Subsetting with Date produces error

[renkun-ken]

It is known that using a numeric vector in i will be treated as subsetting by index. However, if the supplied numeric vector has some classes (e.g. Date), then it might make more sense not to subset by index but join (just like a character vector).

Consider the following example:

library(data.table)
test <- data.table(
  date = seq.Date(as.Date("2020-01-01"), as.Date("2020-07-01"), by = "day"),
  key = "date"
)
test[, x := rnorm(.N)]
test[as.Date("2020-01-02")]

The following error is produced:

> test[as.Date("2020-01-02")]                                                                   
Error in `[.data.table`(test, as.Date("2020-01-02")) : 
  i has evaluated to type double. Expecting logical, integer or double.
Backtrace:
1: stop("i has evaluated to type ", typeof(i), ". Expecting logical, integer or double.")
2: `[.data.table`(test, as.Date("2020-01-02"))
3: test[as.Date("2020-01-02")]

User might expect it to behave like

> test[.(as.Date("2020-01-02"))]                                                                
Key: <date>
         date          x
       <Date>      <num>
1: 2020-01-02 -0.2380267

[MichaelChirico]

Hmm I think I would prefer to just implement #3506

[ColeMiller1]

This is slightly different. This is suggesting using NSE to transform a bare date into a list date, similar to what happens with bare characters.

data.table/R/data.table.R

Lines 397 to 405 in db61844

	if (is.character(i)) {
	isnull_inames = TRUE
	i = data.table(V1=i) # for user convenience; e.g. DT["foo"] without needing DT[.("foo")]
	} else if (identical(class(i),"list") && length(i)==1L && is.data.frame(i[[1L]])) { i = as.data.table(i[[1L]]) }
	else if (identical(class(i),"data.frame")) { i = as.data.table(i) } # TO DO: avoid these as.data.table() and use a flag instead
	else if (identical(class(i),"list")) {
	isnull_inames = is.null(names(i))
	i = as.data.table(i)
	}

It would be easy to accommodate if this is something that should be supported. It would be good to have #4585 merged first as there would be some conflicts otherwise although I realize the other PR might be too much change.