Odd test variations based on download source

[sambler]

When 0.9.0 was released I saw some extra test failures compared to rc1. I have since found that the test results from github tag of 0.9.0 are different than the 0.9.0 release available from pypi.

Comparing the tarball from github and pypi, I see setup.cfg varies in formatting (spaces/tabs), while _version.py has >500 lines different. As the shorter _version.py should be output from the repo script, I don't expect either should cause variations in tests.

This is running on FreeBSD 12.1 amd64

0.9.0-github
1 failed, 113 passed, 56 skipped, 3 xfailed, 1 xpassed

0.9.0-pypi
1 failed, 109 passed, 56 skipped, 3 xfailed, 1 xpassed, 1 warnings, 4 error

[bashtage]

Which tests specifically fail?

[sambler]

the two links are log files of the test output.

[bashtage]

Seems to be something to do with the pandas and/or pytest you are using. There is no version information available for installed packages in the log.

[sambler]

pandas-0.24.2 - pytest 4.5.0 so both ports need updating.

But besides the updates, how can the 0.9.0 tarball from github and 0.9.0 from pypi give different results when they are using the same modules?

Between the two tests the pandas-datareader source tarball is the only changing factor.
Do you get the same results from both tarballs? or is it just me?

[bashtage]

What is the syntax fo the pytest command is used to involke the test?

[sambler]

The cmd used with the logs is cd $WRKSRC && python3.7 -m pytest pandas_datareader/tests

[bashtage]

And what is $WRKSRC?

[bashtage]

I have found the issue; it won't be fixed until 0.9.1.