On Wed, Apr 19, 2017 at 1:03 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
david.g.johnston@gmail.com writes: > This doesn't, and should since the number of elements in the non-empty array > shouldn't change the dimensionality logic.
> SELECT array_agg(CASE WHEN a = ARRAY[]::text[] THEN ARRAY['N/A']::text[] > ELSE a END) > FROM ( VALUES (1, ARRAY[]::text[]), (1, ARRAY['1','2']::text[]) ) vals (v, > a)
Why do you think that should work? You're asking array_agg to accumulate a 1-D length-1 array and then a 1-D length-2 array. There's no way to make a rectangular 2-D array out of that, except perhaps by inventing entries which isn't in array_agg's charter.
I was being too narrow-minded in interpreting the word dimension.
SELECT array_ndims(ARRAY['1','2']::text[]); => 1; therefore it is a one dimensional array - having a length of 2. One can, and we do in other places, define that array as having a dimension of 1x2.
SELECT '{{N/A},{1,2}}'::text[] --> fails with "sub-arrays with matching dimensions"...
Apparently the annoyance I posted on the other thread isn't so simple to resolve - just having an empty "1-dimensional" array is not particularly useful given the length dimension must also match.
This helps explain why my first attempt:
SELECT array_dims('{1,2}'::text[]); --> yields [1:2] which is what is being keyed off here.
All this and all I really want is a friggin' "array_append / array_concat" aggregate function that accepts either scalars or matching "primary dimension" arrays - and treats empty arrays as no-ops.
