reproducibility in s_sfno wrt to decomp #909
Description
General issue (not with your code): To achieve reproducibility independent of decomposition, I think we're going to need to initialize the global random state (with the same seed on each rank, or on root and then scatter) and then sub-select the local slice. This would need to be as a flag we basically just use for testing, since in production especially at high resolutions this isn't desired, and gaussian noise with a different seed on each rank is perfectly fine (I think Distributed does seed randomness differently on each rank? Maybe?)
Feel free to open issues related to this if you want to track it.
Originally posted by @mcgibbon in #900 (comment)