-- test/sql/pending_distinct.sql
--
-- aminsert writes one pending entry per DISTINCT trigram of a row, as the
-- bulk build does.  It used to write one per occurrence; nothing reads the
-- position, so repeats were pure waste, and on whole source files (each
-- trigram ~9x) the pending list reached ~24x the size of the text it
-- indexed (field report 2026-09-29: 137 GB of index over 163 MB).
CREATE EXTENSION IF NOT EXISTS pg_tre;
psql:test/sql/pending_distinct.sql:9: NOTICE:  extension "pg_tre" already exists, skipping
CREATE EXTENSION
DROP TABLE IF EXISTS pdi;
psql:test/sql/pending_distinct.sql:11: NOTICE:  table "pdi" does not exist, skipping
DROP TABLE
CREATE TABLE pdi (id serial PRIMARY KEY, body text)
    WITH (autovacuum_enabled = off);
CREATE TABLE
SET client_min_messages = warning;
SET
CREATE INDEX pdi_tre ON pdi USING tre (body);
CREATE INDEX
RESET client_min_messages;
RESET
-- 'abcabc...' has three distinct trigrams (abc, bca, cab) and ~30,000
-- occurrences.  One entry each fits a single pending page; one per
-- occurrence took ~90.
INSERT INTO pdi (body) VALUES (repeat('abc', 10000));
INSERT 0 1
SELECT n_pages AS pending_pages
FROM tre_page_kind_histogram('pdi_tre') WHERE page_kind = 'pending';
 pending_pages 
---------------
             1
(1 row)

-- Still found, via the pending overlay and after the merge.
SET enable_seqscan = off;
SET
SELECT count(*) AS found_pending FROM pdi WHERE body ~ 'cabca';
 found_pending 
---------------
             1
(1 row)

RESET enable_seqscan;
RESET
VACUUM pdi;
VACUUM
SET enable_seqscan = off;
SET
SELECT count(*) AS found_merged FROM pdi WHERE body ~ 'cabca';
 found_merged 
--------------
            1
(1 row)

RESET enable_seqscan;
RESET
DROP TABLE pdi;
DROP TABLE