NA in subsetter - Inconsistent behavior

Question

Consider the following:

seq(from=10,to=30)[c(4,8)]
[1] 13 17

seq(from=10,to=30)[c(NA,8)]
[1] NA 17

seq(from=10,to=30)[c(NA,NA)]
[1] NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA

To me the result of the latter is inconsistent with how the other two examples are handled by R and has cost me quite unpleasant debugging pains.

Might this be considered a bug?

Claudication

A bug, no. None of the seq(10, 30) is NA. [ is an extraction function. The NA you are trying to extract are not present in the sequence. – Pesach

Try seq(from=10,to=30)[c(TRUE, FALSE)] and see what you get. – Pesach

Or try (10:30)[NA] to see what you get. R's rules about dealing with NA are quite explicit and are in fact consistent. – Glib

I don't get that. (10:30)[c(NA,8)] might be paraphrased into: "Return the 'not applicable' and 8th element", so why is "return the 'not applicable' element twice" not returning c(NA,NA)? – Claudication

Answer 1

Not sure if it is a bug - I suspect that depends on your point of view, but it is a subtlety of how indexing works. The quick solution is to change your third example to:

seq(from=10,to=30)[as.numeric(c(NA,NA))]

The reason is that c(NA,NA) is a logical vector, so the logical subsetting (which involves recycling the vector) is used, whereas having at least one non-NA causes the vector to be promoted to an integer vector. Likewise this could be implemented as:

seq(from=10,to=30)[c(NA_integer_,NA_integer_)]

See ?'[' for specifics of indexing if you're not familiar with them.

Blackdamp

That makes it very clear - I had not appreciated the type difference the subsetting vector is subject to and always thought about NA as a kind of neutral type. Thank you very much. – Claudication