fcoalesce doesnt play nice w/ date and idate

[ethanbsmith]

data.table loads dates as IDate. not an issue, as these can be simply converted to native Date with :=(as.Date())
problem is that the mode on these converted dates is still integer and this causes fcoalesce to puke

very annoying to have to micromanage this all over the place when mixing data sourced both from and not from fread
sort of feels like a self-own. Any way this could be made less cumbersome? maybe fcase special casing Date subclass comparisons?

data.table::fcoalesce(Sys.Date(), as.IDate("9999-12-31")) 
#Error in data.table::fcoalesce(Sys.Date(), as.IDate("9999-12-31")) :    Item 2 is type integer but the first item is type double. Please coerce before coalescing.  
data.table::fcoalesce(Sys.Date(), as.Date(as.IDate("9999-12-31"))) 
#Error in data.table::fcoalesce(Sys.Date(), as.Date(as.IDate("9999-12-31"))) :    Item 2 is type integer but the first item is type double. Please coerce before coalescing.  
data.table::fcoalesce(Sys.Date(), as.Date(as.numeric(as.IDate("9999-12-31")))) 
#[1] "2025-12-29"

[ben-schwen]

Ty for the report. Note that Date can also come with both flavors of storage mode integer and double.

We have patched a similar issue in #6602. Need to take a careful look on all of these "corner" cases

[ethanbsmith]

also, u can't go through integer. u have to go through IDate

fcoalesce(as.Date(as.integer(Sys.Date())), as.Date(as.IDate("9999-12-31")))
#Error in fcoalesce(as.Date(as.integer(Sys.Date())), as.Date(as.IDate("9999-12-31"))) : 
  Item 2 is type integer but the first item is type double. Please coerce before coalescing.

not sure from the linked issue if this is a windows only thing. my setup:

R version 4.5.2 (2025-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26100)

data.table_1.18.0

[aksh08022006]

This seems to stem from a storage-mode mismatch rather than a class mismatch: Date objects can have storage mode double or integer, while IDate is always integer. Even after coercing via as.Date(as.IDate(...)), the underlying mode can remain integer, which causes fcoalesce() to fail its strict type checks.

One possible approach would be to special-case Date-like inputs (Date / IDate) in fcoalesce(): if all arguments inherit from either class, normalize them to a common internal representation (e.g. IDate) before the type check, then restore the output class based on the inputs. This would avoid users needing to manually micromanage storage modes while keeping behavior consistent with base R.

If this approach sounds reasonable, I’d be happy to open a PR with a minimal fix and accompanying tests.

[MichaelChirico]

@aksh08022006 that sounds correct to me, a PR would be most welcome!

I think I'd be fine with allowing integer->double coercion in fcoalesce generally, no need to make this a Date class thing specifically.

[ethanbsmith]

I think I'd be fine with allowing integer->double coercion in fcoalesce generally, no need to make this a Date class thing specifically.

this approach feels very R. just want to make sure u dont end up doing something like returning an IDate w/ mode double

[MichaelChirico]

I guess something to consider is if the vector has double storage but equivalent representation as integer, to do double->integer coercion instead. More expensive, though.

Safer is to just implement coalesce(<Date>, <IDate>) always gives Date output.

[ethanbsmith]

my mental model of fcoalesce is that it returns the type,class,mode,attributes of the first input.
not sure this is correct. if its is, maybe make the contract explicit

if not, return value should at least be consistent. i.e. type,class,mode,attributes from only one column, not mix/matching these
in this approach where the int -> double conversion rule takes precedence (which is quite reasonable), then just use the type,class,mode,attributes from the first double column