dcast outputs column name V1 for empty string

[tdhock]

Hi! I expected that dcast(DT, .~column) should always output a data table that has columns named unique(paste(DT$column)) but here is a counter-example:

> x=c("NA","",NA,"foo");DT=data.table(x);out=dcast(DT,.~x,length)
> x
[1] "NA"  ""    NA    "foo"
> names(out)
[1] "."   "NA"  "V1"  "NA"  "foo"
> out
   . NA V1 NA foo
1: .  1  1  1   1

The issue above is that the empty string value is mapped to the column name V1, which is confusing. I would propose to fix by changing that output column name to empty string, as below:

> setnames(out,c(".","NA","","NA","foo"))
> out
   . NA   NA foo
1: .  1 1  1   1

Another thing I noticed is that both NA (missing value) and "NA" (string) map to two different columns (with the same name, "NA"), is that normal? (should there be a warning if dcast outputs columns with the same name?)

[ben-schwen]

I guess the initial intention was to avoid duplicate names similar to make.names(names, unique=TRUE), hence, I find it quite surprising that you can sneak in the double "NA".

[Divendra2006]

Hi @tdhock,

I came across this issue, which was opened in March 2023, and I’m interested in contributing. However, I wanted to check if this issue is still relevant and needs to be addressed.

If it’s still valid, I’d be happy to work on it. Please let me know.

[tdhock]

yes still relevant, please go ahead.

[tdhock]

from #6795 (comment) I don't think behavior should be changed after all.
maybe we could update the docs to clarify?