I came up with an algorithm and implemented it in Rust.
I want to ensure the code is of sufficient quality before releasing it on Crates.io and GitHub. You can also find more info about the algorithm and its purpose there.

I’m eager to hear your feedback and thoughts and am open to ANY suggestions that may help me improve the code further.
The code should be performant, clean, easy to use, idiomatic, etc.
I’m also looking for potentially better names for certain variables and stuff.

Note that some things have changed a bit since the last release (for example - the hash calculation has been changed from blake3(blake3(OBJECT) || blake3(DATE)) to blake3::keyed_hash(DATE, OBJECT) to improve performance by eliminating 2 out of 3 hash calculations), but the README is still valid (and hopefully at least somewhat decent source of info about the algorithm - I did my best trying to explain stuff, but I’m still looking to improve it further in the future).

The functions can panic if they are given a date where the year is outside of the 0000 to 9999 range.
I’m not sure if I should make the function return the result/option instead of panicking since I’d like to avoid having to unwrap() the output and would prefer if the function returned just a simple u32/chrono::NaiveTime (but I’m open to changing my mind on this one).

I’m also curious if there is a cleaner way to create a key from a date. Previously, I’ve been using format(“%Y-%m-%d”) to convert it into a string slice and copy it into the key array, but I’ve found that approach to be very slow (over 50% of the CPU time was being spent just for that alone), so I opted out for an approach you can see below.

The newest function rdvt() is not yet documented anywhere, but I’ll do my best to explain it here:   In addition to an object and a date (same as rdv()), it also takes a rank (which is a positive integer - u32).
The function calculates the keyed blake3 hash of an object, using the date and a rank as key, and uses the resulting (pseudorandom) bits to generate and return a random uniform time (chrono::NaiveTime) between 0h and 24h (down to nanosecond precision).
Time is calculated by taking the first bit of the hash and in case it’s a binary one - 12h is added to the time, then we add 6h if the 2nd bit of the hash is a one, 3h for the 3rd bit, 1.5h for 4th and so on until the increment reaches the small enough value where it doesn’t contribute anything to the time (when it becomes less than 1ns, essentially).   This means if all of the bits in the hash were zeros - time would be zero, and if they were all ones - time would be 23:59:59:999:999:999h (the very last and highest possible value). The code short-circuits and stops earlier than going through all 256 bits since we usually only need around 46 bits before the increment becomes smaller than 1ns (the code stops only in case the sum of tiny sub 1ns increments can’t contribute enough to change the last digit in the total time (even if all of the rest of the bits in the hash were to be ones))):

Feel free to ask ANY questions regarding the code, the algorithm, its function, use cases, or anything else you’d like explained or clarified.

Here’s the code (I haven’t written any tests yet, so they are not included for review):

//! The official Rust implementation of the [RANDEVU](https://github.com/TypicalHog/randevu) algorithm
//!
//! # Example
//! ```rust
//! use chrono::Utc;
//! use randevu::{rdv, rdvt};
//!
//! fn main() {
//!     let object = "THE_SIMPSONS";
//!     let date = Utc::now();
//!     let rdv = rdv(object, &date);
//!     let rdvt = rdvt(0, object, &date);
//!
//!     println!("Object {} has RDV{} today with RDVT0 at {:?}", object, rdv, rdvt);
//! }
//! ```

use blake3;
use chrono::{DateTime, Datelike, NaiveTime, TimeDelta, Utc};
use itoa;

/// Returns the 32-byte KEY `[u8; 32]` created from a given DATE `&DateTime<Utc>` and an optional RANK `Option<u32>`
fn create_key(date: &DateTime<Utc>, rank: Option<u32>) -> [u8; 32] {
    let mut key = [0u8; 32];

    let mut year = Datelike::year(date);
    let mut month = Datelike::month(date);
    let mut day = Datelike::day(date);

    let mut year_len = 4;
    let mut prefix_len = 0;

    // Add a prefix (-/+) if the year is not between 0 and 9999 (-YYYY-MM-DD / +YYYY-MM-DD)
    if year < 0 {
        key[0] = b'-';
        prefix_len = 1;

        year = year.abs(); // Make year positive
    } else if year > 9999 {
        key[0] = b'+';
        prefix_len = 1;
    }

    // Adjust year_len for very large years (both positive and negative)
    if year > 9999 {
        year_len += 1;
        if year > 99999 {
            year_len += 1;
        }
    }

    let full_year_len = prefix_len + year_len;

    // If a rank is provided, write it into the key after the date, separated by an '_'
    if rank != None {
        let mut buffer = itoa::Buffer::new();
        let rank_str = buffer.format(rank.unwrap());
        key[7 + full_year_len..7 + full_year_len + rank_str.len()]
            .copy_from_slice(&rank_str.as_bytes()[..rank_str.len()]);

        key[6 + full_year_len] = b'_';
    }

    // Write the day into the key
    key[5 + full_year_len] = b'0' + (day % 10) as u8;
    day /= 10;
    key[4 + full_year_len] = b'0' + day as u8;

    key[3 + full_year_len] = b'-';

    // Write the month into the key
    key[2 + full_year_len] = b'0' + (month % 10) as u8;
    month /= 10;
    key[1 + full_year_len] = b'0' + month as u8;

    key[full_year_len] = b'-';

    // Write the year into the key
    for i in (prefix_len..full_year_len).rev() {
        key[i] = b'0' + (year % 10) as u8;
        year /= 10;
    }

    key
}

/// Returns the RDV value `u32` for an OBJECT `&str` on a specific DATE `&DateTime<Utc>`
///
/// **RDV = number of leading zero bits in blake3::keyed_hash(key: DATE, data: OBJECT)**
pub fn rdv(object: &str, date: &DateTime<Utc>) -> u32 {
    let hash = blake3::keyed_hash(&create_key(date, None), object.as_bytes());

    // Count the number of leading zero bits in the hash
    let mut rdv = 0;
    for &byte in hash.as_bytes() {
        rdv += byte.leading_zeros();

        if byte != 0 {
            break;
        }
    }

    rdv
}

/// Returns the RDVT time `DateTime<Utc>` of a given RANK `u32` for an OBJECT `&str` on a specific DATE `&DateTime<Utc>`
pub fn rdvt(rank: u32, object: &str, date: &DateTime<Utc>) -> DateTime<Utc> {
    let hash = blake3::keyed_hash(&create_key(date, Some(rank)), object.as_bytes());

    // Calculate the time using bits from the hash
    let mut total: f64 = 0.0;
    let mut increment = 12.0 * 60.0 * 60.0 * 1_000_000_000.0; // 12h in nanoseconds
    for (i, byte) in hash.as_bytes().iter().enumerate() {
        for j in (0..8).rev() {
            let bit = (byte >> j) & 1;
            if bit == 1 {
                total += increment;
            }
            increment /= 2.0;
        }
        // Stop once increments become too small to affect the total
        if i > 4 && (2.0 * increment) < (1.0 - total.fract()) {
            break;
        }
    }

    // Construct the RDVT time from total
    let rdvt = date.with_time(NaiveTime::MIN).unwrap() + TimeDelta::nanoseconds(total as i64);

    rdvt
}

  • lad@programming.dev
    link
    fedilink
    arrow-up
    0
    ·
    6 months ago

    Regarding NaiveDate: if there is a type for date and time that enforces UTC or includes timezone information, use it. Documenting that the user must pass UTC time is a weak spot, whatever you can enforce in types, you should enforce in types.

    Regarding the stopping condition: yeah, I got it, it may be a good place to add a comment about why (although it is quite straightforward, so I might have been too stupid with that)

    Regarding the macro usage: well, I’m not sure how that should be, but I imagine using a macro to precalculate what are possible coefficients multiplied by increment. Since coefficients will be calculated in compile-time, you may ignore the extra precision (it will not require additional computation) and only check per byte.

    Also, it may be so, that removing precision check and doing the math for all the positions you care about will be more efficient because of how optimisations work. Less branching is usually better, and separate bytes may be processed in parallel. That depends on the compiler and the target processor, and I’m far from being expert, so you may want to research that.

    • TypicalHog@lemm.eeOP
      link
      fedilink
      English
      arrow-up
      2
      ·
      edit-2
      6 months ago

      Hi, I’ve modified my code and implemented your suggestion about enforcing UTC with types. I edited the post to include the newest code. I’m just not sure if I should still call the variables/arguments “date” since now it’s a DateTime.

      • lad@programming.dev
        link
        fedilink
        English
        arrow-up
        2
        ·
        6 months ago

        If you feel like datetime or date_time would be clearer, go with it

        I will try to take a fresh look and tell how it feels some time later