On 15/12/2021, at 11:51 AM, Tom Lane <t...@sss.pgh.pa.us> wrote:
> You should
> probably palloc temp arrays right here and then use construct_md_array
> directly instead of dealing with an ArrayBuildState.
OK, I gave that a go [1] this time in a vec_agg_sum() aggregate, that operates
much the same as the vec_agg_mean() one and is behaving in the same way. For
example this shows the correct values:
SELECT vec_agg_sum(nums) FROM measurements;
NOTICE: sum 0 = 4.92
NOTICE: sum 1 = 5.91
NOTICE: sum 2 = 14.80
vec_agg_sum
-------------------
{4.92,5.91,14.80}
(1 row)
But adding unnest() goes back to seemingly memory address values:
SELECT unnest(vec_agg_sum(nums)) FROM measurements;
NOTICE: sum 0 = 4.92
NOTICE: sum 1 = 5.91
NOTICE: sum 2 = 14.80
unnest
----------------
94069010663032
94069010663044
94069010663056
(3 rows)
Any other ideas I could look into?
— m@
[1]
https://github.com/SolarNetwork/aggs_for_vecs/blob/df8b950c73d9b32fa6f016b44dd8859e9341fe0f/vec_agg_sum.c#L50-L52
<https://github.com/SolarNetwork/aggs_for_vecs/blob/df8b950c73d9b32fa6f016b44dd8859e9341fe0f/vec_agg_sum.c#L50-L52>
