Converting 'day' value to a timestamp

Question

Org agendas use 'day' values in various APIs. I'd like to continue using the wonderful ts.el package to keep all my own automation using a timestamp implementation that clicks in my own head. (Elisp's various manifestations of 'time', and an apparent inconsistency in documentation around these manifestations, is a constant source of confusion for me.)

I can't seem to find a way to use the value I get from (date-to-day "2023-08-14") into a useful time value. Naively using days-to-time on this value yields a date in 3992. In fact, both of the following forms evaluate to that same date:

(format-time-string "%FT%T%z" (days-to-time (date-to-day "2023-08-14")))
(format-time-string "%FT%T%z" (days-to-time (time-to-days (current-time))))

even though one might initially expect days-to-time and time-to-days to be inverses.

What's going on here? How can I get a ts value out of the value returned by (date-to-day "2023-08-14")?

Answer 1

There is date-to-time:

(make-ts :unix (string-to-number
                (format-time-string "%s" (date-to-time "2023-08-13"))))

This will assume that 2023-08-13 denotes midnight in your local timezone.

---

(let ((day (date-to-day "2023-08-13"))
      (epoch (date-to-day "1970-01-01")))
  (make-ts :unix (* (- day epoch) 86400)))

This will assume that 2023-08-13 denotes midnight in UTC.

Answer 2

I think @d125q has the right solution for you, but let me explain what's happening. Firstly date-to-day doesn't like your input. Via it's docstring (C-h f date-to-day)

Return the absolute date of DATE, a date-time string. The absolute date is the number of days elapsed since the imaginary Gregorian date Sunday, December 31, 1 BC.

;; *scratch* buffer
(date-to-day "2023-08-14")
730119
(date-to-day "3023-08-14")
730119

Oops. Something is wrong, we should have a 365250 day difference there. So what's 730119 days to years? 2000. What's 2000 years since December 31, 1 BC? December 31, 1999.

Now breaking down the call stack for date-to-day we have something like

;; date-to-day stack
(time-to-days ((encode-time (parse-time-string
               (timezone-make-date-arpa-standard date))))

So, using our innermost function

;; try our given input
(timezone-make-date-arpa-standard "2023-08-14")
"31 Dec 1999 19:00:00 -0500"
;; Oops

;; be more precise with our input
(timezone-make-date-arpa-standard "2023-08-14T00:00:00")
"13 Aug 2023 20:00:00 -0400"
(date-to-day "3023-08-14T00:00:00")
738746
(format-time-string "%FT%T%z" (days-to-time (date-to-day "2023-08-14")))
"3968-12-30T19:00:00-0500"
;; correct date based on our input to format-time-string

We've corrected our output, but it's still unexpectedly too far in to the future. This leads us to the second problem, our mis-application of format-time-string. From its docstring (C-h f format-time-string)

This function treats seconds as time since the epoch of 1970-01-01 00:00:00 UTC

Oops. We've calculated seconds (as in integer list) from December 31, 1 BC (date-to-day) and fed it to a function that does its calculations beginning on January 01, 1970. Obviously, we're going to be a just few (thousand) years off.

So, I see a couple of options,

• use @d125q's excellent and concise solution (or something based on it) (preferred)
• rewrite your function using functions that only operate on the UNIX epoch (meh)
• keep doing what you're doing but account for the difference in time (ugh)