recover() not called If the first log entry is rolled back

[losfair]

If the first entry is rolled back, LogManager::recover() is never called for the entire WAL.

To reproduce:

use okaywal::{LogManager, WriteAheadLog};

#[derive(Debug)]
struct TestLogManager;

impl LogManager for TestLogManager {
    fn recover(&mut self, entry: &mut okaywal::Entry<'_>) -> std::io::Result<()> {
        entry.read_all_chunks().unwrap();
        println!("recover: {:?}", entry);
        Ok(())
    }

    fn checkpoint_to(
        &mut self,
        _last_checkpointed_id: okaywal::EntryId,
        _checkpointed_entries: &mut okaywal::SegmentReader,
        _wal: &okaywal::WriteAheadLog,
    ) -> std::io::Result<()> {
        Ok(())
    }
}

fn main() {
    let _ = std::fs::remove_dir_all("./testdata");
    let log = WriteAheadLog::recover("./testdata", TestLogManager).unwrap();

    // this entry is rolled back
    log.begin_entry().unwrap();

    let mut entry = log.begin_entry().unwrap();
    entry.write_chunk(b"test").unwrap();
    entry.commit().unwrap();

    log.shutdown().unwrap();

    WriteAheadLog::recover("./testdata", TestLogManager).unwrap();
}

This prints nothing. If the first log.begin_entry() is commented out, recover() is called correctly.

[ecton]

Thank you for reporting this! I will be releasing v0.3.1 with a fix for this shortly.

[ecton]

The release is out, and I've yanked all previous versions on crates.io due to the risk of data loss. Thank you again for reporting this and providing such a simple test case.