gwenn / r2d2-sqlite Goto Github PK

Hi, I'm happily using Pool::new(SqliteConnectionManager::file(foo)) in my production code and everything is working fine.

However now I'm writing unit tests which I want to run on an in-memory instance of sqlite.

I create my connection pool like:

pub struct DbConnection(Pool<SqliteConnectionManager>);
impl DbConnection {
    pub fn test() -> Self {
        let manager = SqliteConnectionManager::memory();
        let pool = Pool::new(manager).unwrap();
        let db_conn = DbConnection(pool);
        db_conn.initialise().unwrap();
        db_conn
    }

   fn initialise(&self) -> Result<(), Error> {
        let conn = &*self.0.clone().get()?;
        conn.execute_batch("create table if not exists foo (...)")?;
        Ok(())
   }
}

and my unit tests (which just insert into that table) fail with "table foo does not exist".

If I instead change it to

        let manager = SqliteConnectionManager::memory();
        let pool = Pool::builder().max_size(1).build(manager).unwrap();

then my test passes. If I make that max_size(2) then it fails again.

The only conclusion I can come to is that when the initialisation and the test use the same connection then everything is fine, but if they use different ones then the test does not see the result of the initialisation call.
Is this intended? It seems like very surprising behaviour if so.

The full code can be found here: https://github.com/djmcgill/dominions-5-status/blob/master/src/db/mod.rs

P. S. I'm fairly new to Rust so if I'm using something unidiomatically please do let me know.