nanval

A no_std, zero-dependency crate for the creation and handling of NaN-tagged 64-bit floating-point values.

Inspired by this article and this crate.

How does this work?

When a 64-bit floating-point number is set to NaN/0x7FF8000000000000, its bits are as follows:

s111 1111 1111 1qxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx
^               ^\____________________________________________________________/
|               |                             ^
| Sign Bit      | Quiet Bit                   | Data Bits

As long as the data bits aren't all set to 0, indicating the original/sentinel NaN value, they can be literally anything else! This gives us 50 bits to mess with/use as we please...

UInts / Unsigned Integers

Look at the module crate::uint for this.

TODO: Add explanation.

Cells / Pointers

Look at the module crate::cell for this.

Since it doesn't matter what the sign-bit s is set to, we can use it as a flag/marker that indicates that the value is some kind of cell or ptr.

Combine this with the fact that basically all x64-platforms only use the lower 48 or 50 bits for addressing (ignoring CHERI shenanigans), we are left with 3 bits (that includes the 'quiet' bit) to store some kind of type-tag for the cell; look at the crate::cell::CellTag.

References

• https://float.exposed/0x7ff8000000000000
• https://sean.cm/a/nan-boxing
• https://github.com/Marwes/nanbox/blob/master/src/lib.rs
• https://github.com/SerenityOS/serenity/blob/master/Userland/Libraries/LibJS/Runtime/Value.h