Unexpected result of using in DSN string collation and charset parameters at the same time.

[soslanco]

Issue description

Result of using in DSN string charset and non default collation parameters together not as expected.

In file github.com/go-sql-driver/mysql/connection.go executed SET NAMES sql command without COLLATE parameter. Therefore connection collation = default collation for charset.

Example code

sql.Open("mysql", "user:password@/db?charset=utf8mb4&collation=utf8mb4_bin")

Result

collation_connection: utf8mb4_general_ci

Configuration

Driver version (or git SHA): bc14601

Go version: go1.9.3 linux/amd64

Server version: MariaDB 10.2.12

Server OS: RHEL7

[dolmen]

Please add the Go code that shows how you query the current charset and the current collation.

[dolmen]

var collation string
db.QueryRow(`SELECT @@collation_connection`).Scan(&collation)
fmt.Println("Collation:", collation)

[dolmen]

Note that the documentation for charset says:

Usage of the charset parameter is discouraged because it issues additional queries to the server. Unless you need the fallback behavior, please use collation instead.

The workaround is to set only collation and not charset.

[soslanco]

No problems while using charset and collation parameters separately.
Issue is when using both parameters together.
Possible solution: SET NAMES 'charset_name' [COLLATE 'collation_name']

[soslanco]

Go code:
db, _ = sql.Open("mysql", "user:password@/db?charset=utf8mb4&collation=utf8mb4_bin")
rows, _ = db.Query("SHOW VARIABLES LIKE 'collation%'")

Output:
collation_connection: utf8mb4_general_ci
collation_database: utf8mb4_bin
collation_server: utf8mb4_bin

[methane]

charset is deprecated, and not recommended. Why do you use both?

[soslanco]

It's not my code: https://github.com/go-sql-driver/mysql/blob/master/connection.go#L62

Using both of them not deprecated: https://dev.mysql.com/doc/refman/5.7/en/charset-connection.html

Setting character_set_connection to charset_name also implicitly sets collation_connection to the default collation for charset_name. It is unnecessary to set that collation explicitly. To specify a particular collation, use the optional COLLATE clause:

SET NAMES 'charset_name' COLLATE 'collation_name'

[methane]

@soslanco I meant about parameter of DSN. collation=utf8mb4_bin doesn't use SET NAMES.

See here. https://github.com/go-sql-driver/mysql#charset

Usage of the charset parameter is discouraged because it issues additional queries to the server. Unless you need the fallback behavior, please use collation instead.

[dolmen]

@methane Users can be wrong, despites a good intent (specifying both charset and collation in a connection string looks good until you read the documentation). Worse: the fact that charset overrides collation is not documented.

So the driver should catch this user error because it can: it should ignore charset if collation is set.

[methane]

I think there are two long-term solution:

• Remove charset entirely
• Return error when both are set.
  • When only charset is specified, use default collation for the charset. Don't use SET NAMES.

[kaihendry]

I had issues with my collation_connection being utf8mb4_general_ci with charset=utf8mb4&collation=utf8mb4_unicode_ci. Removing charset & keeping just &collation=utf8mb4_unicode_ci in the DSN correctly results in collation_connection being utf8mb4_unicode_ci.

That's quite a gotcha.