Assignment is inconsistent with list

[ColeMiller1]

Updating a data.table by reference provides inconsistent results for single-row data.tables:

library(data.table) ##1.12.8 - same results in master
dt = data.table(a = 3L)
dt[, b:= 3L]
dt[, b2 := .(3L)]
dt[, `:=`( b3 = 3L)]
dt[, `:=`(b4 = .(3L))]
dt
#>        a     b    b2    b3     b4
#>    <int> <int> <int> <int> <list>
#> 1:     3     3     3     3      3

Basically, the var := .(...) version seems to coerce the RHS to an atomic if ... evaluates to an atomic and is the same number of rows as the data.table. This differs from `:=`(var = .(...)) which appears to always make var a list column

dt = data.table(a = c(1L,3L))
dt[, b2 := .(c(2L, 3L))]
dt[, `:=`(b4 = .(c(2L, 3L)))]
dt

##       a    b2     b4
##   <int> <int> <list>
##1:     1     2    2,3
##2:     3     3    2,3

##non-equal lengths
dt = data.table(a = c(1L, 3L, 5))
dt[, b2 := .(c(2L, 3L))]
dt

##       a     b2
##   <num> <list>
##1:     1    2,3
##2:     3    2,3
##3:     5    2,3

I have not had a chance to dig deep, but this line from [.data.table may help explain why `:=`(var = .(...)) results in a list.

data.table/R/data.table.R

Lines 1047 to 1053 in ad7b67c

	} else {
	# `:=`(c2=1L,c3=2L,...)
	lhs = names(jsub)[-1L]
	if (any(lhs=="")) stop("In `:=`(col1=val1, col2=val2, ...) form, all arguments must be named.")
	names(jsub)=""
	jsub[[1L]]=as.name("list")
	}

This was observed while investigating #4568 although I am not sure how related they are.

[jangorecki]

Related:
point 4 in https://github.com/Rdatatable/data.table/blob/master/NEWS.md#datatable-v1124--03-oct-2019
#3626
#3925
my comment on that issue: #3626 (comment)

[ColeMiller1]

Thanks for the related items.

I see LHS := RHS similar to transform(), within(), or $<-. In each case, the RHS is taken as is. Therefore, I expect V1 := 1L to be an atomic whereas V1 := list(1L) to be a list column.

[mb706]

I assume this behaviour comes from b2 := list(3L) being an edge-case of c("b2", "c2", "d2") := list(3L, 4L, 5L)?

If n is set to an integer, then

dt <- data.table()
lettercols <- head(letters, n)
lettervals <- as.list(seq_len(n))
dt[, (lettercols) := lettervals]

should not give a list column in the n == 1 case. The `:=`(col = val) notation does not work like this, therefore there is no need to remove the outermost list().