Select the last exchange value for each currency. SQL - sql

Question

Let's consider we have table of currencies. The task is to derive the last update of a price for each currency (name).
My endeavor:
SELECT name, date, price
FROM (
SELECT *,
ROW_NUMBER() OVER (
PARTITION BY name
ORDER BY date DESC
) as RN
FROM currency
) X where RN = 1;
My solution virtually cover everything, except two dollar's fields:
| USD | 2006-03-04 | 8 | and | USD | 2007-03-04 | 8 |
Technically, I comprehend why does it happen. I define the RN = 1, which selects first row for each currency (considering that the dates for each currency are in descending order). In the problem, described above the last update occurs 2006-03-04, and generally its order is second.
However, I have no idea, how to formulate a request to choose the MIN of the dates in the last subgroup.
If you have an idea how to do that - I'ill be very thankful!

Answer 1

I think you want the rows where the currency changes. If that interpretation is correct, use lag():
select c.*
from (select c.*, lag(name) over (order by date) as prev_name
from currency c
) c
where prev_name is null or prev_name <> name;
Note that standard SQL has a way to simplify the where clause using a NULL safe comparison:
where prev_name is distinct from name
However, not all databases support this (or similar) syntax.
EDIT:
I think my above interpretation is incorrect. You want the second to last row because the price does not change. So:
select c.*
from (select c.*, lag(price) over (partition by name order by date) as prev_price
from currency c
) c
where prev_price is null or prev_price <> price;
Then to get one row per currency:
select c.*
from (select c.*,
row_number() over (partition by name order by date desc) as seqnum
from (select c.*, lag(price) over (partition by name order by date) as prev_price
from currency c
) c
where prev_price is null or prev_price <> price
) c
where seqnum = 1
EDIT II:
In Postgres, the last query is more simply written as:
select distinct on (name) c.*
from (select c.*, lag(price) over (partition by name order by date) as prev_price
from currency c
) c
where prev_price is null or prev_price <> price
order by name, date desc