Fabian Giesen

@rygorous @aras_p Hehe, @sheredom pointed out the same thing. But glad he's considering new representations, there's room for improvement!

@alexjc @aras_p @sheredom I don't think this is new though? People used log-scale FP (as opposed to semi-log like IEEE) for a while.

@rygorous @sheredom It's not scale that makes unums interesting, but the fact they are rigorously defined operations on sets (no NAN).

@alexjc @sheredom Nope! The set operations are on the *SORN*s. The unums are just a lattice on the real line, like regular floats.

@alexjc @sheredom (well, projective real line in his case.)

@alexjc @sheredom If you want the set arithmetic on 8-bit unums, you get 256-bit SORNs. For 16-bit unorms: 65536-bit SORNs!

@alexjc @sheredom (his only, laconic, comment on the matter: "General SORNs need run-length encoding". Yuuup.)

@alexjc @sheredom He mentions in passing a SORN encoding for "connected sets" (storing start/end pts for an interval, probably? Not clear.)