[Rd] Should `as.difftime()` convert integer input to double?

Bill Dunlap w||||@mwdun|@p @end|ng |rom gm@||@com
Tue Aug 9 21:10:53 CEST 2022


Forcing difftime's payload to be numeric would avoid anomalies like

> as.difftime(123456789L, units="secs") * 100L
Time difference of NA secs
Warning message:
In e2 * unclass(e1) : NAs produced by integer overflow
> as.difftime(123456789, units="secs") * 100L
Time difference of 12345678900 secs

Note that there are packages (e.g., fst, a serialization package) which can
(in C++ code) understand and create difftimes with integer payloads, so you
may have trouble completely getting rid of such things.

-Bill

On Tue, Aug 9, 2022 at 11:26 AM Davis Vaughan <davis using rstudio.com> wrote:

> Hi all,
>
> Currently, `as.difftime()` with an integer input will produce a difftime
> object that internally is built on an integer vector, i.e.:
>
> x <- as.difftime(1L, units = "secs")
> typeof(x)
> #> [1] "integer"
>
> x <- as.difftime(1, units = "secs")
> typeof(x)
> #> [1] "double"
>
> I feel like difftime objects should always be built on *double* vectors.
> There are a few reasons I feel like this should be true:
>
> - There is an `as.double.difftime()` method, but no `as.integer.difftime()`
> method, which implies something about what the underlying storage type is
> assumed to be.
>
> - AFAIK, there is no other way to produce a difftime object with integer
> storage using the exposed API (aside from abusing the internal .difftime()
> helper). Even `.Date(1L) - .Date(1L)` produces a difftime with double
> storage.
>
> - `seq.Date()` used to be able to produce dates with integer storage, but
> as of recently even that edge case has been altered to always produce dates
> with double storage. So a change to also force difftime to have double
> storage would feel consistent with that.
>
> https://github.com/wch/r-source/commit/0762ee983813c4df9b93f6b5ee52c910dcd3ab39
>
> It looks like a patch to `as.difftime()` would be fairly straightforward
> (i.e. ensuring that integer input is coerced to double), so I'd be happy to
> attempt one if someone else agrees that this should be changed.
>
> Thanks,
> Davis
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

	[[alternative HTML version deleted]]



More information about the R-devel mailing list