Hexadecimal Floating Point

[i-am-linja]

There is a standard hexadecimal syntax for floating point numbers: 0x*.*p*, in which the hex numeral after the p represents a power of two by which to multiply the pointed hex numeral before it. I think it would be good if Janet had this syntax as well.

[sogaiu]

I think there is also an explanation here along with a specific example.

Increasing the exponent by 1 multiplies by 2, not 16: 20p0 = 10p1 = 8p2 = 4p3 = 2p4 = 1p5. Usually, the number is normalized so that the hexadecimal digits start with 1. (zero is usually 0 with no P).

Example: 1.3DEp42 represents 1.3DE(base 16) × 2^42(base 10).

[bakpakin]

We do already support something similar with the & character. For example, 0xAB.CD&5 is equivalent to 0xAB.CD * 16^5, or equivalently 0xABCD * 16^3.

Would be fairly easy to special case the 0x prefix to handle a different exponentiation, but it's not strictly needed. Also where is this standard syntax coming from, is it just C++ or something similar? IEEE?

[sogaiu]

The following is from Wikipedia:

Just as decimal numbers can be represented in exponential notation, so too can hexadecimal numbers. P notation uses the letter P (or p, for "power"), whereas E (or e) serves a similar purpose in decimal E notation.

...

P notation is required by the IEEE 754-2008 binary floating-point standard and can be used for floating-point literals in the C99 edition of the C programming language.

via the first and last paragraphs of the aforementioned section.

There are some footnotes at the original article that I did not reproduce here, I thought it might be better to see them in context.

[bakpakin]

Thanks for the reference, I was able to get a quick implementation working. I'm pretty confident it's fine, but I will add some more testing at some point before merging.

[sogaiu]

On a related note, I verified a variety of values at janet's repl and recorded them as tests for a tree-sitter grammar here.