-- CREATE INDEX splits what a segment cannot hold: the postings of a term and the dictionary of a segment
-- have a 4 GB limit (a uint32 length and offset), so the build writes more segments instead of failing.
-- sift.test_blob_max and sift.test_dict_max lower the limits to bytes a test can reach; the runs of the
-- build merge only while their terms and dictionaries fit them together.
CREATE EXTENSION pg_sift;
SET max_parallel_maintenance_workers = 0;

CREATE TABLE bl (id serial, body text);
INSERT INTO bl (body)
SELECT 'all g' || (i % 100) || ' u' || i || ' v' || (i * 7919 % 100003) || ' w' || (i % 50)
FROM generate_series(1, 40000) i;

CREATE FUNCTION answers(OUT n bigint, OUT ids text) AS $$
	SELECT (SELECT count(*) FROM bl WHERE body @@> 'all'),
	       (SELECT md5(string_agg(id::text, ',' ORDER BY id)) FROM bl WHERE body @@> 'g5 OR u777 OR w3 AND g8')
$$ LANGUAGE sql;
CREATE FUNCTION ranked() RETURNS text AS $$
	SELECT string_agg(id::text || ':' || round(s::numeric, 5), ' ' ORDER BY s DESC, id) FROM
	(SELECT id, sift.score(ctid) AS s FROM bl WHERE body @@> 'g5 OR u777 OR w3 AND g8'
	 ORDER BY s DESC, id LIMIT 30) t
$$ LANGUAGE sql;
CREATE INDEX bl_ref ON bl USING sift (body);
CREATE TABLE ref AS SELECT a.*, ranked() AS top FROM answers() a;
DROP INDEX bl_ref;
CREATE FUNCTION same() RETURNS boolean AS $$
	SELECT a.n = r.n AND a.ids = r.ids AND ranked() = r.top FROM answers() a, ref r
$$ LANGUAGE sql;

-- the postings of 'all' take more than 40,000 bytes in one segment
SET sift.test_run_docs = 1000;
SET sift.test_merge_fanin = 8;
SET sift.test_blob_max = 40000;
CREATE INDEX bl_blob ON bl USING sift (body);
RESET sift.test_blob_max;
SELECT count(*) BETWEEN 4 AND 128 AS blob_segments FROM sift.segments('bl_blob');
SELECT sift.verify('bl_blob', true);
SELECT same() AS blob_same;
DROP INDEX bl_blob;

-- the dictionary of the table takes more than 150,000 bytes in one segment
SET sift.test_dict_max = 150000;
CREATE INDEX bl_dict ON bl USING sift (body);
RESET sift.test_dict_max;
SELECT count(*) BETWEEN 4 AND 128 AS dict_segments FROM sift.segments('bl_dict');
SELECT sift.verify('bl_dict', true);
SELECT same() AS dict_same;
DROP INDEX bl_dict;

-- both, in parallel and with forced boundaries
SET sift.test_blob_max = 40000;
SET sift.test_dict_max = 150000;
SET max_parallel_maintenance_workers = 2;
ALTER TABLE bl SET (parallel_workers = 2);
CREATE INDEX bl_par ON bl USING sift (body);
SELECT count(*) BETWEEN 4 AND 128 AS parallel_segments FROM sift.segments('bl_par');
SELECT sift.verify('bl_par', true);
SELECT same() AS parallel_same;
DROP INDEX bl_par;
CREATE INDEX bl_bound ON bl USING sift (body) WITH (build_segments = 3);
SELECT count(*) BETWEEN 4 AND 128 AS bound_segments FROM sift.segments('bl_bound');
SELECT sift.verify('bl_bound', true);
SELECT same() AS bound_same;

DROP TABLE bl, ref;
DROP FUNCTION same(), ranked(), answers();
DROP EXTENSION pg_sift;