When exactly does data.table preserve column names?

Question

Some J expressions preserve column names, and some don't:

library(data.table)
d = data.table(hello = 1)
d[, .(hello)]
#    hello
# 1:     1
d[, c(.(hello))]
#    V1
# 1:  1
d[, {.(hello)}]
#    hello
# 1:     1
d[, {1; .(hello)}]
#    V1
# 1:  1
d[, .(get("hello"))]
#    V1
# 1:  1
d[, mget("hello")]
#    hello
# 1:     1
d[, c(mget("hello"))]
# Error: value for ‘hello’ not found  # WTF?
d[, {1; mget("hello")}]
#    hello
# 1:     1

What's the general rule?

Answer 1

This really isn't very mysterious. First note that . is a shorthand for list in data.table. With that in mind, this is what's happening.

Every single expression you have produces an unnamed list or vector, except for mget which spits out a named list. This is why mget examples have named results. In all of the other examples data.table only deduces the name in very simple expressions, which are equivalent to list(items). There are a few more cases where you'd get name deduction that would involve .SD, but generally speaking - if you didn't name it (as mget does) and it's more complicated than list(items), then don't expect name deduction.