The Oracle equivalent of that would be "SELECT count(*) FROM dual". Does it make more sense to you thought of that way?
For a user, Oracle's case makes perfect sense, since the command is querying a single-row table. In Postgres' case, there's nothing being queried, so the result's got to be either 0 or NULL.
> I agree it's an odd thing for someone to query, but I feel it should return > 0, and not 1.
For that to return zero, it would also be necessary for "SELECT 2+2" to return zero rows. Which would be consistent with some views of the universe, but not particularly useful. Another counterexample is
regression=# select sum(42); sum ----- 42 (1 row)
which by your argument would need to return NULL, since that would be SUM's result over zero rows.
Hmm.. Now that you put it that way, I agree it's a useful feature, or shall I say, a quirk with useful side effect.