@khinsen

#Maneage directory policy allows neat shifting of output dirs. Intra-install (no recompilation) tracing shows exact results-reproducibility backwards :) to the step running calc_dens.py [1]. This uses Voronoi + ConvexHull from scipy.spatial. Re-running this step via our workflow [2] gives different results on the same N-body simulation.

Are scipy.spatial #Voronoi or #ConvexHull known to give non-reproducible results?

@civodul

[1] https://archive.softwareheritage.org/browse/revision/41507a706e16f04075c0c35853ddc226d3c72df5/?path=reproduce/analysis/python/calc_dens.py

[2] https://archive.softwareheritage.org/browse/revision/41507a706e16f04075c0c35853ddc226d3c72df5/?path=reproduce/analysis/make/calc_densities.mk