Le 26/10/2020 à 21:47, Emmanuel Medernach a écrit : > > Le 25/10/2020 à 21:53, Marc Nieper-Wißkirchen a écrit : >> If you allow me to comment on this: >> >> Whether a portable implementation is possible or not is one thing (and >> doesn't have to be a show stopper), but it is quite another thing if >> the specification cannot make any guarantees in order to allow at >> least non-portable implementations on most existing and future Schemes >> (which, in this case, has to take nanboxing into account). > > Then maybe we could propose that 'nan-payload' > returns some marker data that may or may not be > the 51 bits, letting implementations to decide > what is the best choice for them. > > Conversely to be useful in calculations I think we > need to be able to add a function to create NaNs > with a given marker. This is exactly the NaN functions interface John proposed in his FlonumsCowan, I think this is the Right Thing to do. > Emmanuel >
