0.56.0

Release notes

Enhancements

• Added DecodeRecordRef::decode_buf_iter() for lazily iterating owned RecordBufs
  via a standard Iterator
• Added AsyncDecodeRecordRef::decode_stream() for lazily decoding owned RecordBufs
  via a futures::Stream
  • Added async-stream and futures-core dependencies (gated on async feature)
• Improved DbnFsm decode throughput by making AlignedBuffer shifts lazy,
  deferring the internal move to buffer refill boundaries (credit: @wtn)
• Added new publisher values for Cboe Titanium Cboe Global Indices Feed
• Added support for splitting files by year to TimeSplitter
• Improved Python DBNDecoder.decode() performance with pre-allocated output buffers
• Added DBNDecoder.write_and_decode() to combine write and decode in a single call
• Upgraded pyo3 to 0.28.3
  • Declared free-threaded Python support on the databento_dbn extension module. Not
    yet building free-threaded wheels in CI

0.55.0

Release notes

Enhancements

• Added new publisher values for OPRA MEMX MX2 Options and IEX Options
• Improved DbnFsm decode throughput on current-version data and AsIs workloads by
  caching whether the upgrade policy-version combination requires upgrading, skipping
  the per-record upgrade_record dispatch on the fast path (credit: @wtn)

Bug fixes

• Removed unsound Copy and Clone implementations for RecordRefMut

0.54.0

Release notes

Enhancements

• Added RecordBuf, an owned stack-allocated buffer for holding a DBN record of
  dynamic type. Complements RecordRef (borrowed, dynamic) and concrete types
  (owned, static). Supports get(), try_get(), set(), upgrade(), and
  cross-capacity PartialEq
• Added RecordRefMut for mutable non-owning references to DBN records
• Added RecordRef::to_owned() and RecordRefMut::to_owned() for converting
  borrowed records to an owned RecordBuf
• Added Hash, PartialEq, and Eq for RecordRef, including cross-type equality
  between RecordBuf and RecordRef
• Changed RecordHeader to be Copy

Release notes

Enhancements

• Added RecordBuf, an owned stack-allocated buffer for holding a DBN record of
  dynamic type. Complements RecordRef (borrowed, dynamic) and concrete types
  (owned, static). Supports get(), try_get(), set(), upgrade(), and
  cross-capacity PartialEq
• Added RecordRefMut for mutable non-owning references to DBN records
• Added RecordRef::to_owned() and RecordRefMut::to_owned() for converting
  borrowed records to an owned RecordBuf
• Added Hash, PartialEq, and Eq for RecordRef, including cross-type equality
  between RecordBuf and RecordRef
• Changed RecordHeader to be Copy

0.53.0

Release notes

Enhancements

• Made ts_out a permanent field on all Python record types, replacing the
  dynamic __dict__ attribute. ts_out returns an int (UNDEF_TIMESTAMP when
  not set)
• Removed __dict__ from all Python record classes, eliminating a separate
  per-instance allocation
• FlagSet setter methods are now const, making it easier to define constants, e.g.
  const FLAGS: FlagSet = FlagSet::empty().set_snapshot().set_last()

Breaking changes

• record.ts_out now returns UNDEF_TIMESTAMP instead of raising AttributeError
  when ts_out is not set. Code using hasattr(record, "ts_out") should check
  record.ts_out != UNDEF_TIMESTAMP instead
• bytes(record) now includes the 8-byte ts_out suffix when the record was
  created with ts_out. Code that separately appends ts_out bytes after
  bytes(record) should be updated to avoid double-writing
• Arbitrary attribute assignment on record instances is no longer supported

Bug fixes

• Fixed memory leak around ts_out assignment

Release notes

Enhancements

• Made ts_out a permanent field on all Python record types, replacing the
  dynamic __dict__ attribute. ts_out returns an int (UNDEF_TIMESTAMP when
  not set)
• Removed __dict__ from all Python record classes, eliminating a separate
  per-instance allocation
• FlagSet setter methods are now const, making it easier to define constants, e.g.
  const FLAGS: FlagSet = FlagSet::empty().set_snapshot().set_last()

Breaking changes

• record.ts_out now returns UNDEF_TIMESTAMP instead of raising AttributeError
  when ts_out is not set. Code using hasattr(record, "ts_out") should check
  record.ts_out != UNDEF_TIMESTAMP instead
• bytes(record) now includes the 8-byte ts_out suffix when the record was
  created with ts_out. Code that separately appends ts_out bytes after
  bytes(record) should be updated to avoid double-writing
• Arbitrary attribute assignment on record instances is no longer supported

Bug fixes

• Fixed memory leak around ts_out assignment

0.52.1

Release notes

Enhancements

• Added __index__ to all int-representable and char-backed Python enums, enabling
  use with int(), hex(), and as sequence indices

Bug fixes

• Fixed memory leak in Python bindings where every record object leaked ~64 bytes
  due to a pyo3 0.28 regression in #[pyclass(dict)] deallocation. Downgraded
  pyo3 to 0.27.2
• Removed unnecessary dict from BidAskPair and ConsolidatedBidAskPair Python
  classes
• Fixed Python type stubs to reflect that record_size() is a method, not a property
• Fixed Python type stubs for record fields to indicate which fields are writable
  (e.g. publisher_id, instrument_id, price, size) and added @setter stubs
  for enum fields (e.g. action, side)

0.52.0

Release notes

Enhancements

• Added SYMBOL_CSTR_LEN constant and versioned variants (SYMBOL_CSTR_LEN_V1,
  SYMBOL_CSTR_LEN_V2, SYMBOL_CSTR_LEN_V3) to Python, including in each versioned
  module (v1, v2, v3) as SYMBOL_CSTR_LEN
• Added v1, v2, and v3 submodule imports to databento_dbn.__init__ so they are
  accessible as attributes (e.g. databento_dbn.v1)
• Upgraded pyo3 version to 0.28.2

0.51.0

Release notes

Enhancements

• Added logic to set code when upgrading version 1 ErrorMsg to newer versions
• Upgraded async-compression dependency version to 0.4.41
• Upgraded time version to 0.3.47

Bug fixes

• Fixed miri issue with potential for unaligned data when decoding DBN through DbnFsm

0.50.0

Release notes

Enhancements

• Added SkippedRecordsAfterSlowReading to the ErrorCode enum for gateway errors
  originating from slow client catch-up

0.49.0

Release notes

Enhancements

• Added encode_records to EncodeRecord and AsyncEncodeRecord traits for more
  efficient batch encoding
• Added encode_record_refs to EncodeRecordRef and AsyncEncodeRecordRef traits with
  more efficient vectored I/O implementation for DBN
• Added support for decompressing Zstd in the Python DBNDecoder and new optional
  compression parameter

Breaking changes

• Moved encode_records method from the EncodeDbn trait to EncodeRecord

0.48.0

Release notes

Enhancements

• Added initial support for splitting DBN files:
  • Added new SplitEncoder that supports synchronous and asynchronous encoding
  • Added new Splitter trait that allows for extensible splitting of files while
    reusing the SplitEncoder boilerplate.
  • Added SchemaSplitter, SymbolSplitter, and TimeSplitter which allow for
    different methods of splitting DBN files
  • Added split support to the CLI. For example:
    dbn mbo.dbn --split-by week --output-pattern '{date}.json' --json
• Added new publisher for Blue Ocean ATS (OCEA_MEMOIR_OCEA)

Bug fixes

• Fixed issue where AsyncDynReader instances created from with_buffer() would only
  decode the first frame of multi-frame Zstandard files