Respect `nomatch = 0L` for integer i

[mllg]

Currently nomatch seems to be ignored if i in dt[i, j, nomatch = 0L] is integer. Values of i which are out of bounds or NA result in rows filled with NA:

dt = data.table(x = 1:10)
dt[c(1L, 11L, NA_integer_), x, nomatch = 0L]
> 1 NA NA

I see no reason to not support nomatch = 0L for integer i.

Also note that the current handling of negative integer i seems to be inconsistent with not-joins:

# mixing negation/not-join with nomatch
dt[-1L, nomatch = 0L] # works, nomatch ignored
dt[!list(1L), on = "x", nomatch = 0L] # exception

# negation/not-join with out-of-bounds values
dt[-11L] # warning
dt[!list(11), on = "x"] # no warning

[franknarf1]

For the second part, the inconsistent behavior is good, right?

I generally don't want a warning when part of an anti-join matches nothing, but do want it when I pass a row number outside of range. There are other safeguards on row numbers that don't have join analogies too (eg, dt[c(-1,1)] errors).

[mllg]

I generally don't want a warning when part of an anti-join matches nothing, but do want it when I pass a row number outside of range.

Well, I'm not sure about it. Base R also does not give a warning:

x = 1:10
x[-11L] # no warning, works fine
x[c(-1L,1L)] # also an exception

[jangorecki]

please use nomatch=NULL from now on