fix: move loading httpfs and setting extension directory to beginning of connection

[rebasedming]

Ticket(s) Closed

• Closes #

What

Moves loading httpfs and setting the extension directory to the start of the DuckDB connection so that the settings are available to the rest of the connection.

Why

How

Tests

[kysshsy]

Happy to see you already submitted a PR to fix this issue. 😁
And I am investigating why the CI didn't pass, my preliminary findings suggest that calling execute within init_globals might be causing the issue. This is because execute triggers the initialization of the connection, which in turn calls init_globals

[kysshsy]

And maybe we don't need to INSTALL httpfs manually as Duckdb will load it automatically

[kysshsy]

It appears that the deadlock in the connection is indeed caused by the recursive call to init_global within call_once. We can refactor it. like

pub fn set_duckdb_extension_directory(conn: &Connection) -> Result<usize> {
    let data_dir = unsafe {
        CStr::from_ptr(pgrx::pg_sys::DataDir)
            .to_str()
            .map_err(|e| anyhow::anyhow!("Failed to convert DataDir to &str: {}", e))?
    };
    conn.execute(format!("SET extension_directory = '{data_dir}'").as_str(), []).map_err(|err| anyhow!("{err}"))
}

[philippemnoel]

It appears that the deadlock in the connection is indeed caused by the recursive call to init_global within call_once. We can refactor it. like

pub fn set_duckdb_extension_directory(conn: &Connection) -> Result<usize> {
    let data_dir = unsafe {
        CStr::from_ptr(pgrx::pg_sys::DataDir)
            .to_str()
            .map_err(|e| anyhow::anyhow!("Failed to convert DataDir to &str: {}", e))?
    };
    conn.execute(format!("SET extension_directory = '{data_dir}'").as_str(), []).map_err(|err| anyhow!("{err}"))
}

Thank YOU for fixing this!