Setting collation_connection with init_connect regression

I am using the collation "utf8mb4_swedish_ci" in MySQL server and to get the correct collation for clients I have used a init_connect setting that sets the collation_connection variable to the servers collation

[mysqld]

character_set_server=utf8mb4

collation_server=utf8mb4_swedish_ci

init_connect='SET collation_connection = @@collation_database;'

After upgrading to at least maria db connector/j version 3.2.0 if I check the collation_connection variable its value is utf8mb4_0900_ai_ci which leads to a lot of illegal mix of collation errors when running my application.

From show variables like '%collation%'

|Variable_name                |Value             |

|-----------------------------|------------------|

|collation_connection         |utf8mb4_0900_ai_ci|

|collation_database           |utf8mb4_swedish_ci|

|collation_server             |utf8mb4_swedish_ci|

|default_collation_for_utf8mb4|utf8mb4_0900_ai_ci|

I have done some digging and the reason seems to be the implicit SET NAMES call that was introduced in CONJ-1092, since that call doesn't have a COLLATE it collation_connection becomes the default value. (commit 0c13897eea649a4195161a27b3f8884cbeda8415)

As far as I can tell init_connect statements are executed after that SET NAMES call though, so I don't understand why my init_connect doesn't work to reset collation_connection to the correct value.

I have also tried using SET NAMES with COLLATE in my init_connect but I get the same behavior (collation_connection is still utf8mb4_0900_ai_ci)