The need for comparing two separately instantiated nodes seems to be historic
(related to an already-gone mode of pytest-xdist which would re-collect nodes)
and not actually needed anymore.
in a test invocation will have a corresponding FixtureDef instance.
also fixes issue246 (again).
simplify parametrized fixture teardown by making it work lazy:
during the setup of a parametrized fixture instance any previously
setup instance which was setup with a different param is torn down
before setting up the new one.
don't use autouse fixtures for now because it would cause a proliferation
and overhead for the execution of every test. Rather introduce a
node.addfinalizer(fin) to attach a finalizer to the respective node
and call it from node.setup() functions if the setup phase succeeded
(i.e. there is no setup function or it finished successfully)
When an autouse fixture in a plugin was encountered None was stored as nodeid
where it used to be ''. This broke the lookup of autouse fixtures later on.
This also adds another test for the normal fixture ordering which was slightly
wrong: a fixture without location was always added at the front of the fixture
list rather then at the end of the fixtures without location but before the
fixtures with location.
Conftests are plugins with a location attached to them while other
plugins do not have a location. When ordering fixturedefs those from
plugins without a location need to be listed first.
- When not specifying ids, let None and bools use their native string form (like str, int, float) rather than obfuscated form used for objects
- When specifying ids, explicitly raise a ValueError if a different number of ids are specified compared to the test cases
- Add tests for both these items.
address some comments by @hpk42 on 0b9d82e :
- move tests into their own class, rename
- add test showing metafunc.parametrize called in pytest_generate_tests rather than as decorator
- add test and fix single-argname case
- convert two loops into one in parametrize()
also
- renamed 'input' to 'n', since 'input' is a built-in
pytest-rerunfailures plugins) by re-initializing and removing
request/funcargs information in runtestprotocol() - which is a slightly
odd place to add funcarg-related functionality but it allows all
pytest_runtest_setup/teardown hooks to properly see a valid
request/funcarg content on test items.
holger krekel
Anatoly Bubenkov
James Lan
Floris Bruynooghe
Mathieu Agopian
holger krekel
variedthoughts
Brianna Laugher
Ronny Pfannschmidt
Graham Horler