CTE jn
    ->  Nested Loop  (cost=27.02..692876.20 rows=479997 width=89) (actual time=48.524..664771.377 rows=119324 loops=1)
          Buffers: shared hit=5984023 read=152
          ->  CTE Scan on prjdat  (cost=0.00..17.90 rows=895 width=72) (actual time=1.033..1127.682 rows=1150 loops=1)
                Buffers: shared hit=12462
          ->  Bitmap Heap Scan on fok_keywords k1  (cost=27.02..768.78 rows=536 width=49) (actual time=57.808..577.039 rows=104 loops=1150)
                Recheck Cond: ((lexize_trim_to_uniq_array(keyword) <@ prjdat.av_tokens) AND (cardinality(fok_codes_wghts) <= 37))
                Rows Removed by Index Recheck: 13424
                Heap Blocks: exact=5415056
                Buffers: shared hit=5971561 read=152
                ->  Bitmap Index Scan on fok_keywords_lexize_trim_to_uniq_array_cardinality_keyword_idx  (cost=0.00..26.89 rows=536 width=0) (actual time=26.935..26.935 rows=13528 loops=1150)
                      Index Cond: ((lexize_trim_to_uniq_array(keyword) <@ prjdat.av_tokens) AND (cardinality(fok_codes_wghts) <= 37))
                      Buffers: shared hit=556505 read=152
  ->  Nested Loop  (cost=7.50..20407.37 rows=479997 width=48) (actual time=1756.383..1345239.164 rows=11532 loops=1)
        Buffers: shared hit=6310560 read=152, temp written=8701
        ->  CTE Scan on jn  (cost=0.00..9599.94 rows=479997 width=72) (actual time=48.530..664974.568 rows=119324 loops=1)
              Buffers: shared hit=5984023 read=152, temp written=8701
        ->  Function Scan on find_keyword_hunsp fk  (cost=7.50..7.51 rows=1 width=8) (actual time=5.700..5.700 rows=0 loops=119324)
              Filter: (fk > '0'::double precision)
              Rows Removed by Filter: 1
              Buffers: shared hit=326537
Planning time: 2.666 ms
Execution time: 1345276.880 ms

  CTE jn
    ->  Nested Loop  (cost=11.02..84169.39 rows=479997 width=89) (actual time=49.888..131849.918 rows=119324 loops=1)
          Buffers: shared hit=8202201 read=8835 written=1144
          ->  CTE Scan on prjdat  (cost=0.00..17.90 rows=895 width=72) (actual time=18.886..5731.687 rows=1150 loops=1)
                Buffers: shared hit=11864 read=598 written=1
          ->  Bitmap Heap Scan on fok_keywords k1  (cost=11.02..88.66 rows=536 width=124) (actual time=66.897..109.631 rows=104 loops=1150)
                Recheck Cond: (kw_tokens <@ prjdat.av_tokens)
                Rows Removed by Index Recheck: 28715
                Filter: (cardinality(fok_codes_wghts) <= 37)
                Rows Removed by Filter: 108
                Heap Blocks: exact=7677353
                Buffers: shared hit=8190337 read=8237 written=1143
                ->  Bitmap Index Scan on fok_keywords_kw_tokens_idx  (cost=0.00..10.89 rows=872 width=0) (actual time=38.387..38.387 rows=28927 loops=1150)
                      Index Cond: (kw_tokens <@ prjdat.av_tokens)
                      Buffers: shared hit=454446 read=13 written=2
  ->  Nested Loop  (cost=7.50..20407.37 rows=479997 width=48) (actual time=620.177..809146.403 rows=11532 loops=1)
        Buffers: shared hit=8528738 read=8835 written=1144, temp written=8701
        ->  CTE Scan on jn  (cost=0.00..9599.94 rows=479997 width=72) (actual time=49.895..132060.603 rows=119324 loops=1)
              Buffers: shared hit=8202201 read=8835 written=1144, temp written=8701
        ->  Function Scan on find_keyword_hunsp fk  (cost=7.50..7.51 rows=1 width=8) (actual time=5.673..5.673 rows=0 loops=119324)
              Filter: (fk > '0'::double precision)
              Rows Removed by Filter: 1
              Buffers: shared hit=326537
Planning time: 2.610 ms
Execution time: 809179.745 ms

  CTE jn
    ->  Nested Loop  (cost=15.37..105543.42 rows=483331 width=89) (actual time=65.591..308701.015 rows=20728 loops=1)
          Buffers: shared hit=10207597 read=14797 written=1972
          ->  CTE Scan on prjdat  (cost=0.00..17.90 rows=895 width=72) (actual time=2.001..1858.651 rows=1150 loops=1)
                Buffers: shared hit=11874 read=600 written=3
          ->  Bitmap Heap Scan on fok_keywords k1  (cost=15.37..112.51 rows=540 width=187) (actual time=178.421..266.807 rows=18 loops=1150)
                Recheck Cond: (kw_tokens <@ prjdat.av_tokens)
                Rows Removed by Index Recheck: 29917
                Filter: (cardinality(fok_codes_wghts) <= 37)
                Rows Removed by Filter: 56
                Heap Blocks: exact=9052124
                Buffers: shared hit=10195723 read=14197 written=1969
                ->  Bitmap Index Scan on fok_keywords_kw_tokens_idx  (cost=0.00..15.24 rows=872 width=0) (actual time=107.563..107.563 rows=29992 loops=1150)
                      Index Cond: (kw_tokens <@ prjdat.av_tokens)
                      Buffers: shared hit=1096942 read=15 written=4
  ->  Nested Loop  (cost=7.50..20549.07 rows=483331 width=48) (actual time=508.131..403618.232 rows=12185 loops=1)
        Buffers: shared hit=10253288 read=14797 written=1972
        ->  CTE Scan on jn  (cost=0.00..9666.62 rows=483331 width=72) (actual time=65.595..308725.633 rows=20728 loops=1)
              Buffers: shared hit=10207597 read=14797 written=1972
        ->  Function Scan on find_keyword_hunsp fk  (cost=7.50..7.51 rows=1 width=8) (actual time=4.577..4.577 rows=1 loops=20728)
              Filter: (fk > '0'::double precision)
              Rows Removed by Filter: 0
              Buffers: shared hit=45691
Planning time: 3.871 ms
Execution time: 403644.954 ms

--DROP FUNCTION lexize_uniq_array2 (in txt text);
CREATE OR REPLACE FUNCTION lexize_uniq_array2 (in txt text)
returns text[]
AS 
$x$
--var without 12
SELECT ARRAY_AGG(trim(u) ORDER BY u)FILTER(WHERE trim(u)<>'') FROM (

select string_to_array(
	regexp_replace(string_agg(lexeme,'' order by o),'[\.,\|;:] ',' ','ig') --cut potential separators
	,' ') arr
FROM (
SELECT 
	o::int,tokid,token
	,case tokid when 12 then ' ' else (ts_lexize('russian_all_stem',token))[1] -- стеммер, т.к. нужно попасть в любые словоформы => симпл не подойдет
		-- если ханспелл -- придется собирать все варианты дублей. пока отложим
	end as lexeme
FROM ts_parse('default',regexp_replace(txt,'[\s'||chr(160)||']+',' ','ig')) with ordinality ot (tokid,token,o)
where tokid not in (15,5) -- out twice -- as own and as parts
order by o) tokenz
) foo 
,lateral (with u AS (SELECT * FROM unnest(arr) with ordinality AS u(u,o))
			SELECT concat (v.u,' ',w.u) FROM u as v join u as w ON v.o+1 = w.o
					UNION
					SELECT u FROM u
			
		) u(u)
;
$x$
immutable strict
language sql
COST 1000
;