FR: as.hms.units()

[krlmlr]

Simply as.hms.units <- function(x, ...) as.hms(as.dt(x, ...)), but probably needs to import hms.

[edzer]

I now realize that hms is used for time of day, not duration. The way you suggest to implement it simply coerces a difftime object (time duration) into a hms, but that means something different. I will remove this until we've sorted it out. udunits does have support for absolute time; we could coerce that to POSIXct, and then drop day and retain a hms; the other way around is impossible without knowing the day.

[krlmlr]

hms wraps difftime, and so can be used for both: duration (as difference of two times) and time of day (like in Sys.time() - as.POSIXct(Sys.Date())).

[edzer]

I can see that, but that is not how the hms package explains what hms values mean. So: syntactically yes, semantically no, and measurement units are about meaning. Subclassing time-of-day from a time duration class is a conceptual mistake: is 12:00:00 twice as late as 06:00:00? hms thinks so:

> library(hms)
> hms(hours=6)
06:00:00
> hms(hours=6)*2
Time difference of 43200 secs

This should of course have triggered an error.

[krlmlr]

Thanks, I shall adapt the README then. The hms class doesn't specify preferred usage, it's just a container with nice formatting intended to be compatible to difftime.

See tidyverse/hms#18 for the conversion issue.

[edzer]

The better fix would be to make hms unambiguous (and not derive from difftime)

[edzer]

O.K., you clarified that the meaning of hms 09:00:00 can be one of

• a duration of 9 hours (since 00:00 midnight)
• 9 o'clock, as of "the shop opens at 9 o'clock in the morning", reflecting time-of-day

Since this cannot be be converted unambiguously in an SI unit of time duration (as package units understands them), I close this issue.