Testing code with Scipp#


It is possible to write tests in terms of regular functions provided by Scipp such as scipp.identical() or scipp.allclose(). While this works, error messages produced with those functions contain little information and it is often necessary to inspect the comapred objects manually in case of a test failure.

For this reason, the scipp.testing.assertions (also available in scipp.testing) module provides a function to compare objects:

scipp.testing.assert_identical(a, b)

It is intended for use with pytest and is equivalent to

assert scipp.identical(a, b, equal_nan=True)

but produces errors that precisely indicate why the inputs are not equal.

There currently is no equivalent for scipp.allclose().


To get the most out of it assert_identical, you need to configure pytest with the following. For example, by placing it in your conftest.py:


This tells pytest that scipp.testing.assertions contains test assertions. pytest will then change those assertions the same way as assertions in test files to produce more detailed error messages.

On top, we recommend the following (also in conftest.py) to improve errors messages further:

def pytest_assertrepr_compare(op: str, left: Any, right: Any) -> List[str]:
    if isinstance(left, sc.Unit) and isinstance(right, sc.Unit):
        return [f'Unit({left}) {op} Unit({right})']
    if isinstance(left, sc.DType) or isinstance(right, sc.DType):
        return [f'{left!r} {op} {right!r}']

See also

Scipp’s own conftest.py.

Input generation#

Scipp’s containers have many flexible components that can produce many different and sometimes unexpected combinations. It is difficult to write tests that cover all cases by hand. So Scipp provides tools to generate data structures in the form of Hypothesis strategies in scipp.testing.strategies. See in particular scipp.testing.strategies.variables(). Support for anything beyond variables is currently limited and binned data is not supported.

To use the strategies, install hypothesis and, in pytest, write, e.g.,

from hypothesis import given
import scipp.testing.strategies as scst

def test_abs_preserves_shape(var):
    assert abs(var).shape == var.shape

The variables strategy generates arbitrary non-binned variables with different units, dims, shapes, dtypes, values, and variances. It has several arguments that can be used to steer generation. For example, to generate only two-dimensional variables with floating point dtypes, use

from hypothesis import given, strategies as st
import scipp.testing.strategies as scst

                      dtype=st.sampled_from(('float64', 'float32'))))
def test_mean_reduces_ndim(var):
    assert var.mean(dim=var.dims[0]).ndim == 1