TL;DR:
1. Has someone an idea of usage of the module pkg_resourses,
specifically the iterator pkg_resources.iter_entry_points,
for fetching local resources?
2. Is someone aware of recent (as in the past year or so)
breaking changes in the way the pkg_resources module works?
Greetings,
While attempting to give a hand with maintenance of various
debian-med python packages, various sources were failing their
test suites due to diverse failures to import their test modules
through the mechanism provided by the module pkg_resources.
Andreas Tille first noted this in python-cobra:
https://lists.debian.org/debian-med/2020/09/msg00369.html
Then I hit some issues involving pkg_resources as well in Qiime,
while attempting to update it, and it turned out even the
current version does not manage to pass tests, due to failure to
load a dummy-plugin with this system:
https://lists.debian.org/debian-med/2020/10/msg00014.html
https://salsa.debian.org/med-team/qiime
Note the package will incorrectly report that the QIIMETEST
variable should be set, although it actually is. Here is a
sample error from a build attempt of qiime_2019.10.0-1.dsc:
======================================================================
ERROR: test_import_root (qiime2.tests.test_artifact_api.TestImports)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/tmp/autopkgtest.MTf3Wq/build.GrO/src/.pybuild/cpython3_3.8_qiime/build/qiime2/tests/test_artifact_api.py", line 23, in setUp
get_dummy_plugin()
File "/tmp/autopkgtest.MTf3Wq/build.GrO/src/.pybuild/cpython3_3.8_qiime/build/qiime2/core/testing/util.py", line 19, in get_dummy_plugin
raise RuntimeError(
RuntimeError: When running QIIME 2 unit tests, the QIIMETEST environment variable must be defined so that plugins required by unit tests are loaded. The value of the QIIMETEST environment variable can be anything. Example command: QIIMETEST=1 nosetests
The get_dummy_plugin function is a wrapper around a class which
ultimately calls an iterator like this to build a for loop:
pkg_resources.iter_entry_points(group='qiime2.plugins')
where qiime2/plugins.py is available at the root of the source
code. My main problem is: I did not manage to get this
construction to return a proper object to point to the dummy
plugin stored at that location for the test suite, even when
making sure the sys.path includes the current working directory
prior to including the pkg_resources module.
Nilesh Patra also followed up with something related to
pkg_resources on the package gubbins and the open bug #971111,
although the situation seemed different this time, as this
package shows issues in the requirements parser of pkg_resources
and not the distribution locator:
https://lists.debian.org/debian-med/2020/10/msg00015.html
https://bugs.debian.org/971111
I even saw something somewhat similar in the present list
archives recently, although altering the PYTHONPATH, or even the
sys.path directly, did not help in my case:
https://lists.debian.org/debian-python/2020/10/msg00005.html
I'm afraid I'm not at ease with dealing with pkg_resources,
although I did try to make sense of how it is supposed to work,
by referring to its documentation, which I believe is here:
https://setuptools.readthedocs.io/en/latest/pkg_resources.html
All these test failures show slightly different symptoms, I even
tested Qiime building on Buster, and the issue was already
present, while former versions of python-cobra or gubbins made
it to Buster, so it is quite possible they are unrelated after
all. But just in case, is someone aware of recent breaking
changes in the way pkg_resources module works?
Kind Regards,
--
Étienne Mollier <etienne.mollier@mailoo.org>
Old rsa/3072: 5ab1 4edf 63bb ccff 8b54 2fa9 59da 56fe fff3 882d
New rsa/4096: 8f91 b227 c7d6 f2b1 948c 8236 793c f67e 8f0d 11da
Sent from /dev/pts/3, please excuse my verbosity.
Attachment:
signature.asc
Description: PGP signature