As documented in #990, I suggest to use conda instead, which is much simpler and faster than compiling from source.

If we were to add these instructions for compiling from source in this case, I suggest to prefer documenting it as another edge case in the doc rather than in the README 🙂

Coverage remained the same at 78.922% when pulling 2feb3e1 on numpy-readme-m1 into 7c56423 on master.

Thanks, I'll leave this as draft then until moving the contribution to the doc.

Moved to openfisca/openfisca-doc#274