Running tests¶
There are two basic categories of tests in TARDIS: 1) the unit tests, and 2) the integration tests. Unit tests check the outputs of individual functions, while the integration tests check entire runs for different setups of TARDIS.
The unit tests run very quickly and thus are executed after every suggested change to TARDIS. The integration tests are much more costly and thus are only executed every few days on a dedicated server.
All of them are based on the excellent astropy-setup-helpers
package and
pytest.
Running the Unit Tests¶
This is very straightforward to run on your own machine. For very simple unit tests, you can run this with:
> pytest tardis
Running the more advanced unit tests requires TARDIS Reference data that can be downloaded (tardis-refdata). Git LFS is used to download the large refdata files in the tardis-refdata repository.
However, it is not required to download the entire repository. Firstly it is important to identify the refdata files that are needed. Sometimes, it is possible that a preused fixture that is also being used in the current tests is using some refdata. So, it is advised to check for such cases beforehand.
After identifying the refdata files to be used in the unit tests, those particular
files can be downloaded using git lfs
> git lfs pull --include=filename
It is important to maintain the same directory structure as the tardis-refdata repo i.e. the lfs files should be in the same directory tree exactly as in tardis-refdata repository.
Finally, the tests can be run using the following command
> pytest tardis --tardis-refdata=/path/to/tardis-refdata/
Or, to run tests for a particular file or directory
> pytest tardis/path/to/test_file_or_directory --tardis-refdata=/path/to/tardis-refdata/
Warning
The tests workflow runs on
pull requests and on
push events.
To prevent leaking LFS quota, tests have been disabled on forks.
If, by any chance, you need to run tests on your fork, make sure to run the tests workflow on master branch first.
The LFS cache generated in the master branch should be available in all child branches.
You can check if cache was generated by looking in the Restore LFS Cache
step of the workflow run.
Cache can also be found under the “Management” Section under “Actions” tab.
Running Syrupy Tests¶
The tests module is currently being restructured to use Syrupy.
These tests generate individual HDF(for Pandas) and .npy
or .npz
(for NumPy) files(or snapshots) for each test case.
For other objects, the plugin serialises them and saves them as .ambr
files. There is a custom extension in the
main local conftest file to get this to work.
The plugin currently only supports assert_allclose
for NumPy and assert_series_equal
and assert_frame_equal
for
Pandas. Snapshots can be generated by using the same --generate-reference
flag and are compared automatically.
You however need to provide the location to your snapshot directory using the --tardis-snapshot-data
flag.
The snapshots are also saved in tardis-sn/tardis-regressions.
Generating Plasma Reference¶
You can generate Plasma Reference by the following command:
> pytest -rs tardis/plasma/tests/test_complete_plasmas.py
--tardis-refdata="/path/to/tardis-refdata/" --generate-reference
Running the Integration Tests¶
These tests require reference files against which the results of the various tardis runs are tested. So you first need to either download the current reference files (here) or generate new ones.
Both of these require a configuration file for the integration tests:
atom_data_path: "~/projects/tardis/integration/"
# Path to directory where reference HDF files will be generated and
# saved during the test run. Use "--generate-reference" flag in command
# line args for the purpose, for other cases this will denote path
# to the directory containing reference HDF files.
reference: "~/projects/tardis/integration/"
# Speeds up test execution by reducing amount of packets per iteration,
# useful for debugging problems in testing infrastructure itself.
# Use "--less-packets" in command line args, for other cases this will be
# simply ignored. This section is not mandatory.
less_packets:
no_of_packets: 400
last_no_of_packets: 1000
Inside the atomic data directory there needs to be atomic data for each of
the setups that are provided in the test_integration
folder.
If no references are given, the first step is to generate them.
The --less-packets
option is useful for debugging purposes and will just
use very few packets to generate the references and thus make the process much
faster — THIS IS ONLY FOR DEBUGGING PURPOSES. The -s
option ensures that
TARDIS prints out the progress:
> python setup.py test --args="--integration=integration.yml -m integration
--generate-reference --less-packets"
To run the test after having run the --generate-references
, all that is
needed is:
> python setup.py test --args="--integration=integration.yml -m integration
--less-packets" --remote-data