Regression in 1.4 working_ssb_compareLogOnly/sub/order_limit_sub

working_ssb_compareLogOnly/sub/order_limit_sub, the second query gives the wrong answer:

select * from
(select lo_orderkey, count
from lineorder where lo_orderkey in
(select *
from (select lo_orderkey
from lineorder
join dateinfo
on lo_orderdate = d_datekey
where lo_orderkey <= 900
group by lo_orderkey
order by sum(lo_ordtotalprice), lo_orderkey desc
limit 5
) alias1
)
group by lo_orderkey
order by 1, 2 asc
limit 4
) alias2
order by 2, 1 desc;

---------------------+

lo_orderkey

count

---------------------+

---------------------+

In 1.2 we get the correct answer:
---------------------+

lo_orderkey

count

---------------------+

---------------------+

I ran the interior queries separately and got correct answers. It doesn't seem to break until the final wrapper.