Re: Surprising results from array concatenation

On Tue, Apr 25, 2017 at 9:26 AM, Mike Blackwell <mike(dot)blackwell(at)rrd(dot)com>
wrote:

> The docs (section 9.18 for PG 9.6) show as an example for array
> concatenation
>
> ARRAY[4,5,6] || 7
>
> which works fine. However, trying the same with an array of text doesn't
> work:
>
> # select array['a','b','c'] || 'd';
> ERROR: malformed array literal: "d"
> LINE 1: select array['a','b','c'] || 'd';
> ^
> DETAIL: Array value must start with "{" or dimension information.
>
> Casting the second value to TEXT works.
>
> # select array['a','b','c'] || 'd'::TEXT;
> ?column?
> -----------
> {a,b,c,d}
> (1 row)
>
> The assumption that the second argument is an array constant seems
> surprising
>

​It has to assume something. And for better and worse it has to assume it
without looking at the actual value.​ Choosing the scalar variant here
would be more convenient but choosing the same type as the left-hand side
is logical. Note that the concatenation operator/function isn't the one
complaining - if it was then intelligence could be inserted. The type
conversion code doesn't have the luxury.

I don't suppose one would get far arguing to modify the array input
function to convert a value that doesn't look like an array into a single
element text array. The "implicit conversion" that involves is something
we've gotten away from and seems like it would be worse that requiring the
explicit typing.

David J.