IntelPython / dpnp / 38588419042

83%

push

github

Run dpnp.tensor tests in coverage workflow (#3094)

This PR extends the coverage workflow to also run `dpnp.tensor` tests,
which are currently excluded by `SKIP_TENSOR_TESTS=1`.

Because of that, `dpnp/tensor` code is measured only through what dpnp
tests happen to reach, and some lines end up covered by chance.

For example, line 809 of `dpnp/tensor/_elementwise_common.py` is reached
deterministically only by
`dpnp/tests/tensor/elementwise/test_add.py::test_add_order`. In the
coverage run it is reached only by `TestDotFor0Dim` in the third-party
tests, and only when the random dtype-pair sampling of
`for_all_dtypes_combination` happens to include a `float32` x
`int32`/`int64` pair. The line is missed in about 1/3 of runs (251 of
744 Coveralls builds since `dpnp.tensor` was introduced in #2856), which
shows up as ±0.003% coverage noise on PRs without any source changes.

Note, enabling `--tensor-cpp-coverage` in CI is left for a follow-up. It
may also need `-ignore-filename-regex` rules for `libtensor` kernel
headers, similar to the existing ones for `backend/kernels`.

1608 of 3000 branches covered (53.6%)

Branch coverage included in aggregate %.

29395 of 34528 relevant lines covered (85.13%)

19623.98 hits per line