Count all columns in a table individually - sql

Question

Trying to do some data exploration on a table to count all columns and all unique strings in the row individually like the below output:
+------------+-------------+------------+----------+
| table_name | column_name | distinct | count(*) |
| | | row_string | |
+------------+-------------+------------+----------+
| customer | state | WA | 15 |
+------------+-------------+------------+----------+
| customer | state | NSW | 786 |
+------------+-------------+------------+----------+
| customer | state | SA | 51 |
+------------+-------------+------------+----------+
| ... | ... | ... | ... |
+------------+-------------+------------+----------+
| customer | zip_code | 3563 | 33 |
+------------+-------------+------------+----------+
| customer | zip_code | 7583 | 52 |
+------------+-------------+------------+----------+
| customer | zip_code | 3453 | 553 |
+------------+-------------+------------+----------+
| customer | zip_code | 2132 | 211 |
+------------+-------------+------------+----------+
| ... | ... | ... | ... |
+------------+-------------+------------+----------+
I've been doing something like this:
select state, count(*)
from customer
group by state
union
select zip_code, count(*)
from customer
group by zip_code
union
...
however this is not efficient assuming you have heaps of columns in the table. Is there a more effective way to achieve this?

Answer 1

(The below was posted before I looked at the link that Martin Smith posted in the comments above. That approach using XML appears simpler and likely has roughly similar performance.)
You can try generating dynamic SQL that looks like:
select
cast(N'customer' AS sysname) AS table_name,
C.column_name,
C.value,
count(*) AS [count]
from [customer] T
cross apply (
values
(cast(N'state' AS sysname), cast(T.[state] AS nvachar(max)),
(cast(N'zip_code' AS sysname), cast(T.[zip_code] AS nvachar(max))
) C(column_name, value)
group by column_name, value
The CAST() operations might be seem like overkill, but they may be necessary to prevent truncation if shorter data is followed by longer data. For some data types like DATETIME or FLOAT, you might want to be more precise in the formatting, perhaps even intentionally limiting excess precision.
When injecting table names and column names into the generated SQL, use QUOTENAME(...) and QUOTENAME(..., '''') to safely [quote] or 'quote' the injected values.