NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
pub.dev
pub.dev
Adaptive arithmetic coding for efficient lossless compression of bytes in pure Dart.
Latest stable
1.0.5
Last release 5 months ago
03 Jun 2026
Too new to tell
only 2 release windows
Nearly every release is documented
notes for 5 of 5 stable releases
Nothing withdrawn
no release was ever pulled
7 months old
5 releases · first in 2026
Drag across the ruler to compare, or tap a release to open it.
Performance: optimized encode/decode for order-1, order-2 (and order-3) context models.
ContextModel and ContextState:
_LazyContextModel base class that allocates one Fenwick tree per
context lazily, on first use, instead of eagerly creating a tree for
every possible context.init/reset now cost is proportional to the contexts actually used
rather than to the full context space (size² for order-2, more for
order-3), eliminating millions of wasted operations per call.ContextModelOrder1/Order2/Order3 now extend _LazyContextModel and
flatten their context space into a single slot table; output is
byte-identical to the previous implementation.Fenwick:
freqOf) are O(1). This removes one of the two cumulative tree
walks per symbol in encode.findWithLow, which decodes a symbol and returns its lower
cumulative bound (sum(symbol - 1)) in a single descent, removing two
tree walks per symbol in decode.init now builds the all-ones tree directly in O(n) (each node i holds
i & -i) instead of O(n log n), the dominant cost when many contexts are
visited (e.g. high-order models on high-entropy data: ~4–5x faster).rescale and toFrequencyList now use the direct-frequency array
(O(n) instead of O(n log n)).benchmark/arithmetic_coder_benchmark.dart to track compression ratio
and encode/decode throughput across orders and datasets for future metrics.test/fenwick_test.dart covering the direct-frequency array,
freqOf, findWithLow, the O(n) init, rescale/rebuild/reset sync,
and the sum/cumulative invariants (including through many rescales).test/context_model_test.dart covering the factory, totalSize,
contextIndex mapping, lazy allocation, order-3 bucketing, and pool reuse
across reset/init.test/arithmetic_coder_test.dart with cached-model reuse across
sequential calls, deterministic re-encoding, and rescale-triggering inputs.Your coding agent can read these notes before it upgrades. Set up the MCP server →
One column per month.
Exposed buildContextModel method to allow customization of the context model creation.
ArithmeticCoder:
buildContextModel method to allow customization of the context model creation._buildModels to _buildContextModelCached and updated to use buildContextModel.decode method to use models.eof symbol dynamically instead of hardcoded 256.ContextModel and ContextState:
ContextStateOrder3 with three previous symbols.ContextModelOrder3 with a 3D Fenwick tree structure and a shrink factor to reduce memory usage.order getter to ContextState and ContextModel to indicate the model order.maxContextOrder constant (value 3) to ContextModel.ContextModel to support order 3.order getter.order getter.ContextModelOrder3:
Updated _tree field initialization to use typed lists (Uint8List, Uint16List, Uint32List, Uint64List) based on the bit length of maxTotal for memory e
Fenwick - reduce memory usage:
_tree field initialization to use typed lists (Uint8List, Uint16List, Uint32List, Uint64List) based on the bit length of maxTotal for memory efficiency._buildIntList to create appropriate typed list for frequency storage.computeMaxTotal to return (1 << (precision ~/ 2)) - 1 instead of 1 << (precision ~/ 2) to correctly compute maximum total frequency._tree initialization from List.filled to use _buildIntList in constructor.Added context modeling support to ArithmeticCoder:
Added context modeling support to ArithmeticCoder:
order parameter to specify finite-order Markov models (0, 1, or 2).ContextModel abstraction with adaptive Fenwick trees per context.toString override to ArithmeticCoder showing order and model size.ContextModel:
ContextModel with factory constructor for orders 0, 1, and 2.ContextModelOrder0, ContextModelOrder1, and ContextModelOrder2 with corresponding ContextState classes.arithmetic_coder.dart:
ContextModel for symbol frequency management.example/arithmetic_coder_example.dart:
ArithmeticCoder(order: 2)._listEquals for byte list comparison.test/arithmetic_coder_test.dart:
ArithmeticCoder instantiations inside tests.pubspec.yaml:
1.0.1.## 1.0.0 - Initial version.