Do not leave files after running unit tests

[mfeurer]

Currently, the unit tests leave a few files. This clutters the development directory and makes it harder to keep an overview in git.

[rth]

That would be fairly straightforward with pytest #327

[Neeratyoy]

Hi,
On taking a preliminary look at openml/testing.py, the setUp() and tearDown() does a neat job of creating a temp directory and cleaning it up. It works.
However, all the excess files which show up untracked for git, are in the directory test/files/.
In that case, can I use os.walk() before and after the testing, to look for new files that were not there in this directory and get them cleaned? If not, can someone just nudge me to see where and how test/files is getting updated?

I do need to take a deeper look but pytest still seems to have a related open issue, so this might need some book-keeping and forced deletions.

[mfeurer]

@Neeratyoy I'm afraid that the current implementation does not work, i.e. it changes the working directory before running the tests, executing the 2nd git status outside of the git repository. Could you please create a new PR fixing this?

[Neeratyoy]

Sure, I'll take a look at this.

[Neeratyoy]

@mfeurer could you please guide me as to how I can test this locally?
When I execute bash ci_scripts/test.sh, or even check echo $TEST_DIR at the beginning of run_tests(), it gives a mkdir: missing operand error. The result of echo $TEST_DIR is blank.
How can I recreate locally how CI runs this?

executing the 2nd git status outside of the git repository.

Can I in that case store the current directory curr_dir=`pwd` and before the second git status execute cd $curr_dir ?

[mfeurer]

How can I recreate locally how CI runs this?

That's tricky. The easiest solution would be to do export TEST_DIR=/tmp/test/ prior to running bash ci_scripts/test.sh.

Can I in that case store the current directory curr_dir=pwd and before the second git status execute cd $curr_dir ?

That sounds like a good solution.

[mfeurer]

Added in #721.