a test")
    sys.stderr.write("err")

The stdout and stderr files end up empty after running that test. Using this set of pytest dependencies

certifi (2017.11.5)
pip (9.0.1)
pluggy (0.6.0)
py (1.5.2)
pytest (3.3.1)
setuptools (38.2.4)
six (1.11.0)
wheel (0.30.0)

Steps of replication are

1. Create the above test
2. Run pytest on it.
3. cat /tmp/std* (they will be empty)
4. Downgrade pytest ( pip install autohil==3.2.5 )
5. Run the test again
6. cat /tmp/std*

[pytestbot]

GitMate.io thinks the contributor most likely able to help you is @nicoddemus.

[nicoddemus]

Thanks @trrichard for the report!

I simplified the test a bit:

import sys
import pytest

@pytest.fixture()
def fix(capfd):
    yield
    out, err = capfd.readouterr()
    assert out == 'lolcatz\n'
    assert err == 'err'

def test_a(fix):
    print("lolcatz")
    sys.stderr.write("err")

I confirm that this fails in pytest >=3.3 and works on 3.2.5.

[trrichard]

I lean towards it being related to this but I'm just now looking at the code

#2709

[nicoddemus]

A quick bisect run shows you are right:

9919269ed048b6e9147ee3301532e3591b9a112b is the first bad commit
commit 9919269ed048b6e9147ee3301532e3591b9a112b
Author: Bruno Oliveira <[email protected]>
Date:   Tue Sep 26 02:34:41 2017 -0300

    Allow to use capsys and capfd in other fixtures

    Fix #2709

:040000 040000 004bc037efc094f23211836b45a0c908944c0e5f 4239d2322f4631055d5a64c5178c5d277c831370 M      _pytest
:040000 040000 d900a2d158ea9a69c319ece6ca72db5913e93d7e 7b88a85bb535d2d562a5e6030ba0bfd856dc03c6 M      changelog
:040000 040000 3aa3d4630888de1ce3b6f51d16c0c5639622fcf6 66399dcb8b52ab95ecca77bdd9d7ebf1963f8e18 M      testing
bisect run success

[trrichard]

@nicoddemus I'm looking at the code for this and for the previous change and I'm not clear on what the behavior of this should be.

Is the expectation that each call to .readouterr() clears the output buffer? Tests like the one here: nicoddemus@9919269#diff-6221f80f4e6d91715381f8d4229b645cR520 Seem to indicate that.

Similarly how should the capture behave when -s is present on the command line? Does the output go to both places (both the fixture and the output) or just one?

Let me know what you think. I still haven't figured out what actually broke yet, so any insight into that would be appreciated.

[trrichard]

@nicoddemus I found the error.

nicoddemus@9919269#diff-d8641d58661f0fa63a36c6f6e85715d9R163

It's a 1 liner. Remove activate_fixture(item) from teardown. It causes a double start which effectively clears the results before they show up in the fixture.

[AlexanderTyrrell]

@trrichard Thanks for finding this bug. I am facing the same issue, but removing activate_fixture(item) from teardown does not work for me. I get an error message:

test.logFile.write(capfd.readouterr().out)
E AttributeError: 'tuple' object has no attribute 'out'

Do you see the same issue? Thanks.

[nicoddemus]

There's #3066 open with a possible fix, but it brakes tests. Btw sorry @trrichard for not looking into it yet, but it is in my todo list. 👍

[nicoddemus]

@AlexanderTyrrell #3830 will also fix that AttributeError: 'tuple' object has no attribute 'out' error. 👍