Trace processing functions can (and do) change the dtype of traces they process. Not of the inputs but of the outputs. This should be at least documented and a sensible strategy should be chosen to:
Make sure the results make sense and have enough precision.
Make sure the results are not unnecessarily large/precise.
(Potentially) Follow similar conventions as numpy/scipy to not surprise people, as we usually just wrap some of their functionality.
Trace processing functions can (and do) change the dtype of traces they process. Not of the inputs but of the outputs. This should be at least documented and a sensible strategy should be chosen to: