Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I certainly don't want to have "char" emulate the misbegotten
> decision to have explicit and implicit coercions behave differently.
> So it looks to me like the argument to make "char" work like char(1)
> doesn't actually help us much to decide if an error should be thrown
> here or not. On the whole, throwing an error seems better from a
> usability perspective.
I feel that the behavior of "char" in at least this case should match
char(1) (or just plain char):
test=# select case when true then 'xxx' else 'a'::"char" end from t;
case
------
x
(1 row)
test=# select case when true then 'xxx' else 'a'::char(1) end from t;
case
------
xxx
(1 row)
test=# select case when true then 'xxx' else 'a'::char end from t;
case
------
xxx
(1 row)
Much as the reason for the behavior of "char" may seem clear when
inside the code looking out, it is astonishing for someone writing
application code.
-Kevin