Due to the use of VCL based metaprogramming and the way CPU dispatching is done in distopia, some very high precision tests may return different results based on the CPU you have.
For example the analysis/test_distances.py::TestDist::test_pairwise_dist_offset_effect test fails on my ancient local machine but is fine on CI in #3914 with a relative difference of 10^-7 (insignificant IMO).
This shouldn't be an issue as distopia is an optional dependency but I thought I would draw this to people's attention.
See #3914, addition of distopia did not require accuracy thresholds to be changed on any tests on any of the CI runners, so doesn't look like much of an issue (at least for now).
Discussion point for #3914.
Due to the use of VCL based metaprogramming and the way CPU dispatching is done in distopia, some very high precision tests may return different results based on the CPU you have.
For example the
analysis/test_distances.py::TestDist::test_pairwise_dist_offset_effect
test fails on my ancient local machine but is fine on CI in #3914 with a relative difference of 10^-7 (insignificant IMO).This shouldn't be an issue as distopia is an optional dependency but I thought I would draw this to people's attention.