date_time_parse_UTC()? A more flexible date_time_parse_RFC_3339()

[DavisVaughan]

date_time_parse_RFC_3339() is unique because it parses the RFC 3339 format, like:

2019-01-01T00:00:00Z
2019-01-01T00:00:00+0430
2019-01-01T00:00:00+04:30

Notably it allows you to parse and use the +04:30 info. But it's pretty strict about the format itself. You can only customize separator = "T" and offset = "Z".

This is an example where the abstraction over sys-time leaks:
https://stackoverflow.com/questions/79043367/how-to-read-in-a-character-datetime-with-a-timezone-offset-in-r/79044161#79044161

2019-01-01 00:00+04:30

Notably, no seconds! We have to dip down to sys_time_parse() to parse this

library(clock)

x <- c(
  "2023-10-29 00:00+02:00",
  "2023-10-29 01:00+02:00",
  "2023-10-29 02:00+02:00",
  "2023-10-29 02:00+01:00",
  "2023-10-29 03:00+01:00",
  "2023-10-29 04:00+01:00"
)

# Parse into (roughly) UTC, respecting `%Ez`, i.e. the `+HH:MM` bit
x <- sys_time_parse(
  x,
  format = "%Y-%m-%d %H:%M%Ez",
  precision = "minute"
)
x
#> <sys_time<minute>[6]>
#> [1] "2023-10-28T22:00" "2023-10-28T23:00" "2023-10-29T00:00" "2023-10-29T01:00"
#> [5] "2023-10-29T02:00" "2023-10-29T03:00"

# Convert to POSIXct with your expected time zone
as_date_time(x, zone = "Europe/Paris")
#> [1] "2023-10-29 00:00:00 CEST" "2023-10-29 01:00:00 CEST"
#> [3] "2023-10-29 02:00:00 CEST" "2023-10-29 02:00:00 CET" 
#> [5] "2023-10-29 03:00:00 CET"  "2023-10-29 04:00:00 CET"

# Or UTC if you wanted that
as_date_time(x, zone = "UTC")
#> [1] "2023-10-28 22:00:00 UTC" "2023-10-28 23:00:00 UTC"
#> [3] "2023-10-29 00:00:00 UTC" "2023-10-29 01:00:00 UTC"
#> [5] "2023-10-29 02:00:00 UTC" "2023-10-29 03:00:00 UTC"

Possibly we need date_time_parse_UTC() as one more convenience parser to fill this gap. I think this has come up one other time in the past.

---

We'd end up with:

	UTC offset YES	UTC offset NO
Full TZ name YES	date_time_parse_complete	date_time_parse_abbrev*
Full TZ name NO	date_time_parse_UTC	date_time_parse

With:

• date_time_parse_abbrev() being the UTC offset NO + Full TZ name YES combo, which isn't exactly accurate, but just the full tz name like America/New_York is not enough to disambiguate a time that sits in the "fall back" overlap, so we don't include that case and instead include the abbrev case, because an abbreviation combined with the supplied zone is enough to disambiguate
• date_time_parse_RFC_3339() being a special case of date_time_parse_UTC() that is restricted to just the common RFC format

But then it seems like we actually cover the whole spectrum of possible formats to parse!

This table would probably be pretty useful to include in the help docs of ?date_time_parse