Scientifically accurate date and time handling with guaranteed nanosecond precision for 32,768 years before 01 January 1900 and 32,767 years after that reference epoch.
Formally verified to not crash on operations on epochs and durations using the Kani model checking.
2.hours() + 3.seconds()), subtraction (e.g. 2.hours() - 3.seconds()), round/floor/ceil operations (e.g. 2.hours().round(3.seconds()))Epochs and Durations)no-std and const fn where possibleThis library is validated against NASA/NAIF SPICE for the Ephemeris Time to Universal Coordinated Time computations: there are exactly zero nanoseconds of difference between SPICE and hifitime for the computation of ET and UTC after 01 January 1972. Refer to the leap second section for details. Other examples are validated with external references, as detailed on a test-by-test basis.
UNIX
Time-agnostic / date-only epochs. Hifitime only supports the combination of date and time, but the Epoch::{at_midnight, at_noon} is provided as a helper function.
Put this in your Cargo.toml:
toml
[dependencies]
hifitime = "3.7"
```rust use hifitime::{Epoch, Unit, TimeUnits}; use core::str::FromStr;
{ // Initialization from system time is only available when std feature is enabled let now = Epoch::now().unwrap(); println!("{}", now); }
let mut santa = Epoch::fromgregorianutchms(2017, 12, 25, 01, 02, 14); asserteq!(santa.tomjdutcdays(), 58112.043217592590); asserteq!(santa.tojdeutc_days(), 2458112.5432175924);
asserteq!( santa + 3600 * Unit::Second, Epoch::fromgregorianutchms(2017, 12, 25, 02, 02, 14), "Could not add one hour to Christmas" );
asserteq!( santa + 60.0.minutes(), Epoch::fromgregorianutchms(2017, 12, 25, 02, 02, 14), "Could not add one hour to Christmas" );
asserteq!( santa + 1.hours(), Epoch::fromgregorianutchms(2017, 12, 25, 02, 02, 14), "Could not add one hour to Christmas" );
let dt = Epoch::fromgregorianutchms(2017, 1, 14, 0, 31, 55); asserteq!(dt, Epoch::fromstr("2017-01-14T00:31:55 UTC").unwrap()); // And you can print it too, although by default it will print in UTC asserteq!(format!("{}", dt), "2017-01-14T00:31:55 UTC".to_string());
```
Comparing times will lead to a Duration type. Printing that will automatically select the unit.
```rust use hifitime::{Epoch, Unit, Duration, TimeUnits};
let atmidnight = Epoch::fromgregorianutcatmidnight(2020, 11, 2); let atnoon = Epoch::fromgregorianutcatnoon(2020, 11, 2); asserteq!(atnoon - atmidnight, 12 * Unit::Hour); asserteq!(atnoon - atmidnight, 1 * Unit::Day / 2); asserteq!(atmidnight - at_noon, -1.days() / 2);
let deltatime = atnoon - atmidnight; asserteq!(format!("{}", deltatime), "12 h".tostring()); // And we can multiply durations by a scalar... let delta2 = 2 * deltatime; asserteq!(format!("{}", delta2), "1 days".tostring()); // Or divide them by a scalar. asserteq!(format!("{}", delta2 / 2.0), "12 h".to_string());
// And of course, these comparisons account for differences in time scales let atmidnightutc = Epoch::fromgregorianutcatmidnight(2020, 11, 2); let atnoontai = Epoch::fromgregoriantaiatnoon(2020, 11, 2); asserteq!(format!("{}", atnoontai - atmidnightutc), "11 h 59 min 23 s".tostring()); ```
Timeunits and frequency units are trivially supported. Hifitime only supports up to nanosecond precision (but guarantees it for 64 millenia), so any duration less than one nanosecond is truncated.
```rust use hifitime::{Epoch, Unit, Freq, Duration, TimeUnits};
// One can compare durations assert!(10.seconds() > 5.seconds()); assert!(10.days() + 1.nanoseconds() > 10.days());
// Those durations are more precise than floating point since this is integer math in nanoseconds let d: Duration = 1.0.hours() / 3 - 20.minutes(); assert!(d.abs() < Unit::Nanosecond); assert_eq!(3 * 20.minutes(), Unit::Hour);
// And also frequencies but note that frequencies are converted to Durations! // So the duration of that frequency is compared, hence the following: assert!(10 * Freq::Hertz < 5 * Freq::Hertz); assert!(4 * Freq::MegaHertz > 5 * Freq::MegaHertz);
// And asserts on the units themselves assert!(Freq::GigaHertz < Freq::MegaHertz); assert!(Unit::Second > Unit::Millisecond); ```
Finally, something which may come in very handy, line spaces between times with a given step.
rust
use hifitime::{Epoch, Unit, TimeSeries};
let start = Epoch::from_gregorian_utc_at_midnight(2017, 1, 14);
let end = Epoch::from_gregorian_utc_at_noon(2017, 1, 14);
let step = 2 * Unit::Hour;
let time_series = TimeSeries::inclusive(start, end, step);
let mut cnt = 0;
for epoch in time_series {
println!("{}", epoch);
cnt += 1
}
// Check that there are indeed six two-hour periods in a half a day,
// including start and end times.
assert_eq!(cnt, 7)
No software is perfect, so please report any issue or bugs on Github.
Under the hood, a Duration is represented as a 16 bit signed integer of centuries (i16) and a 64 bit unsigned integer (u64) of the nanoseconds past that century. The overflowing and underflowing of nanoseconds is handled by changing the number of centuries such that the nanoseconds number never represents more than one century (just over four centuries can be stored in 64 bits).
Advantages: 1. Exact precision of a duration: using a floating point value would cause large durations (e.g. Julian Dates) to have less precision than smaller durations. Durations in hifitime have exactly one nanosecond of precision for 65,536 years. 2. Skipping floating point operations allows this library to be used in embedded devices without a floating point unit. 3. Duration arithmetics are exact, e.g. one third of an hour is exactly twenty minutes and not "0.33333 hours."
Disadvantages:
1. Most astrodynamics applications require the computation of a duration in floating point values such as when querying an ephemeris. This design leads to an overhead of about 5.2 nanoseconds according to the benchmarks (Duration to f64 seconds benchmark). You may run the benchmarks with cargo bench.
When Durations are printed, only the units whose value is non-zero is printed. For example, 5.hours() + 256.0.milliseconds() + 1.0.nanoseconds() will be printed as "5 h 256 ms 1 ns".
```rust use hifitime::{Duration, Unit, TimeUnits}; use core::str::FromStr;
assert_eq!( format!( "{}", 5.hours() + 256.0.milliseconds() + 1.0.nanoseconds() ), "5 h 256 ms 1 ns" );
assert_eq!( format!( "{}", 5.days() + 1.0.nanoseconds() ), "5 days 1 ns" );
asserteq!( Duration::fromstr("5 h 256 ms 1 ns").unwrap(), 5 * Unit::Hour + 256 * Unit::Millisecond + Unit::Nanosecond ); ```
The Epoch is simply a wrapper around a Duration. All epochs are stored in TAI duration with respect to 01 January 1900 at noon (the official TAI epoch). The choice of TAI meets the Standard of Fundamental Astronomy (SOFA) recommendation of opting for a glitch-free time scale (i.e. without discontinuities like leap seconds or non-uniform seconds like TDB).
Epochs can be formatted and parsed in the following time scales:
{epoch}{epoch:x}{epoch:X}{epoch:e}{epoch:E}{epoch:p}{epoch:o}```rust use hifitime::{Epoch, TimeScale}; use core::str::FromStr;
let epoch = Epoch::fromgregorianutc_hms(2022, 9, 6, 23, 24, 29);
asserteq!(format!("{epoch}"), "2022-09-06T23:24:29 UTC"); asserteq!(format!("{epoch:x}"), "2022-09-06T23:25:06 TAI"); asserteq!(format!("{epoch:X}"), "2022-09-06T23:25:38.184000000 TT"); asserteq!(format!("{epoch:E}"), "2022-09-06T23:25:38.182538909 ET"); asserteq!(format!("{epoch:e}"), "2022-09-06T23:25:38.182541259 TDB"); asserteq!(format!("{epoch:p}"), "1662506669"); // UNIX seconds assert_eq!(format!("{epoch:o}"), "1346541887000000000"); // GPS nanoseconds
// RFC3339 parsing with time scales asserteq!( Epoch::fromgregorianutchms(1994, 11, 5, 13, 15, 30), Epoch::fromstr("1994-11-05T08:15:30-05:00").unwrap() ); asserteq!( Epoch::fromgregorianutchms(1994, 11, 5, 13, 15, 30), Epoch::fromstr("1994-11-05T13:15:30Z").unwrap() ); // Same test with different time systems // TAI asserteq!( Epoch::fromgregoriantaihms(1994, 11, 5, 13, 15, 30), Epoch::fromstr("1994-11-05T08:15:30-05:00 TAI").unwrap() ); asserteq!( Epoch::fromgregoriantaihms(1994, 11, 5, 13, 15, 30), Epoch::fromstr("1994-11-05T13:15:30Z TAI").unwrap() ); // TDB asserteq!( Epoch::fromgregorianhms(1994, 11, 5, 13, 15, 30, TimeScale::TDB), Epoch::fromstr("1994-11-05T08:15:30-05:00 TDB").unwrap() ); asserteq!( Epoch::fromgregorianhms(1994, 11, 5, 13, 15, 30, TimeScale::TDB), Epoch::fromstr("1994-11-05T13:15:30Z TDB").unwrap() ); ```
Leap seconds allow TAI (the absolute time reference) and UTC (the civil time reference) to not drift too much. In short, UTC allows humans to see the sun at zenith at noon, whereas TAI does not worry about that. Leap seconds are introduced to allow for UTC to catch up with the absolute time reference of TAI. Specifically, UTC clocks are "stopped" for one second to make up for the accumulated difference between TAI and UTC. These leap seconds are announced several months in advance by IERS, cf. in the IETF leap second reference.
The "placement" of these leap seconds in the formatting of a UTC date is left up to the software: there is no common way to handle this. Some software prevents a second tick, i.e. at 23:59:59 the UTC clock will tick for two seconds (instead of one) before hoping to 00:00:00. Some software, like hifitime, allow UTC dates to be formatted as 23:59:60 on strictly the days when a leap second is inserted. For example, the date 2016-12-31 23:59:60 UTC is a valid date in hifitime because a leap second was inserted on 01 Jan 2017.
Prior to the first leap second, NAIF SPICE claims that there were nine seconds of difference between TAI and UTC: this is different from the Standard of Fundamental Astronomy (SOFA). SOFA's iauDat function will return non-integer leap seconds from 1960 to 1972. It will return an error for dates prior to 1960. Hifitime only accounts for leap seconds announced by IERS in its computations: there is a ten (10) second jump between TAI and UTC on 01 January 1972. This allows the computation of UNIX time to be a specific offset of TAI in hifitime. However, the prehistoric (pre-1972) leap seconds as returned by SOFA are available in the leap_seconds() method of an epoch if the iers_only parameter is set to false.
In theory, as of January 2000, ET and TDB should now be identical. However, the NASA NAIF leap seconds files (e.g. naif00012.tls) use a simplified algorithm to compute the TDB:
Equation [4], which ignores small-period fluctuations, is accurate to about 0.000030 seconds.
In order to provide full interoperability with NAIF, hifitime uses the NAIF algorithm for "ephemeris time" and the ESA algorithm for "dynamical barycentric time." Hence, if exact NAIF behavior is needed, use all of the functions marked as et instead of the tdb functions, such as epoch.to_et_seconds() instead of epoch.to_tdb_seconds().
Huge thanks to @gwbres who put in all of the work for this release. These usability changes allow Rinex to use hifitime, check out this work. + timescale.rs: derive serdes traits when feasible by @gwbres in https://github.com/nyx-space/hifitime/pull/167 + timecale.rs: introduce format/display by @gwbres in https://github.com/nyx-space/hifitime/pull/168 + readme: fix BeiDou typo by @gwbres in https://github.com/nyx-space/hifitime/pull/169 + epoch: derive Hash by @gwbres in https://github.com/nyx-space/hifitime/pull/170 + timescale: identify GNSS timescales from standard 3 letter codes by @gwbres in https://github.com/nyx-space/hifitime/pull/171 + timescale: standard formatting is now available by @gwbres in https://github.com/nyx-space/hifitime/pull/174 + epoch, duration: improve and fix serdes feature by @gwbres in https://github.com/nyx-space/hifitime/pull/175 + epoch, timescale: implement default trait by @gwbres in https://github.com/nyx-space/hifitime/pull/176
min and max function which respectively returns a copy of the epoch/duration that is the smallest or the largest between self and other, cf. #164.>, >=, <, <=, ==, and !=. Epoch now supports init_from_gregorian with a time scape, like in Rust. Epochs can also be subtracted from one another using the timedelta function, cf. #162.test_add_durations_over_leap_seconds test.Epoch::from_str("1994-11-05T08:15:30-05:00"), cf. #73.maturin and then build with the python feature flag. For example, maturin develop -F python && python will build the Python package in debug mode and start a new shell where the package can be imported.ceil-ed, floor-ed, and round-ed according to the time scale they were initialized in, cf. #145.from_gregorian, from_gregorian_hms, from_gregorian_at_noon, from_gregorian_at_midnight.Duration::MIN (i.e. minus thirty-two centuries).no-std.regex.as_* functions become to_* and in_* also becomes to_*, cf. #155.leap_seconds() -> Option<f64> function on an instance of Epoch. Importantly, no difference in the behavior of hifitime should be noticed here: the prehistoric leap seconds are ignored in all calculations in hifitime and only provided to meet the SOFA calculations.Epoch and Duration now have the C memory representation to allow for hifitime to be embedded in C more easily.Epoch and Duration can now be encoded or decoded as ASN1 DER with the asn1der crate feature (disabled by default).Duration, which in turn guarantees that Epoch operations cannot panic, cf. #127len and size_hint for TimeSeries, cf. #131, reported by @d3v-null, thanks for the find!Epoch now implements Eq and Ord, cf. #133, thanks @mkolopanis for the PR!Epoch can now be printed in different time systems with format modifiers, cf. #130as_utc_duration in Epoch is now public, cf. #129num-traits thereby skipping the explicit use of libm. Basically, operations on f64 look like normal Rust again, cf. #128libm for non-core f64 operationsEq and some derive Ord (where relevant) #118Duration and Epoch, respectively with toparts and totai_parts #122Add ceil, floor, round operations to Epoch and Duration
Add #![no_std] support
to_parts to Duration to extract the centuries and nanoseconds of a durationEpoch from its duration and parts in TAI systemAdd pure nanosecond (u64) constructor and getter for GPST since GPS based clocks will count in nanoseconds
Errors::ParseError no longer contains a String but an enum ParsingErrors instead. This is considered possibly breaking because it would only break code in the cases where a datetime parsing or unit parsing was caught and handled (uncommon). Moreover, the output is still Display-able.
Backend rewritten from TwoFloat to a struct of the centuries in i16 and nanoseconds in u64. Thanks to @pwnorbitals for proposing the idea in #107 and writing the proof of concept. This leads to at least a 2x speed up in most calculations, cf. this comment.
Epoch by @cjordanas_jde_tdb_days() in Epoch. In version 2.2.1, the ephemeris time and TDB days were identical down to machine precision. After a number of validation cases in the rotation equations of the IAU Earth to Earth Mean Equator J2000 frame, the new formulation was shown to lead to less rounding errors when requesting the days. These rounding errors prevented otherwise trivial test cases. However, it adds an error of 40.2 nanoseconds when initializing an Epoch with the days in ET and requesting the TDB days.Note: this was originally published as 2.2.2 but I'd forgotten to update one of the tests with the 40.2 ns error.