Oh boy… good catch.
Here was the output:
LANG=fr_FR.UTF-8
LANGUAGE=
LC_CTYPE="fr_FR.UTF-8"
LC_NUMERIC="fr_FR.UTF-8"
LC_TIME="fr_FR.UTF-8"
LC_COLLATE="fr_FR.UTF-8"
LC_MONETARY="fr_FR.UTF-8"
LC_MESSAGES="fr_FR.UTF-8"
LC_PAPER="fr_FR.UTF-8"
LC_NAME="fr_FR.UTF-8"
LC_ADDRESS="fr_FR.UTF-8"
LC_TELEPHONE="fr_FR.UTF-8"
LC_MEASUREMENT="fr_FR.UTF-8"
LC_IDENTIFICATION="fr_FR.UTF-8"
LC_ALL=
French does use comma as a decimal separator so using export LC_ALL=C.UTF-8 made it run. However it is still unclear to me how the MPI rank 0 could read it in the other context (which added a layer of Python on top of this). I’ll check if this also solves the other problem. Edit: it does. So loading ovito changes the locale used by the C code.