Epoch::from_time_of_week returns wrong time in GPST

[ljbade]

When using Epoch::from_time_of_week to initialise Epoch from a GPS week number and time of week, the resulting epoch has a 19 second offset. Below is a sample program that shows the bug:

use hifitime::{Epoch, TimeScale};

fn main() {
    let t = Epoch::from_time_of_week(1983, 0, TimeScale::GPST);
    println!("{:#?}", t);
    // Should print "2018-01-07T00:00:00 GPST"

    let t = Epoch::from_time_of_week(1982, 604800000000000 - 19000000000, TimeScale::GPST);
    println!("{:#?}", t);
}

Currently this prints:

2018-01-07T00:00:19 GPST
2018-01-07T00:00:00 GPST

[ljbade]

Also, as a feature request it would be handy to have a version of from_time_of_week and from_time_of_week that works on f64 seconds instead of nanoseconds

[ChristopherRabotin]

Thanks for the bug report. This looks similar to an issue we had in the PR that added GPST. I know we fixed it, but I forget how right now. How urgently does this bug need fixing for you?

…

On Sat, Mar 18, 2023, 03:44 Leith Bade ***@***.***> wrote: Also, as a feature request it would be handy to have a version of from_time_of_week and from_time_of_week that works on f64 seconds instead of nanoseconds — Reply to this email directly, view it on GitHub <#209 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABEZV2HHXI5MQOOCWVG6UODW4V7YPANCNFSM6AAAAAAV7J3MJA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ljbade]

It's not urgent, I was able to work around it by just subtracting the 19 second offset to value I give to from_time_of_week

[ljbade]

Possibly related to this bug, if you provide the same time in two different GNSS timescales you get the wrong values in UTC. For example if I use GPST and BDT then I get the same incorrect time in UTC for both.

A test:

se hifitime::{Epoch, TimeScale};

fn main() {
    let t = Epoch::from_time_of_week(1983, 0, TimeScale::GPST);
    println!("{:#?}", t);
    // Should print "2018-01-07T00:00:00 GPST"

    let t = Epoch::from_time_of_week(1982, 604800000000000 - 19000000000, TimeScale::GPST);
    println!("{:#?}", t);


    let t_gps = Epoch::from_gregorian(2023, 03, 12, 00, 00, 00, 0, TimeScale::GPST);
    println!("{}", t_gps);
    // Should print "2018-01-06T23:59:42 UTC"

    let t_bdt = Epoch::from_gregorian(2023, 03, 12, 00, 00, 00, 0, TimeScale::BDT);
    println!("{}", t_bdt);
    // Should print "2018-01-06T23:59:56 UTC"

    let delta_t = t_bdt - t_gps;
    println!("{}", delta_t.in_seconds());
    // Should print "14"
}

This prints currently:

2023-03-11T23:59:23 UTC
2023-03-11T23:59:23 UTC
0

The UTC time printed seems to be always 37 seconds off from the input time. 37 seconds I think is also the difference between UTC and TAI, so perhaps both times are being interpreted as TAI?

[ChristopherRabotin]

Thanks for the report, that does look like there is a wrong interpretation of the leap seconds.

@gwbres Any thoughts on this issue?

[gwbres]

Thank you @ljbade for the bug report !

If i remember correctly, 19 is the amount of leap seconds when GPST was initiated (1980 Jan 5th to 6th midnight).
My first guess would be a problem of bad TAI referencing? we're supposed to always refer to TAI in our calculations

[ljbade]

Yes GPS is 19 seconds offset from TAI, and BDT is 33 seconds.

[ChristopherRabotin]

Hey @ljbade , if that's OK, I'll work on this for version 4.0.0. As explained in #237 , I think that things get confused in between the GNSS time scales.

[ChristopherRabotin]

Hi @ljbade ,

I think that I've fixed the bug in #303 . Here is the test I'm about to add to the regression tests and the output I got. The first case matches exactly what you said it should be, and I'd like your confirmation that the second output is also correct.

Thanks

#[test]
fn regression_test_gh_209() {
    let t = Epoch::from_time_of_week(1983, 0, TimeScale::GPST);

    println!("{t}");
    // Should print

    assert_eq!(format!("{t}"), format!("2018-01-07T00:00:00 GPST"));

    let t = Epoch::from_time_of_week(1982, 604800000000000 - 19000000000, TimeScale::GPST);
    println!("{t}");
}

Output:

---- regression_test_gh_209 stdout ----
2018-01-07T00:00:00 GPST
2018-01-06T23:59:41 GPST


successes:
    regression_test_gh_209

[ChristopherRabotin]

Done in #305 .