[run]
source=torch_scatter
[report]
exclude_lines =
    pragma: no cover
    cuda
    forward
    backward
    apply
    raise
    min_value
    max_value