Ed: solved with the help of the async_stream crate.

I’m struggling with the borrow checker!

My problem: I’m using actix-web and rusqlite. I want to return an unlimited number of records from an rusqlite query, and actix provides a Stream trait for that kind of thing. You just impl the trait and return your records from a poll_next() fn.

On the rusqlite side, there’s this query_map that returns an iterator of records from a query. All I have to do is smush these two features together.

So the plan is to put the iterator returned by query_map into a struct that impls Stream. Problem is the lifetime of a var used by query_map. How to make the var have the same lifetime as the iterator??

So here’s the code:

pub struct ZkNoteStream<'a, T> {
  rec_iter: Box<dyn Iterator<Item = T> + 'a>,
}

// impl of Stream just calls next() on the iterator.  This compiles fine.
impl<'a> Stream for ZkNoteStream<'a, serde_json::Value> {
  type Item = serde_json::Value;

  fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
    Poll::Ready(self.rec_iter.next())
  }
}

// init function to set up the ZkNoteStream.
impl<'a> ZkNoteStream<'a, Result<ZkListNote, rusqlite::Error>> {
  pub fn init(
    conn: &'a Connection,
    user: i64,
    search: &ZkNoteSearch,
  ) -> Result<Self, Box<dyn Error>> {
    let (sql, args) = build_sql(&conn, user, search.clone())?;

    let sysid = user_id(&conn, "system")?;
    let mut pstmt = conn.prepare(sql.as_str())?;

    // Here's the problem!  Borrowing pstmt.
    let rec_iter = pstmt.query_map(rusqlite::params_from_iter(args.iter()), move |row| {
      let id = row.get(0)?;
      let sysids = get_sysids(&conn, sysid, id)?;
      Ok(ZkListNote {
        id: id,
        title: row.get(1)?,
        is_file: {
          let wat: Option<i64> = row.get(2)?;
          wat.is_some()
        },
        user: row.get(3)?,
        createdate: row.get(4)?,
        changeddate: row.get(5)?,
        sysids: sysids,
      })
    })?;

    Ok(ZkNoteStream::<Result<ZkListNote, rusqlite::Error>> {
      rec_iter: Box::new(rec_iter),
    })
  }
}

And here’s the error:

error[E0515]: cannot return value referencing local variable `pstmt`
   --> server-lib/src/search.rs:170:5
    |
153 |       let rec_iter = pstmt.query_map(rusqlite::params_from_iter(args.iter()), move |row| {
    |                      ----- `pstmt` is borrowed here
...
170 | /     Ok(ZkNoteStream::<Result<ZkListNote, rusqlite::Error>> {
171 | |       rec_iter: Box::new(rec_iter),
172 | |     })
    | |______^ returns a value referencing data owned by the current function

So basically it boils down to pstmt getting borrowed in the query_map call. It needs to have the same lifetime as the closure. How do I ensure that?

  • hallettj@beehaw.org
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    1 year ago

    I’m glad you found a workaround. I didn’t want to be defeated by the callback idea that I had yesterday so I worked on it some more and came up with a similar-but-different solution where ZnsMaker stores pstmt paired with a closure that borrows it. This is again a simplified version of your code that is self-contained:

    https://play.rust-lang.org/?version=stable&amp;mode=debug&amp;edition=2021&amp;gist=5bd6fb7c1cbf1c9c44c8f4bdbb1e8074

    The closure solution avoids the need to pass conn to make_stream. I don’t know if it would fix the new borrowing error that you got since I did not reproduce that error. But I think it might.

    Does znsstream need to outlive znsm? If so I think my solution solves the problem. Does znsstream need to outlive conn? That would be more complicated.

    Edit: Oh! You don’t need to put both the pstmt and a closure in ZnsMaker. Instead you can just store a closure if you move ownership of pstmt into the closure. That ensures that pstmt lives as long as the function that produces the iterator that you want:

    https://play.rust-lang.org/?version=stable&amp;mode=debug&amp;edition=2021&amp;gist=491258a7dc9bcad9dab08632d68c026b

    Edit: Lol you don’t need ZnsMaker after all. See my other top-level comment. I learned some things working on this, so time well spent IMO.