-
Notifications
You must be signed in to change notification settings - Fork 68
Numbers
Our numerical models could use substantial improvements.
The Hacker's guide should be updated to include a description of:
- our numerical classes and how they relate to each other
- how / where they are implemented
- how to add a new one
Having names like <integer> which may be 30 bits or 62 depending on architecture, or <double-integer> (always 64 bit, but not valid for values that fit into an <integer>) and so on gets very confusing.
It would be nice to consider at least having the raw types correspond to how large they are in bits, like <raw-32-bit-integer>.
This is an open question and not a certainty.
It would be nice if we had a working <raw-double-integer> that was a 64 bit value and that we could use with <double-integer> in the future.
Related to the above, but we need an easy way to pass 64 bit integers in and out of the C-FFI for interop with C APIs.
We suspect that using <double-float> has more overhead when storing values back to slots than when using <float>.
This needs investigation.
Fill in more here.
Fill in more here.