Re: Statistics and selectivity estimation for ranges

It looks like there are still some problems with this patch.

  CREATE TABLE foo(ir int4range);
  insert into foo select 'empty' from generate_series(1,10000);
  insert into foo select int4range(NULL, g, '(]')
    from generate_series(1,1000000) g;
  insert into foo select int4range(g, NULL, '[)')
    from generate_series(1,1000000) g;
  insert into foo select int4range(g, ((g*1.01)+10)::int4, '[]')
    from generate_series(1,1000000) g;
  CREATE TABLE bar(ir) AS select * from foo order by random();
  ANALYZE bar;

Now:
  EXPLAIN ANALYZE SELECT * FROM bar
    WHERE ir @> int4range(10000,20000);

The estimates are "-nan". Similar for many other queries.