Converting from VARCHAR to DATE - sql

Question

I have a db where I have converted 2 date columns to varchar for the purpose of getting 1 column substringed into another. However, now I don't seem to be able to convert the datatype when I try to use:
ALTER TABLE datacomplete
ALTER COLUMN yearmonth TYPE DATE; /*Can't find a way to specify a format*/
It throws this error:
ERROR: column "yearmonth" cannot be cast automatically to type date
Hint: You might need to specify "USING yearmonth::date".
I'm not sure how to use that command at all, could anyone potentially assist?
My first column is in the format of yyyy-mm-dd, however I'd like it to be yyyymm only, but I'm guessing this is easier once I convert the datatype to date and I can somehow switch formats.
The second column only shows the year so I need to convert it to date as format 'yyyy'.
UPDATE: The first one was solved, now I need to convert the second to 'yyyy'
ALTER TABLE pscomplete_1 ALTER COLUMN "year" TYPE DATE USING "year"::date;
It throws this error
15:12:51 [ALTER - 0 rows, 1.062 secs] [Code: 0, SQL State: 22007] ERROR: >invalid input syntax for type date: "2016"
... 1 statement(s) executed, 0 rows affected, exec/fetch time: 1.062/0.000 sec >[0 successful, 1 errors]

Answer 1

The USING keyword allows you to give the translation function to PostgreSQL.
For your first column it is easy for you already have a correct DATE format:
ALTER TABLE datacomplete ALTER COLUMN yearmonth TYPE DATE USING yearmonth::DATE;
For your second column it is unclear for PostgreSQL which exact date you want. Let's say we want the first of January of the given year:
ALTER TABLE datacomplete ALTER COLUMN year TYPE DATE USING (year || '-01-01')::DATE;

Answer 2

I understood you were starting from a VARCHAR column in the format 'YYYY-MM-DD'.
So I'll do the same.
And you want a column in the date format, and you want a yearmonth in all-digits format.
If I'm not forced to use leading zeroes in an all-digits column, I prefer INT to string columns.
And I prefer to make the same derivation only once if I can do that.
This is why I use a WITH clause (global table expression) to cast the varchar to date, and then I use the resulting date for the DATE_PART() function I use to create the yearmonth column. I have seen very often that date arithmetics are safer and often faster than subtring-ing the date literal (remember, the Americans format dates differently from the Europeans, there are different formats within Europe, and also in Asia, and not all like the ISO date format). So I derive yearmonth as the year multiplied by 100, plus the month, and as DATE_PART() returns a float, I cast the whole expression to INT.
So here goes:
WITH foo(varchar_dt) AS (
SELECT '2017-01-11'
UNION ALL SELECT '2016-12-11'
UNION ALL SELECT '2016-11-11'
UNION ALL SELECT '2016-10-11'
)
, foo_with_date AS (
SELECT
varchar_dt
, CAST(varchar_dt AS DATE) AS the_date
FROM foo
)
SELECT
varchar_dt
, the_date
, CAST(DATE_PART('year',the_date)*100 + DATE_PART('month',the_date) AS INT) AS yearmonth
FROM foo_with_date
;
varchar_dt|the_date |yearmonth
2017-01-11|2017-01-11| 201,701
2016-12-11|2016-12-11| 201,612
2016-11-11|2016-11-11| 201,611
2016-10-11|2016-10-11| 201,610
I can't help myself - I find this much cleaner, and filtering by yearmonth would become filtering by an integer, which is always at least a little bit faster than strings.
Happy playing
Marco the Sane