v0.0.41

Performance

• (decode) decode StreamingDecoder read_to_end in place (#429)
• (hc) lazy-skip + separate dictMatchState beat C on small-window dict compress (#427)
• (encode) close small-window dict compress gap (hash-chain matchfinder + borrowed dict kernel) (#426)
• (huff0) skip FSE weight description for streams it cannot encode (#424)

Refactored

• purge reference-impl terminology + isolate C bindings in ffi-bench (#430)

v0.0.40

Performance

• (huff0) hoist bit-stream state into locals in the encode loop (#423)
• borrowed in-place over-window scan for dfast + row + btlazy2 (#422)
• (encode) cap HC/BT history mirror near the live window (#421)
• (encode) cap dfast history buffer near the live window (#420)
• (encode) cap row history buffer near the live window (#418)

v0.0.39

Performance

• (encode) cap row match-finder table to upstream hashLog (#417)
• (encode) single price arena + interleaved cache in the optimal parser (#415)

v0.0.38

Added

• (c-api) complete the stable ZSTDLIB_API surface + btopt/btultra perf (#413)

v0.0.37

Added

• (c-api) dictionary attach surface + estimates + fastCover; wasm prepared dictionary (#409)

v0.0.36

Performance

• dict decode peak + direct path + repcode gate + entropy table build (#403)

v0.0.35

0.0.35 - 2026-06-11

Added

• (c-api) advanced parameter API + streaming surface (compressStream2 / decompressStream) (#400)

Performance

• (encode) size the output reservation from the observed ratio (#398)
• (encode) allocation-free copy-mode dictionary restore (#397)

v0.0.34

Added

• dictionary CDict-equivalent (encoder + C ABI + wasm + CLI); L22 dict memory 8.2x->1.13x (#387)
• (c-api) C ABI core (cdylib + vendored headers + simple/context/error/frame/dict wrappers) (#386)
• content-checksum modes + npm bindings, RingBuffer wrapped inline match-copy (#385)

Performance

• (encode) zero per-frame allocations on the reused-compressor path (#395)
• (decode) hash each block while cache-hot on the direct path (#394)
• (encode) [breaking] cut dict-compress per-frame overhead; shrink wasm payload 18% (#393)
• (encode) monolithize dfast per-match helpers into the kernel (#391)
• (encode) 4-byte gate before HC chain common_prefix_len (#392)
• (encode) reshape dfast match-find toward donor structure (#390)
• (encode) close the level-4 dfast speed outlier (donor greedy double-fast) (#389)
• (decode) cut per-frame/block overhead (RingBuffer inline exec, FSE arrays, dict copy-on-write) (#381)
• (decode) cut HUF/FSE entropy-build overhead on the decode hot path (#377)

v0.0.33

Performance

• dict-attach across match finders + decode peak-alloc (#359)

v0.0.32

Added

• (decoding) resumable cold partial decode (resume_at via decode_blocks_partial) (#371)

Fixed

• (wasm) add floating version requirement to structured-zstd dep (#373)

Full changelog: v0.0.31...v0.0.32