-- The docs of a write buffer are matched and scored from their records (bm25/wbind.c): the same docs in a promoted segment must give the same ids and the same
-- scores, for a plain term, AND, OR, NOT, a phrase, a prefix, a wildcard and a fuzzy term, with docs that repeat a term; with deleted docs the ids are the same (a buffer
-- counts the docs it holds deleted in its statistics, so the scores differ)
CREATE EXTENSION pg_sift;
CREATE TABLE docs_buf (id int PRIMARY KEY, body text);
INSERT INTO docs_buf
SELECT g, 'alpha' || g % 7 || ' beta' || g % 11 || ' gamma ' || repeat('delta ', g % 4) || CASE WHEN g % 5 = 0 THEN 'epsilon zeta ' ELSE '' END ||
	'tail' || g % 13 || ' common' || CASE WHEN g % 3 = 0 THEN ' alphabet' ELSE '' END
FROM generate_series(1, 3000) g;
CREATE TABLE docs_seg AS SELECT * FROM docs_buf;
ALTER TABLE docs_seg ADD PRIMARY KEY (id);
CREATE INDEX docs_buf_idx ON docs_buf USING sift (body);
CREATE INDEX docs_seg_idx ON docs_seg USING sift (body);
-- the index of docs_buf keeps its rows in the buffer: they were added after the (empty) build
TRUNCATE docs_buf;
INSERT INTO docs_buf SELECT * FROM docs_seg;
SELECT sift.promote('docs_seg_idx') >= 0 AS promoted;
 promoted 
----------
 t
(1 row)

SELECT count(*) AS buffers FROM sift.segments('docs_buf_idx') WHERE kind <> 'immutable';
 buffers 
---------
       1
(1 row)

SELECT count(*) AS segments FROM sift.segments('docs_seg_idx') WHERE kind <> 'immutable';
 segments 
----------
        0
(1 row)

CREATE TABLE q (n serial, q text);
INSERT INTO q (q) VALUES ('alpha1'), ('gamma'), ('delta'), ('alpha1 beta3'), ('alpha1 OR beta3'), ('alpha1 OR beta3 OR tail5'), ('gamma AND NOT delta'), ('alpha2 AND (beta1 OR beta2)'),
	('epsilon zeta'), ('"epsilon zeta"'), ('alph*'), ('alpha*'), ('*pha1'), ('t?il5'), ('alpha1~1'), ('alphabet'), ('common AND NOT alphabet'), ('nothing'),
	('alpha3 OR nothing'), ('alph* beta4'), ('epsilon OR delta OR alpha6 OR tail0');
CREATE FUNCTION answer(tbl text, qq text) RETURNS text LANGUAGE plpgsql AS $$
DECLARE r text;
BEGIN
	EXECUTE format('SELECT count(*) || '':'' || coalesce(string_agg(id || ''/'' || round(s::numeric, 3), '','' ORDER BY s DESC, id) FILTER (WHERE rn <= 25), '''')
		FROM (SELECT id, sift.score(ctid) AS s, row_number() OVER (ORDER BY sift.score(ctid) DESC, id) AS rn FROM %s WHERE body @@> %L) t', tbl, qq) INTO r;
	RETURN r;
END $$;
SELECT q, left(answer('docs_buf', q), 60) AS buffer_answer FROM q WHERE answer('docs_buf', q) IS DISTINCT FROM answer('docs_seg', q) ORDER BY n;
 q | buffer_answer 
---+---------------
(0 rows)

SELECT count(*) AS queries, count(*) FILTER (WHERE answer('docs_buf', q) NOT LIKE '0:%') AS with_matches FROM q;
 queries | with_matches 
---------+--------------
      21 |           20
(1 row)

DELETE FROM docs_buf WHERE id % 10 = 3;
DELETE FROM docs_seg WHERE id % 10 = 3;
VACUUM docs_seg;
CREATE FUNCTION ids(tbl text, qq text) RETURNS text LANGUAGE plpgsql AS $$
DECLARE r text;
BEGIN
	EXECUTE format('SELECT count(*) || '':'' || coalesce(md5(string_agg(id::text, '','' ORDER BY id)), '''') FROM %s WHERE body @@> %L', tbl, qq) INTO r;
	RETURN r;
END $$;
SELECT q FROM q WHERE ids('docs_buf', q) IS DISTINCT FROM ids('docs_seg', q) ORDER BY n;
 q 
---
(0 rows)

DROP FUNCTION answer(text, text), ids(text, text);
DROP TABLE q, docs_buf, docs_seg;
DROP EXTENSION pg_sift;