It would fail if the length of bytes and unicode differs. It now passes non-bytes to `Form` directly.
* py3 fixes for form handling in paste.fixture It uses "not six.PY2" in contrast to other places in the code to be forward-compatible. I've not looked too closely, but it might make sense to decode body/text in the beginning already, instead of having it as bytes internally?! Also, like mentioned in [1] already, it should probably use the correct source encoding?! 1: https://github.com/cdent/paste/blob/36e5b8bd16a6063ec654faf04541f3a20d19f7fe/paste/fixture.py#L820 * Add test, using/fixing SlowConsumer form app
* pytest: fix collection warnings via __test__=False Fixes > "cannot collect test class %r because it has a __init__ constructor Ref: https://github.com/pytest-dev/pytest/issues/2007 * pytest: configure testpaths This is faster with test collection. * pytest: fix warning with doctests Fixes > /usr/lib/python3.7/site-packages/_pytest/python.py:764: > RemovedInPytest4Warning: usage of Generator.Function is deprecated, > please use pytest.Function instead * Minor fixes around s/py.test/pytest/
pytest exposes many warnings, some but not all of which are cleaned up here. The main switch is to use html.escape instead of cgi.escape. This inspired the addition of 'future' to requirements. The remaining warnings are related to pytest deprecations or over-eager test discovery. It is perhaps ironic that the switch to pytest is to avoid nose being mostly dead, and now we are using features in pytest that pytest wants to make dead. These are left for later cleanups, which means that running the tests is noisy.
Switch from nose to pytest
The proxy test relied on pythonpaste.org being up and existing. It is neither. So instead we use httpbin.org which is often used for this kind of thing. Unfortunately httpbin is now a react app which means a lot of the HTML is generated client-side, which means we need to choose wisely. As the original comments indicate, the test is not particularly robust and remains so.