In looking at NZDir to try to figure out why the process_chunk combination was failing, I noticed that there are no test cases for NZDir. Strangely, codecov is just ignoring the file altogether, so it's not affecting the coverage report. There were tests of NZDir before it was moved to rail_sklearn, see https://github.com/LSSTDESC/rail_attic/blob/main/tests/estimation/test_nzestimators.py
Not sure why this test was not copied over.
In looking at NZDir to try to figure out why the process_chunk combination was failing, I noticed that there are no test cases for NZDir. Strangely, codecov is just ignoring the file altogether, so it's not affecting the coverage report. There were tests of NZDir before it was moved to rail_sklearn, see https://github.com/LSSTDESC/rail_attic/blob/main/tests/estimation/test_nzestimators.py
Not sure why this test was not copied over.