Read from the UI

The UI never writes. It reads through its own read-only connection and reads again whenever a source says something changed. Every database is opened in WAL mode with synchronous=NORMAL, so a read never queues behind a source’s commit.

In the chart a source commits 10,000-row passes back to back for three seconds while the UI reads a page of 50 rows and the sync row every millisecond. With pito-sync the UI got 2,305 reads, the slowest 2.84 ms. The same code on a plain connection with SQLite’s defaults (rollback journal, synchronous=FULL) got 88 reads, the slowest stuck behind the writer for 1.23 s.

Open a reader

let Some(ui) = pito_sync::reader(&spec)? else {
    return Ok("not built yet".into());
};

reader(&spec) returns Ok(None) until a source has built the file, and while the version stored in the file differs from the Spec’s, which happens for a moment after an upgrade, before the source’s thread rebuilds it. It returns Err only when the file exists and can’t be read.

Re-read on each Changed

spawn and spawn_checked take the sending half of a channel. The UI owns the receiving half and reads again on every Changed(name):

for Changed(name) in changes {
    println!("{name:<9} {}", show(&spec, &name)?);
}

A desktop or terminal app does the same in its event loop, with try_recv on the channel so the loop never blocks.

Show “syncing…”

A pass sends Changed when it starts and again when it ends. While it runs, the source’s sync row has started_at later than finished_at:

let sync = pito_sync::sync(&ui, name)
    .map_err(|e| e.to_string())?
    .unwrap_or_default();
if sync.started_at > sync.finished_at {
    return Ok("syncing…".into());
}

sync(&conn, source) returns None for a source that has never run; unwrap_or_default() turns that into an empty row.

Read your own tables

The rest is plain SQL on your own tables through the same connection:

let (posts, newest): (i64, Option<String>) = ui
    .query_row(
        "SELECT count(*), (SELECT title FROM posts ORDER BY id DESC LIMIT 1) FROM posts",
        [],
        |row| Ok((row.get(0)?, row.get(1)?)),
    )
    .map_err(|e| e.to_string())?;

With the failure check added, this is the show function of the feed example, and these are the lines it prints:

feed      syncing…
feed      6 posts, newest "One file per app"
feed      syncing…
feed      unreachable: the server did not answer (6 posts kept)