test: clickhouse integration tests

[eitsupi]

Close #2813, based on #2814

Same as #2811

[eitsupi]

Oh, I see. The order of the Clickhouse results seems to be random.
https://stackoverflow.com/questions/54786494/clickhouse-query-row-order-behaviour

I have observed that the order of the results changes each time I run the following query.

WITH table_0 AS (
  SELECT
    13 AS x_int,
    13.0 AS x_float,
    5 AS k_int,
    5.0 AS k_float
  UNION
  ALL
  SELECT
    -13 AS x_int,
    -13.0 AS x_float,
    5 AS k_int,
    5.0 AS k_float
  UNION
  ALL
  SELECT
    13 AS x_int,
    13.0 AS x_float,
    -5 AS k_int,
    -5.0 AS k_float
  UNION
  ALL
  SELECT
    -13 AS x_int,
    -13.0 AS x_float,
    -5 AS k_int,
    -5.0 AS k_float
)
SELECT
  (x_int / k_int)
FROM
  table_0

@max-sixty @aljazerzen Do you think we can add a sort at the end of every PRQL query so that even systems like ClickHouse match the results?

[max-sixty]

@max-sixty @aljazerzen Do you think we can add a sort at the end of every PRQL query so that even systems like ClickHouse match the results?

Yes I think that's OK — it doesn't limit any SQL functionality. Feel free to merge this with most queries excluded and then follow-up with changed queries unexcluded.

I also think we could have the Unsupported dialects look for e.g. # run_clickhouse, rather than the absence of e.g. # skip_clickhouse, if that lets us add queries without worrying about the tail of dialects. Supported dialects would still require explicit skipping.

(Though maybe that's harder than just adding # skip_clickhouse to the others, feel free to do whatever is easier)

[max-sixty]

Great!

[max-sixty]

Depending on how many exclusions we need, we could move this down a rung. (Probably we should do that for MsSql too)

[eitsupi]

Note: ClickHouse doesn't have lag/lead functions.
https://clickhouse.com/docs/en/sql-reference/window-functions

[eitsupi]

The problem that we are currently experiencing is that the results of the following query.

prql/prql-compiler/tests/integration/queries/switch.prql

Lines 1 to 8 in f35c7e5

	from tracks
	sort milliseconds
	select display = case {
	composer != null => composer,
	genre_id < 17 => 'no composer',
	true => f'unknown composer'
	}
	take 10

This is not a connection problem, as the following SQL via MySQL CLI reproduces it.

WITH table_0 AS (
  SELECT
    CASE
      WHEN composer IS NOT NULL THEN composer
      WHEN genre_id < 17 THEN 'no composer'
      ELSE 'unknown composer'
    END AS display,
    milliseconds
  FROM
    tracks
  ORDER BY
    milliseconds
  LIMIT
    10
)
SELECT
  display
FROM
  table_0
ORDER BY
  milliseconds

+---------------+
| display       |
+---------------+
| Samuel Rosa   |
|               |
|               |
|               |
| L. Muggerud   |
|               |
| L. Muggerud   |
|               |
| Gilberto Gil  |
| Chico Science |
+---------------+

[eitsupi]

I expected it to be a case when issue, but the following example works.......

from [
    {x = '3'},
    {x = '1'},
]
select x = case {
    x == '3' => 'fiz',
    true => x,
}

MySQL [dummy]> WITH table_0 AS (
    ->   SELECT
    ->     '3' AS x
    ->   UNION
    ->   ALL
    ->   SELECT
    ->     '1' AS x
    -> )
    -> SELECT
    ->   CASE
    ->     WHEN x = '3' THEN 'fiz'
    ->     ELSE x
    ->   END AS x
    -> FROM
    ->   table_0;
+------+
| x    |
+------+
| fiz  |
| 1    |
+------+
2 rows in set (0.025 sec)

[eitsupi]

Oh, I understand. Perhaps this is being read as an empty character rather than NULL...

WITH table_0 AS (
  SELECT
    CASE
      WHEN composer != '' THEN composer
      WHEN genre_id < 17 THEN 'no composer'
      ELSE 'unknown composer'
    END AS display,
    milliseconds
  FROM
    tracks
  ORDER BY
    milliseconds
  LIMIT
    10
)
SELECT
  display
FROM
  table_0
ORDER BY
  milliseconds

+------------------+
| display          |
+------------------+
| Samuel Rosa      |
| no composer      |
| no composer      |
| no composer      |
| L. Muggerud      |
| no composer      |
| L. Muggerud      |
| unknown composer |
| Gilberto Gil     |
| Chico Science    |
+------------------+

[eitsupi]

https://clickhouse.com/docs/en/operations/settings/formats#format_csv_null_representation
-> Seems not related to...

It seems to be correctly recognized as NULL when reading csv. Perhaps a problem with the table type definition?
https://clickhouse.com/docs/en/sql-reference/data-types/nullable

> SELECT composer FROM file('/var/lib/clickhouse/user_files/chinook/tracks.csv') limit 5;
+------------------------------------------------------------------------+
| composer                                                               |
+------------------------------------------------------------------------+
| Angus Young, Malcolm Young, Brian Johnson                              |
| NULL                                                                   |
| F. Baltes, S. Kaufman, U. Dirkscneider & W. Hoffman                    |
| F. Baltes, R.A. Smith-Diesel, S. Kaufman, U. Dirkscneider & W. Hoffman |
| Deaffy & R.A. Smith-Diesel                                             |
+------------------------------------------------------------------------+
5 rows in set (0.010 sec)

[eitsupi]

Ok, Nullable works...

> WITH table_0 AS (
    ->   SELECT
    ->     CASE
    ->       WHEN composer IS NOT NULL THEN composer
    ->       WHEN genre_id < 17 THEN 'no composer'
    ->       ELSE 'unknown composer'
    ->     END AS display,
    ->     milliseconds
    ->   FROM
    ->     tracks
    ->   ORDER BY
    ->     milliseconds
    ->   LIMIT
    ->     10
    -> )
    -> SELECT
    ->   display
    -> FROM
    ->   table_0
    -> ORDER BY
    ->   milliseconds;
+------------------+
| display          |
+------------------+
| Samuel Rosa      |
| no composer      |
| no composer      |
| no composer      |
| L. Muggerud      |
| no composer      |
| L. Muggerud      |
| unknown composer |
| Gilberto Gil     |
| Chico Science    |
+------------------+
10 rows in set (0.011 sec)

[eitsupi]

It seems working now 🎉

[max-sixty]

Great work!!