Issues with the f8-storage-class
John Cowan 09 Apr 2026 23:18 UTC
1. It is not yet an IEEE standard.
2. There are no less than seven formats in the draft, with from 1 to 7
bits of precision (yes, the binary8p7 format can only represent
integers). All of them have +inf.0 and -inf.0, but no -0.0 value and
only a single +nan.0 value. There are also a number of non-IEEE
formats in use.
I think that before we standardize these as *numeric* vectors, we need
to wait for more of a shakeout, so I suggest that `f8-storage-class`
be deleted from SRFI 231. For now, users can store them in s8vectors
or u8vectors and do their own conversions.