prayertimes data
csv/ is every island's table, taken verbatim from the Salat MV export at
https://github.com/WovenCoast/prayertimes-db. Files are named by island id;
config/prayertimes/islands.conf maps ids to names and picks which are shown.
Home-manager links csv/ to ~/.local/share/prayertimes/.
Each row is day,month,fajr,sunrise,dhuhr,asr,maghrib,isha, times H:MM in
24-hour local time -- the generator folds each island's offset from Male' into
the values.
The day and month columns are a day out
scripts/prayertimes.sh addresses rows by line number, not by the day,month
they carry. Upstream's convert.js labels each row with
moment().dayOfYear(Date).year(2024), which sets the day number against the
current year and only then rewrites the year, so in a common year the
leap-year intent is lost: row 1 is labelled 31 December instead of 1 January,
29 February never appears, and 31 December appears twice.
Date in the upstream salat.db runs 0..365 and Date 0 is 1 January. The
sibling out/ files, prayertime.sh (which indexes them with date +%j - 1)
and the four test vectors in the db's own readme all agree. So, rows numbered
from 1:
row = day_of_year in a leap year
row = day_of_year + 1 in a common year, from 1 March on
The + 1 skips the 29 February row every table carries. That reproduces all
four upstream test vectors for Male' (102) -- 1 January Dhuhr 12:13 and Isha
19:25, 2 March Dhuhr 12:22 and Asr 15:35 -- in leap and common years alike,
which the day,month labels do not. Every file must therefore stay exactly
366 lines, in order, with no header.
Hijri dates
The tooltip's Hijri date comes from a 720 byte Umm al-Qura month-length table embedded in the script, covering 1440-1499 AH (2018-09-11 to 2076-11-26). Arithmetic Hijri calendars are a day off Umm al-Qura roughly half the time, which the table avoids. Umm al-Qura is itself a calculated calendar, so it can still differ from a locally announced sighting.