0.70.0

Release notes

Enhancements

• Added DbnDecoder_process_many to the C API for decoding a batch of records in a
  single call, which is significantly faster than calling DbnDecoder_process for each
  record
• Added DbnDecoder_take_metadata to the C API for taking ownership of the metadata
  after a Metadata result
• Sped up decode_ref() and decode_record_ref() on the sync and async record
  decoders, which now decode records in batches
• Added DbnFsm::process_batch() for decoding a batch of records in a single call,
  along with has_buffered_record() and next_buffered_record() for draining the
  batch
• Added new venue, dataset, and publishers for Databento Core Indices
• Upgraded async-compression to 0.4.44

Breaking changes

• Replaced Record::header() with a method per header field: raw_rtype(),
  publisher_id(), instrument_id(), and raw_ts_event(), joining the existing
  record_size(), rtype(), and publisher(). RecordRef, RecordRefMut,
  RecordBuf, RecordEnum, and RecordRefEnum keep a header() method of their own,
  and concrete record types keep their hd field
• Record::raw_rtype() returns u16 and Record::instrument_id() returns u64, wider
  than the corresponding RecordHeader fields, which are unchanged
• RType::try_into_schema() takes a u16 and returns None for values outside the
  range of the RecordHeader rtype field
• Removed the read_more and metadata out parameters from DbnDecoder_process. Call
  DbnDecoder_take_metadata after a Metadata result

Bug fixes

• Fixed DbnFsm::process_all() and DbnFsm::process_many() returning Record with no
  records instead of ReadMore when the buffer held no complete record
• Fixed unsound RecordRefMut accessors, which returned references outliving the borrow
  of self and let safe code hold two mutable references to the same record, or a
  shared reference aliasing a mutable one. get_mut() and get_mut_unchecked() now
  take &mut self, and header() returns a reference borrowing self
• Fixed undefined behavior when mutating a record through a RecordRefMut, where the
  write went through a pointer that permitted only reads, or that did not span the
  whole record
• Fixed DbnFsm::process_all() and DbnFsm::process_many() discarding the records
  decoded before an invalid record. They now return those records and the error on the
  following call, matching DbnFsm::process()