0.2.2: missing files in pypi sdist tarball: unable to run tests

Hi! I’m currently bringing python-dbus-next to the official Arch Linux repositories. As best practice we usually run tests against the packages (if there are any). Unfortunately this fails due to missing files, when using the sdist tarball on pypi:

============================= test session starts ==============================
platform linux -- Python 3.9.6, pytest-6.2.4, py-1.10.0, pluggy-0.13.1 -- /usr/bin/python
cachedir: .pytest_cache
rootdir: /build/python-dbus-next/src/python-dbus-next-0.2.2, configfile: pytest.ini
plugins: timeout-1.4.2, asyncio-0.15.1
timeout: 5.0s
timeout method: signal
timeout func_only: False
collecting ... collected 74 items / 2 errors / 72 selected

==================================== ERRORS ====================================
_________________ ERROR collecting test/test_introspection.py __________________
test/test_introspection.py:5: in <module>
    example_data = open(f'{os.path.dirname(__file__)}/data/introspection.xml', 'r').read()
E   FileNotFoundError: [Errno 2] No such file or directory: '/build/python-dbus-next/src/python-dbus-next-0.2.2/test/data/introspection.xml'
___________________ ERROR collecting test/test_marshaller.py ___________________
test/test_marshaller.py:20: in <module>
    table = json.load(open(os.path.dirname(__file__) + '/data/messages.json'))
E   FileNotFoundError: [Errno 2] No such file or directory: '/build/python-dbus-next/src/python-dbus-next-0.2.2/test/data/messages.json'
=========================== short test summary info ============================
ERROR test/test_introspection.py - FileNotFoundError: [Errno 2] No such file ...
ERROR test/test_marshaller.py - FileNotFoundError: [Errno 2] No such file or ...
!!!!!!!!!!!!!!!!!!! Interrupted: 2 errors during collection !!!!!!!!!!!!!!!!!!!!
============================== 2 errors in 0.30s ===============================

[dvzrv]

I can run the tests using dbus-run-session -- pytest -v.

Unfortunately in 0.2.3 the sdist tarball on pypi is still missing some files:

==================================== ERRORS ====================================
_________________ ERROR collecting test/test_introspection.py __________________
test/test_introspection.py:5: in <module>
    example_data = open(f'{os.path.dirname(__file__)}/data/introspection.xml', 'r').read()
E   FileNotFoundError: [Errno 2] No such file or directory: '/build/python-dbus-next/src/python-dbus-next-0.2.3/test/data/introspection.xml'
___________________ ERROR collecting test/test_marshaller.py ___________________
test/test_marshaller.py:20: in <module>
    table = json.load(open(os.path.dirname(__file__) + '/data/messages.json'))
E   FileNotFoundError: [Errno 2] No such file or directory: '/build/python-dbus-next/src/python-dbus-next-0.2.3/test/data/messages.json'
=========================== short test summary info ============================
ERROR test/test_introspection.py - FileNotFoundError: [Errno 2] No such file ...
ERROR test/test_marshaller.py - FileNotFoundError: [Errno 2] No such file or ...
!!!!!!!!!!!!!!!!!!! Interrupted: 2 errors during collection !!!!!!!!!!!!!!!!!!!!
============================== 2 errors in 0.31s ===============================

[leycec]

Affirmative. Would it also be feasible to gracefully @pytest.mark.skipif(..., reason='Non-portable dbus test.') the relevant integration tests away, in that case? Tests that can only reliably run within a unique non-portable environment (e.g., a particular Docker container running a particular dbus configuration) should ideally just be skipped with an informative message rather than fatally halting the entire test suite.

Just my 🪙 🪙 as a friendly package maintainer, of course.

You should decide which tests you need to skip based on your application.

B-b-but… I’m packaging for Gentoo Linux. There is no specific “application” as such.

The intended userbase is literally all users running Gentoo Linux. The intended use case is dbus_next installed as a system-wide Python package for use by other downstream reverse dependencies also installed as system-wide Python packages (e.g., bleak).

Again, the test suite itself should gracefully scale by selectively ignoring unit and integration tests that are inapplicable to the current execution environment. Thankfully, @pytest.mark.skip makes that trivial – assuming it’s trivial for your test suite to detect whether it’s running in your Docker container, of course. That should be trivial. That’s usually just a lookup of a unique environment variable exported by your Docker container.

Annnnnyway. I’ve marked our Gentoo ebuild for dbus_next as RESTRICT=test for now, meaning that tests are known to fail and should be ignored. I’m fine with that. It is what it is. 🤷‍♂️