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()

0.69.0

Release notes

Enhancements

• Added UnsubscribeAck to the SystemCode enum for acknowledging unsubscribe requests

Bug fixes

• Fixed unsound AsRef<[u8]> for RecordBuf, which produced a slice longer than the
  buffer when the header length written through raw_buf_mut() exceeded the capacity
• Fixed incorrect count in DBN metadata causing decoding to read into the start of
  records

0.68.0

Release notes

Enhancements

• Added MetadataEncoder::encoded_len() for the number of bytes encode() will write
• Added DbnMetadata_encode and DbnMetadata_encoded_size to the C API for encoding
  metadata, including symbols, symbol mappings, and every symbology field
• Added SType to the generated C header
• Improved documentation for the min_price_increment_amount and inst_attrib_value
  fields

Breaking changes

• Removed encode_metadata from the C API. Use DbnMetadata_encode, which takes a
  DbnMetadataRef describing the full metadata rather than only the dataset, schema,
  and start

0.67.0

Release notes

Enhancements

• Added LISTING_ID, ISSUER_ID, and SECURITY_ID variants to SType, for use
  with the reference data API
• Added DbnDecoder_reset to the C API for decoding another stream with the same
  decoder

Breaking changes

• Updated the minimum supported tokio version to 1.47, which was released one year ago

0.66.0

Release notes

Enhancements

• Improved performance of record field accessors that convert a raw value to an enum, such
  as side(), action() and rtype(), by no longer copying the description when
  constructing the conversion error
• Improved performance of Record::raw_index_ts() for records with an unrecognized rtype
• Relaxed the bound on From<&R> for RecordRef from R: HasRType to R: Record, so a
  RecordRef can be constructed from any record type, including RecordRef itself
• Upgraded pyo3 to 0.29.2

0.65.0

Release notes

Enhancements

• Added ReplayDataAgedOut variant to ErrorCode for when the data for a replay
  subscription is no longer retained and the schema is incompatible with skipping records
• Added new publisher values for JPX
• Added a zstd feature, enabled by default, gating Zstandard compression and
  decompression, the DBZ format, and the Dyn readers, writers, decoders, and
  encoders
• Improved FFI library
  • The static library no longer bundles Zstandard
  • Added a DBN_C_HEADER_DIR environment variable to control where the generated
    dbn.h is written
  • Added native-static-libs.txt to the C static library release archives, recording
    the linker arguments the static library requires
  • Added a shared library to the libdbn_c release archives
• Upgraded time to 0.3.55
• Upgraded pyo3 to 0.29.1

Bug fixes

• Fixed DbnMetadata_end in the C API returning 0 for an unset query end instead of
  UNDEF_TIMESTAMP, the value it's encoded with on the wire

0.64.0

Release notes

Enhancements

• Added a push-based decoder to the C API, where the caller owns the read loop
  and any decompression, feeding bytes in and draining decoded metadata and
  records out
• Added a finish() method to DynWriter to flush data and finalize the
  output stream. For compressed data this will write the zstandard end-of-frame
  block
• Added finish() method to the Python Transcoder and a context manager
  to call this method on exit. This will ensure compressed output contains the
  end-of-frame block
• Upgraded time to 0.3.54

Breaking changes

• Replaced the file-descriptor-based DbnDecoder_* C API with the new
  push-based decoder; DbnDecoder_create now takes a DbnDecoderOptions and
  reports errors through a DbnDecoderError out parameter

Bug fixes

• Fixed AsRef<[u8]> for concrete record types encoding only
  size_of::<Self>() bytes instead of the full record_size() from the header
  length, which dropped appended ts_out bytes and corrupted output when
  re-encoding a typed record decoded from a stream with ts_out

0.63.0

Release notes

Enhancements

• Added an integer representation for VersionUpgradePolicy: the enum is now
  repr(u8) with discriminants starting at 1, derives TryFromPrimitive and
  IntoPrimitive, and in Python gained a value property, a from_int
  constructor, integer construction, and comparison against integers

0.62.0

Release notes

Enhancements

• Added new publisher values for Cboe Titanium Cboe Global Indices Feed
• Added StatType variants MwcbLevel1, MwcbLevel2, and MwcbLevel3
• Added StatType variants AuctionCollarReferencePrice, AuctionCollarUpperPrice,
  and AuctionCollarLowerPrice
• Upgraded time to 0.3.53

Breaking changes

• Renamed the following Venue, Dataset, and Publisher:
  • CGIF to MAIN
  • CGIF.TITANIUM to MAIN.CGIF
  • CGIF.TITANIUM.CGIF to MAIN.CGIF.MAIN
• Changed the default size field to 0 in MboMsg, TradeMsg, Mbp1Msg, Mbp10Msg,
  BboMsg, Cmbp1Msg, and CbboMsg

0.61.0

Release notes

Enhancements

• Added MatchAlgorithm variant Allocation
• Upgraded time version to 0.3.49
• Upgraded pyo3 to 0.29.0