Running the Tests¶
Using setuptools for Testing Iris¶
Warning
The setuptools test
command was deprecated in v41.5.0. See Using Nox for Testing Iris.
A prerequisite of running the tests is to have the Python environment setup. For more information on this see Installing From Source (Developers).
Many Iris tests will use data that may be defined in the test itself, however this is not always the case as sometimes example files may be used. Due to the size of some of the files used these are not kept in the Iris repository. A separate repository under the SciTools organisation is used, see https://github.com/SciTools/iris-test-data.
In order to run the tests with all the test data you must clone the
iris-test-data
repository and then configure your shell to ensure the Iris
tests can find it by using the shell environment variable named
OVERRIDE_TEST_DATA_REPOSITORY. The example command below uses
~/projects
as the parent directory:
cd ~/projects
git clone git@github.com:SciTools/iris-test-data.git
export OVERRIDE_TEST_DATA_REPOSITORY=~/projects/iris-test-data/test_data
All the Iris tests may be run from the root iris
project directory via:
python setup.py test
You can also run a specific test, the example below runs the tests for mapping:
cd lib/iris/tests
python test_mapping.py
When running the test directly as above you can view the command line options
using the commands python test_mapping.py -h
or
python test_mapping.py --help
.
Tip
A useful command line option to use is -d
. This will display
matplotlib figures as the tests are run. For example:
python test_mapping.py -d
You can also use the -d
command line option when running all
the tests but this will take a while to run and will require the
manual closing of each of the figures for the tests to continue.
The output from running the tests is verbose as it will run ~5000 separate tests. Below is a trimmed example of the output:
running test
Running test suite(s): default
Running test discovery on iris.tests with 2 processors.
test_circular_subset (iris.tests.experimental.regrid.test_regrid_area_weighted_rectilinear_src_and_grid.TestAreaWeightedRegrid) ... ok
test_cross_section (iris.tests.experimental.regrid.test_regrid_area_weighted_rectilinear_src_and_grid.TestAreaWeightedRegrid) ... ok
test_different_cs (iris.tests.experimental.regrid.test_regrid_area_weighted_rectilinear_src_and_grid.TestAreaWeightedRegrid) ... ok
...
...
test_ellipsoid (iris.tests.unit.experimental.raster.test_export_geotiff.TestProjection) ... SKIP: Test requires 'gdal'.
test_no_ellipsoid (iris.tests.unit.experimental.raster.test_export_geotiff.TestProjection) ... SKIP: Test requires 'gdal'.
...
...
test_slice (iris.tests.test_util.TestAsCompatibleShape) ... ok
test_slice_and_transpose (iris.tests.test_util.TestAsCompatibleShape) ... ok
test_transpose (iris.tests.test_util.TestAsCompatibleShape) ... ok
----------------------------------------------------------------------
Ran 4762 tests in 238.649s
OK (SKIP=22)
There may be some tests that have been skipped. This is due to a Python decorator being present in the test script that will intentionally skip a test if a certain condition is not met. In the example output above there are 22 skipped tests, at the point in time when this was run this was primarily due to an experimental dependency not being present.
Tip
The most common reason for tests to be skipped is when the directory for the
iris-test-data
has not been set which would shows output such as:
test_coord_coord_map (iris.tests.test_plot.Test1dScatter) ... SKIP: Test(s) require external data.
test_coord_coord (iris.tests.test_plot.Test1dScatter) ... SKIP: Test(s) require external data.
test_coord_cube (iris.tests.test_plot.Test1dScatter) ... SKIP: Test(s) require external data.
All Python decorators that skip tests will be defined in
lib/iris/tests/__init__.py
with a function name with a prefix of
skip_
.
Using Nox for Testing Iris¶
Iris has adopted the use of the nox tool for automated testing on cirrus-ci and also locally on the command-line for developers.
nox is similar to tox, but instead leverages the expressiveness and power of a Python configuration file rather than an .ini style file. As with tox, nox can use virtualenv to create isolated Python environments, but in addition also supports conda as a testing environment backend.
Where is Nox Used?¶
Iris uses nox as a convenience to fully automate the process of executing the Iris tests, but also automates the process of:
building the documentation and executing the doc-tests
building the documentation gallery
running the documentation URL link check
linting the code-base
ensuring the code-base style conforms to the black standard
You can perform all of these tasks manually yourself, however the onus is on you to first ensure that all of the required package dependencies are installed and available in the testing environment.
Nox has been configured to automatically do this for you, and provides a means to easily replicate the remote testing behaviour of cirrus-ci locally for the developer.
Installing Nox¶
We recommend installing nox using conda. To install nox in a separate conda environment:
conda create -n nox -c conda-forge nox
conda activate nox
To install nox in an existing active conda environment:
conda install -c conda-forge nox
The nox package is also available on PyPI, however nox has been configured to use the conda backend for Iris, so an installation of conda must always be available.
Testing with Nox¶
The nox configuration file noxfile.py is available in the root iris
project directory, and
defines all the nox sessions (i.e., tasks) that may be performed. nox must always be executed
from the iris
root directory.
To list the configured nox sessions for Iris:
nox --list
To run the Iris tests for all configured versions of Python:
nox --session tests
To build the Iris documentation specifically for Python 3.7:
nox --session doctest-3.7
To run all the Iris nox sessions:
nox
For further nox command-line options:
nox --help
Note
nox will cache its testing environments in the .nox root iris
project directory.