Reconfigure offset calculation to fix UTC rollover. Fixes: #81, #82, and #83,

[shanedoolane]

In my tests validated against real data, this fixes sunset/sunrise issues for the location i'm working with detailed in #83.

offset comparison no longer causes previous day to be returned when local time rolled over UTC date

Test:

start_date = datetime.date(2023, 1, 1)
latitude = 37.4
longitude = -122.1
observer = astral.sun.Observer(latitude, longitude)

def daylight_hours(current_date):
    sunrise = astral.sun.sunrise(observer, date=current_date)  # returns UTC
    sunset = astral.sun.sunset(observer, date=current_date) # returns UTC

    daylight_hours = (sunset - sunrise).total_seconds() / 3600
    return daylight_hours

Old:

New:

[shanedoolane]

add 5202da6 to apply offset and block value error from being thrown if the date doesn't match by applying delta term. appears to work for PST:

checking date: 2023-06-30 ====================================================
====>sunrise: 2023-06-30 12:XXX+00:00
====>sunset: 2023-07-01 03:XXX+00:00
daylight hours: 14.XXX
checking date: 2023-07-01 ====================================================
====>sunrise: 2023-07-01 12:XXX+00:00
====>sunset: 2023-07-02 03:XXX+00:00
daylight hours: 14.XXX

[shanedoolane]

Ok maybe not as simple of a fix as I thought. Went to test some more locations and miraculously found an edge case that fails the UTC rollover conditions. I randomly picked coords point but somehow stars aligned:

Observer:

# Set location
latitude = 41.908
longitude = -87.655

# Create observer object
observer = astral.sun.Observer(latitude, longitude)

As we can see on 3/18 local date in chicago, sunset is basically at midnight UTC on 3/18. I think because of a floating operation, the offset condition is applied and it bumps the date back to 3/17 UTC causing a singularity. Not sure how best to fix this.

date:  2023-03-15 , sunrise: 2023-03-15 12:02:58.922185+00:00  sunset: 2023-03-15 23:56:44.442507+00:00  daylight hrs:  11.895977867222221
date:  2023-03-16 , sunrise: 2023-03-16 12:01:16.919834+00:00  sunset: 2023-03-16 23:57:52.582803+00:00  daylight hrs:  11.94323971361111
date:  2023-03-17 , sunrise: 2023-03-17 11:59:34.692298+00:00  sunset: 2023-03-17 23:59:00.542142+00:00  daylight hrs:  11.990513845555554
date:  2023-03-18 , sunrise: 2023-03-18 11:57:52.270697+00:00  sunset: 2023-03-17 23:59:00.542142+00:00  daylight hrs:  -11.981035709722223
date:  2023-03-19 , sunrise: 2023-03-19 11:56:09.685995+00:00  sunset: 2023-03-20 00:00:08.330443+00:00  daylight hrs:  12.066290124444444
date:  2023-03-20 , sunrise: 2023-03-20 11:54:26.969014+00:00  sunset: 2023-03-21 00:02:23.381227+00:00  daylight hrs:  12.132336725833333
date:  2023-03-21 , sunrise: 2023-03-21 11:52:44.150441+00:00  sunset: 2023-03-22 00:03:30.716642+00:00  daylight hrs:  12.179601722500001
date:  2023-03-22 , sunrise: 2023-03-22 11:51:01.260846+00:00  sunset: 2023-03-23 00:04:37.920994+00:00  daylight hrs:  12.226850041111112

[jlopezarriaza]

@shanedoolane and @sffjunkie, I'm curious if there's any movement on this or if there are plans on merging this to master?

Happy to try to help wherever I can to solve this edge case.

[shanedoolane]

@jlopezarriaza haven't gotten around to fixing yet. Want to pull the branch and look at it?

[jlopezarriaza]

I'm happy to check it out. I found a workaround by doing things in local time zone rather than UTC, but that doesn't seem optimal.

[Voutsi]

Hi @shanedoolane
thanks a lot for the effort, I am also very interested on this fix.
I tried the PR branch with the following code snippet

from datetime import datetime
import astral
from astral.sun import sun

observer = astral.Observer(latitude=-24.6272, longitude=-70.4039, elevation=2200)
timestamp =  datetime(2023, 9, 23)
helios = sun(observer, timestamp, dawn_dusk_depression=18)
dusk = helios['dusk']
dawn = helios['dawn']
print ("dawn", dawn, "dusk", dusk)

a first comment is that a small fix is needed in the functions calculating dusk and dawn in order to run this test code, i.e.
if tot_date != date -> if tot_date != date - delta

but apart of that I see some confusing behaviour: If I test the code with the fix branch I get this output (adding some printout):

time of transit date 2023-09-25 requested date 2023-09-23
dawn 2023-09-23 09:07:47.206584+00:00 dusk 2023-09-24 00:00:19.982675+00:00

while when I run with the master branch from the main repo I get this:

dawn 2023-09-23 09:07:47.206584+00:00 dusk 2023-09-23 00:00:19.982675+00:00

So there are 3 things that are puzzling me here

• the Time of Transit date is 2 days after the requested date
• the PR branch gives me the dusk for the next day (24th of September) that the one requested
• I am getting identical timestamp for the dusk from the PR branch and the main branch, however the main branch gives me the timestamp for 23 of September while the PR branch from 24. These 2 can't be the same.

[shanedoolane]

@Voutsi @jlopezarriaza let me take a look over the weekend for a fix