Redesign and document the traits/functions related to numbers

[auroranockert]

In #4789 discussion was started about redesigning the traits related to numbers and I think the first step is to design and document the traits we think are required and why.

We probably also depend on #4183 and some other bugs for the actual implementation?

(I have an initial sketch/RFC here, based on the work in #4789 etc. https://gist.github.com/JensNockert/4719287)

[auroranockert]

Also related to #1284 (Complex Numbers), which should be taken into consideration.

Other types that may not fit in core but should fit into the trait system

• Fixed-Point
• Quaternions
• Polynomials
• Interval Arithmetic
• Matrixes and Vectors

[brendanzab]

Good call on making a new bug. Looking forward to seeing progress on this front - it will be of great use to maths libraries.

[auroranockert]

I made a bikeshed, https://github.com/mozilla/rust/wiki/Bikeshed-Numeric-Traits, hoping that it can grow into actual documentation at some point describing the thoughts going into the system. Feel free to add comments, fix bugs &c.

[auroranockert]

#4881, #4208, #4788, #1222, #2198, #3980, #4565, #4219, #3281 and #4231 seem like related issues, I just wanted to add them here for reference so I can see if they get magically closed in relation to this bug.

[brendanzab]

According to @catamorphism and @nikomatsakis (in this comment in #4183), we currently have a blockage at #4678, pushed back until the 0.7 milestone. We might have to figure out some workarounds for the time being and patch things up later once the issues are resolved.

[auroranockert]

Yeah, waiting for 0.7+ seems like a no-go for me too, I think we'll have to start without the operator traits and try to make sure the fix is in as early as possible for 0.7.

[pnkfelix]

�Was happy to see popcount (aka sideways-add). Any chance we could find a more self-documenting name than clz and ctz for the other two functions in the BitCount trait? (Would leadingzeros and trailingzeros be acceptable?)

[auroranockert]

I think you're right, it is a lot more clear with the longer names. I updated the bikeshed.

[brendanzab]

I'll implement pop_count, leading_zeros and trailing_zeros on a BitCount trait on my next pull request. Thanks for the suggestion!

[brendanzab]

If we're going for 'self-documenting', I wonder if either bit_count or bits_set would be better than pop_count. The description in the LLVM docs says, "The llvm.ctpop family of intrinsics counts the number of bits set in a value."

[thestinger]

Population count is self-documenting in my opinion, and it's the domain term for it.

Python has a method called bit_count and it does something completely different (counts the bits needed to represent the number - including zeroes).