Crate datealgo

source ·
Expand description

Low-level date algorithms for libraries

This library aims to provide the highest performance algorithms for date manipulation in an unopinionated way. It is meant to be used by the various date and time libraries which can then provide ergonomic and opinionated interfaces for their users.

Usage

The primary contribution of this crate for date libraries are the conversions between a day number from Unix epoch (January 1st, 1970) and a Gregorian date:

use datealgo::{rd_to_date, date_to_rd};

assert_eq!(date_to_rd((1970, 1, 1)), 0);
assert_eq!(date_to_rd((2023, 5, 12)), 19489);
assert_eq!(rd_to_date(19489), (2023, 5, 12));

For convenience, there is also converters to and from Unix timestamps:

use datealgo::{secs_to_datetime, datetime_to_secs};

assert_eq!(datetime_to_secs((1970, 1, 1, 0, 0, 0)), 0);
assert_eq!(datetime_to_secs((2023, 5, 20, 9, 24, 38)), 1684574678);
assert_eq!(secs_to_datetime(1684574678), (2023, 5, 20, 9, 24, 38));

If the std feature is enabled, there are also converters to and from SystemTime:

use datealgo::{systemtime_to_datetime, datetime_to_systemtime};
use std::time::{Duration, UNIX_EPOCH};

assert_eq!(systemtime_to_datetime(UNIX_EPOCH), Some((1970, 1, 1, 0, 0, 0, 0)));
assert_eq!(systemtime_to_datetime(UNIX_EPOCH + Duration::from_secs(1684574678)), Some((2023, 5, 20, 9, 24, 38, 0)));
assert_eq!(datetime_to_systemtime((2023, 5, 20, 9, 24, 38, 0)), UNIX_EPOCH.checked_add(Duration::from_secs(1684574678)));

Features

The crate works in no_std environments and has no allocations. Most of the functions also work in constant contexts.

  • std (default): Include SystemTime conversions

Background

There are many date and time libraries for Rust for varying use cases as the standard library doesn’t include any utilities for dealing with dates. Most of these libraries contain their own copies of date algorithms, most prominently the conversion from days since an epoch to a Gregorian calendar date (year, month, day). These algorithms have been sourced from various places with various licenses, often translated either by machine or by hand from C algorithms found in different libc variants. The algorithms are usually somewhat optimized for performance, but fall short of fastest algorithms available.

Notes

The library does not expose any kind of Date or DateTime structures, but simply tuples for the necessary values. Bounds checking is done via debug_assert only, which means the methods are guaranteed to not panic in release builds. Callers are required to do their own bounds checking. Datatypes are selected as the smallest that will fit the value.

Currently the library implements algorithms for the Proleptic Gregorian Calendar which is our current calendar extended backwards indefinitely. The Gregorian calendar defines the average year to be 365.2425 days long by defining every fourth year to be a leap year, unless the year is divisible by 100 and not by 400.

The algorithms do not account for leap seconds, as is customary for Unix time. Every day is exactly 86400 seconds in length, and the calculated times do not adjust for leap seconds between timestamps.

We define Rata Die to be integral day numbers counted from 1st of January, 1970, which is the Unix epoch. We use the abbreviation rd to concisely refer to such values. This differs from the epoch originally chosen by Howard Jacobson, but is more convenient for usage.

The Rata Die values are represented as i32 for performance reasons. The needed calculations reduce that to roughly an effective i30 integer range, which means a usable range of roughly -1,460,000 to 1,460,000 years.

Benchmarks

Results on GitHub Codespaces 8-core VM:

Functiondatealgohinnanthttpdatehumantimetimechrono
date_to_rd2.3 ns4 ns3.2 ns3.2 ns17.7 ns7.2 ns
rd_to_date3.6 ns9.3 ns11.3 ns11.3 ns18.8 ns8.2 ns
datetime_to_systemtime8.6 ns9.9 ns9.8 ns57.3 ns50.1 ns
systemtime_to_datetime14.2 ns18.9 ns19 ns54.6 ns226.4 ns

Some code has been adapted from the libraries to produce comparable benchmarks.

Acknowledgements

I do not claim original research on anything that is in this crate.

  • Cassio Neri and Lorenz Schneider: While searching for best method for date conversion, I stumbled upon a research paper which explains a novel way to optimize the performance. These algorithms have been implemented here based on the published article. This wouldn’t be the best performing date conversion library without their work.
  • Howard Hinnant: While searching for “perpetual calendar” algorithms, and having already started my library, I stumbled upon a very similar idea by Howard Hinnant. It remains one of the cleanest and simplest algorithms while still retaining excellent performance.
  • Rich Felker: The original musl __time_to_tm function has spread far and wide and been translated to many languages, and is still the basis of many of the standalone implementations littered among the libraries.
  • Many authors of newlib gmtime_r.c: The newlib implementation has evolved significantly over time and has now been updated based on the work by Howard Hinnant.

Modules

  • consts
    Convenience constants, mostly for input validation

Constants

Functions