BUG #14563: count(*) gives a wrong result in PostgreSQL for some regex with pg_trm/GIN index

The following bug has been logged on the website:

Bug reference: 14563
Logged by: Alexey Isayko
Email address: isayko(dot)alexey(at)gmail(dot)com
PostgreSQL version: 9.6.1
Operating system: Windows 10.0.14393 x64
Description:

When using some RE to filter rows on gin-indexed column with pg_trm,
count(*) could give a wrong result (zero instead of real count):

CREATE EXTENSION pg_trgm;
CREATE TABLE t (
s text
);
CREATE INDEX ON t USING gin (s gin_trgm_ops);
INSERT INTO t VALUES ('12-34');

SET enable_seqscan = OFF; -- to force using the index on our small table
SELECT count(*) FROM t WHERE s~'\d{2}[-]?\d{2}'; -- gives me 0 and it's
wrong

But if we will turn on seq scans, we will get the correct result:

SET enable_seqscan = ON;
SELECT count(*) FROM t WHERE s~'\d{2}[-]?\d{2}'; -- gives me 1 and it's
correct

Same behaviour I get with 9.5.4 (windows and linux) and 9.6.1 (windows)
versions of the PostgreSQL