10 releases
0.2.1 | Dec 20, 2023 |
---|---|
0.2.0 | Oct 31, 2023 |
0.1.1 | Sep 11, 2023 |
0.0.6 | Sep 2, 2023 |
0.0.3 | May 20, 2023 |
#105 in Date and time
80KB
1.5K
SLoC
datealgo
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): IncludeSystemTime
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 Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz:
Function | datealgo | hinnant | httpdate | humantime | time | chrono |
---|---|---|---|---|---|---|
date_to_rd | 2.1 ns | 3.3 ns | 3.3 ns | 3.6 ns | 15.1 ns | 6.5 ns |
rd_to_date | 3.2 ns | 7.6 ns | 13.5 ns | 13.5 ns | 24.3 ns | 8 ns |
datetime_to_systemtime | 5.1 ns | 8.8 ns | 9 ns | 31.3 ns | 22.8 ns | |
systemtime_to_datetime | 17.8 ns | 28.4 ns | 30.9 ns | 44.1 ns | 98.4 ns |
Reliable and reproducible microbenchmarks are extremely hard to obtain with modern processors. And even then, they are of limited use as the surrounding code will dictate a lot about the performance. These benchmarks are not meant to be authoritative, but rather illustrate the likely relative speed differences of the algorithms. Your mileage will vary, so always benchmark the real use case.
Acknowledgements
This crate is based publicly available algorithms from many sources, but most notably Cassio Neri. He has directly contributed to this crate with novel algoritms and has provided valuable feedback and suggestions.
- 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.
Releases
See CHANGELOG
License
Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.