why does split coerce double to integer in R and is there a workaround

Question

One can build dates in R from numeric types, even fractional values. It's not very common, but it happens for instance when averaging dates. Unfortunately, they seem to break split

> as.Date(-1, origin = "1970-01-01")
[1] "1969-12-31"
> as.Date(-1.0001, origin = "1970-01-01")
[1] "1969-12-30"
> split(as.Date(-1, origin = "1970-01-01"), 1)[[1]]
[1] "1969-12-31"
> split(as.Date(-1.0001, origin = "1970-01-01"), 1)[[1]]
[1] "1969-12-31" #this is wrong
> unclass(split(as.Date(-1, origin = "1970-01-01"), 1)[[1]])
[1] -1
> unclass(split(as.Date(-1.0001, origin = "1970-01-01"), 1)[[1]])
[1] -1 #this is "why"

So two dates that were different are made equal by a split. Do people agree that this is a bug or am I missing the deep reason? Any workarounds? Thanks

Answer 1

For whatever reason, split.Date coerces the Date input to integer:

> split.Date
function (x, f, drop = FALSE, ...) 
{
    y <- split.default(as.integer(x), f, drop = drop)
    for (i in seq_along(y)) class(y[[i]]) <- "Date"
    y
}
<bytecode: 0x2effb98>
<environment: namespace:base>

This is at minimum an infelicity between the function and the documentation, since ?Date says, "the date should be an integer, but this is not enforced in the internal representation.". Some might consider it a bug. I'm not sure.

You can avoid that by calling split.default directly.

> split.default(as.Date(-1.0001, origin = "1970-01-01"), 1)[[1]]
[1] "1969-12-30"