PostgreSQL array_agg order for window functions - postgresql

Question

The answer to my question was almost here: PostgreSQL array_agg order
Except that I wanted to array_agg over a window function:
select distinct c.concept_name,
array_agg(c2.vocabulary_id||':'||c2.concept_name
order by c2.vocabulary_id, c2.concept_name)
over (partition by ca.min_levels_of_separation),
ca.min_levels_of_separation
from concept c
join concept_ancestor ca on c.concept_id = ca.descendant_concept_id
and max_levels_of_separation > 0
join concept c2 on ca.ancestor_concept_id = c2.concept_id
where
c.concept_code = '44054006'
order by min_levels_of_separation;
So, maybe this will work in some future version, but I get this error
ERROR: aggregate ORDER BY is not implemented for window functions
LINE 2: select distinct c.concept_name, array_agg(c2.vocabulary_id||...
^
I should probably be selecting from a subquery like the first answer to the quoted question above suggests. I was hoping for something as simple as the order by (in that question's second answer). Or maybe I'm just being lazy about the query and should be doing a group by instead of select distinct.
I did try putting the order by in the windowing function (over (partition by ca.min_levels_of_separation order by c2.vocabulary_id, c2.concept_name)), but I get these sort of repeated rows that way:
"Type 2 diabetes mellitus";"{"MedDRA:Diabetes mellitus"}";1
"Type 2 diabetes mellitus";"{"MedDRA:Diabetes mellitus","MedDRA:Diabetes mellitus (incl subtypes)"}";1
"Type 2 diabetes mellitus";"{"MedDRA:Diabetes mellitus","MedDRA:Diabetes mellitus (incl subtypes)","SNOMED:Diabetes mellitus"}";1
(btw: http://www.ohdsi.org/ if you happen to be curious about where I got the medical vocabulary tables)

Answer 1

Yes, it does look like I was being muddle-headed and didn't need the window function. This seems to work:
select c.concept_name,
array_agg(c2.vocabulary_id||':'||c2.concept_name
order by c2.vocabulary_id, c2.concept_name),
ca.min_levels_of_separation
from concept c
join concept_ancestor ca on c.concept_id = ca.descendant_concept_id
and max_levels_of_separation > 0
join concept c2 on ca.ancestor_concept_id = c2.concept_id
where c.concept_code = '44054006'
group by c.concept_name, ca.min_levels_of_separation
order by min_levels_of_separation
I won't accept my answer for a while since it just avoids the question instead of actually answering it, and someone might have something more useful to say on the matter.

Answer 2

like this :
select distinct c.concept_name,
array_agg(c2.vocabulary_id||':'||c2.concept_name ) over (partition by ca.min_levels_of_separation order by c2.vocabulary_id, c2.concept_name),
ca.min_levels_of_separation
from concept c
join concept_ancestor ca on c.concept_id = ca.descendant_concept_id
and max_levels_of_separation > 0
join concept c2 on ca.ancestor_concept_id = c2.concept_id
where
c.concept_code = '44054006'
order by min_levels_of_separation;