v0.3.1

Fixed

• Rolling back the first entry of a log file caused its data to become skipped
  during recovery due to a small logic bug that rewrote the file's header if no
  entries had been fsynced. Thanks to @losfair for the succinct test case in
  issue #14.

v0.3.0

Added

• WriteAheadLog::checkpoint_active() is a new function that checkpoints the
  currently stored data, regardless of whether the configured thresholds are
  met. This function returns after the active file has been rotated and the
  checkpointing thread has been notified of the file to checkpoint. Thanks to
  @blakesmith for implementing this in #11

v0.2.0

Breaking Changes

• LogManager::checkpoint_to now has an additional parameter, wal: &WriteAheadLog. This is provided for convenience because it may be necessary
  to randomly access information in the WAL while performing a checkpointing
  operation.

Added

• LogPosition::serialize_to/LogPosition::deserialize_from provide methods
  for reading and writing a LogPosition from an arbitrary Write/Read
  implementor (respectively). This uses a fixed-length serialization with a
  length of LogPosition::SERIALIZED_LENGTH -- 16 bytes.

Changed

• WriteAheadLog::shutdown() now no longer requires all instances of
  WriteAheadLog to be dropped to succeed.

Fixed

• When the WAL recycles a segment file, the LogPosition returned is now
  correct. Previously, returned LogPositions would contain the segment file's
  old id, causing those positions to be unreadable.
• When reading from a LogPosition, if the data has not been flushed or
  synchronized to disk yet, the read will be blocked until the sync operation
  finishes.

v0.1.0 - Initial Public Preview

This release is not ready for anyone to use in production. The file format had changes as recently as three days ago, and should be considered unstable at this stage. With that caveat aside, I would love any feedback on this crate's design.