kkappler
commented
8 months ago If we want to have access to the synthetic data we have to move the data within the PyPI scope. This means:
-
[X] either adding
__init__.pyfiles inside tests/ folder or moving data to top-level data/ folder I createdaurora/data/and moved the data that was in tests there, and set the paths to operate on that folder. Note that this does not get us around the need to mkdir for test, but it should at least make the data available. However, for reasons I don't understand (see next bullet) the data are not available after a pip install.Existing options include moving the data folder down a level, or packing the synthetic data into mt_examples. I opted for the later, which meant I had to add asource_folderoption in make_mth5_from_ascii. -
[X] Update MANIFEST.in to contain data folder. When doing this I note that tests is already in the mainfest. I do not understand why tests are in the build/ folder but not in the pip install.
What is currently breaking however is the execution of pathlib's mkdir() command when the places for the test results are being built. Since there are no tests in PyPI or conda-forge , we should be able to get around this by:
- [X] executing the mkdir command inside the setUpClass of the test methods, rather than pre-making the directories in test_utils (OR the folders could also be prebuilt inside tests, with a committed README so that the folders exist in github already) The solution I went with was hybrid, I import a class that can mkdirs into the tests, and mkdir only on executing the tests. That should stop the mkdir errors on gadi when importing methods.
- [X] Handle missing files: config/emtf_band_setup folder is empty, except for the init.py file. This seems to be be because init.py only tags .py files. The .cfg files need to be added to MANIFEST.in
Participants will be running their notebooks from the installed location, but will not have write access to the aurora install directory.
The Jupyter notebooks must be run in that state.