omnizip 0.3.22 → 0.3.23

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 11b7628df01cd834c493cd96ba094783f65997f4c306356b653bd6c7dec27910
-  data.tar.gz: 40853fd564d6fa1286f995f9b04d62e916aa26b56a7d395e067fe32d706e5730
+  metadata.gz: 61cf26f9e324600f286b236ab80e02d720059f32bca2fa3c71640ba3abc4355a
+  data.tar.gz: fa189e081e402ecb4d084490e7874402faacf231a296cd009a15f218fcd9453e
 SHA512:
-  metadata.gz: cb7ed5d3b46761cccd18a471ad1794aad849981f79c822ad7c8c6670cac260b1af28e7cfd368b894afc41613ca7070fb04a5a4bdef10ffff28f662f2e04add1a
-  data.tar.gz: f9396815f4ed33c8813812bed12a0f8b54923eaea1a22d62809d15c826b87e0b32aad00c2ea4cc223cffcb36fab6fc97fa1f8401d68a2fd4c31e1937f485e5d4
+  metadata.gz: 308660d6095a909d72c528bdfb5e828e76d320172b979a53ef7401aa2c1c12a560a0afd57f49991eb5190f87deca1c15cd798e8abae89bd3341ed3bb43c39b3d
+  data.tar.gz: 29ac9066539f2d8cef3973a2eb9cfe15a549a5efead04a1f5c88f28587fb6a2d23ac28c401c1c4187f5c226bfe4b531e3565523400dcd183adc83c842cd2d5dc

data/CHANGELOG.md

@@ -7,6 +7,46 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.3.23] - 2026-08-26
+
+### Changed
+- Zstd encoder hot paths use allocation-free integer reads instead of
+  `String#byteslice` + `unpack1` comparisons (profiling showed GC at
+  52% of encoder time): default-level compression is ~3.4x faster and
+  level 22 ~1.9x on the benchmark corpus, with byte-identical output.
+
+## [0.3.22] - 2026-08-25
+
+### Added
+- Zstandard dictionary compression (`Dictionary.from_raw` /
+  `serialize` / `deserialize`, `compress_with_dict`,
+  `decompress_with_dict`): the dictionary content primes the match
+  finder as shared history and the frame header carries the
+  Dictionary_ID, which the decoder verifies (Phase-1 scope, as in the
+  Rust reference; entropy-table preloading is future work).
+
+## [0.3.21] - 2026-08-25
+
+### Fixed
+- 7-Zip encoder dictionary is now capped at the size announced in the
+  coder properties (prevents matches reading outside the decoder
+  window for inputs above the announced size).
+
+### Changed
+- Zstd lazy levels (6+) gained the rep0 fast-path and backward
+  extension: levels 6-12 improve 0.180 -> 0.154 and levels 19-22
+  0.146 -> 0.126 on the benchmark corpus; the level scale is now
+  monotonic.
+- Zstd adaptive block splitting: heterogeneous chunks of 32 KiB or
+  more split into 16 KiB sub-blocks so entropy tables fit each
+  content regime.
+
+## [0.3.20] - 2026-08-25
+
+### Removed
+- Dead `SevenZipLZMA2`/`XZLZMA2` algorithm wrappers (never
+  autoloaded; one referenced a nonexistent decoder constant).
+
 ## [0.3.19] - 2026-08-25
 
 ### Fixed

data/lib/omnizip/algorithms/zstandard/match_finder.rb

@@ -131,11 +131,29 @@ module Omnizip
             min_match: MIN_MATCH_ECONOMICAL }
         end
 
-        def hash4(src, pos, h_bits)
-          v = src.getbyte(pos) |
+        # Allocation-free little-endian reads; the hot loops call
+        # these millions of times, and String#byteslice + unpack1
+        # showed up as half the encoder's time via GC pressure.
+        def read4(src, pos)
+          src.getbyte(pos) |
             (src.getbyte(pos + 1) << 8) |
             (src.getbyte(pos + 2) << 16) |
             (src.getbyte(pos + 3) << 24)
+        end
+
+        def read8(src, pos)
+          src.getbyte(pos) |
+            (src.getbyte(pos + 1) << 8) |
+            (src.getbyte(pos + 2) << 16) |
+            (src.getbyte(pos + 3) << 24) |
+            (src.getbyte(pos + 4) << 32) |
+            (src.getbyte(pos + 5) << 40) |
+            (src.getbyte(pos + 6) << 48) |
+            (src.getbyte(pos + 7) << 56)
+        end
+
+        def hash4(src, pos, h_bits)
+          v = read4(src, pos)
           # 32-bit wrapping multiply (C/uint32_t), then the high bits.
           ((v * PRIME4_BYTES) & 0xFFFFFFFF) >> (32 - h_bits)
         end
@@ -148,8 +166,8 @@ module Omnizip
           b_size = b.bytesize
           while len + 8 <= limit && a_pos + len + 8 <= a_size &&
               b_pos + len + 8 <= b_size
-            wa = a.byteslice(a_pos + len, 8).unpack1("Q<")
-            wb = b.byteslice(b_pos + len, 8).unpack1("Q<")
+            wa = read8(a, a_pos + len)
+            wb = read8(b, b_pos + len)
             if wa == wb
               len += 8
             else
@@ -289,8 +307,7 @@ module Omnizip
         # rubocop:disable-next Metrics/AbcSize
         def rep0_match(src, ip, rep0, min_match, limit, anchor)
           return nil if rep0 <= 0 || ip <= rep0
-          return nil if src.byteslice(ip, MIN_MATCH) !=
-            src.byteslice(ip - rep0, MIN_MATCH)
+          return nil if read4(src, ip) != read4(src, ip - rep0)
 
           m_len = MIN_MATCH + count_match(
             src, ip + MIN_MATCH, src, ip + MIN_MATCH - rep0,
@@ -346,7 +363,7 @@ module Omnizip
             break if dist >= max_distance
             break if candidate + MIN_MATCH > size
 
-            if src.byteslice(ip, MIN_MATCH) == src.byteslice(candidate, MIN_MATCH)
+            if read4(src, ip) == read4(src, candidate)
               m_len = MIN_MATCH + count_match(src, ip + MIN_MATCH,
                                               src, candidate + MIN_MATCH,
                                               [max_extend - MIN_MATCH, 0].max)
@@ -397,7 +414,7 @@ module Omnizip
           if candidate.positive? && candidate < ip
             dist = ip - candidate
             if dist < max_distance && candidate + MIN_MATCH <= size &&
-                src.byteslice(ip, MIN_MATCH) == src.byteslice(candidate, MIN_MATCH)
+                read4(src, ip) == read4(src, candidate)
               best_len = MIN_MATCH + count_match(
                 src, ip + MIN_MATCH, src, candidate + MIN_MATCH,
                 [limit + MIN_MATCH_ECONOMICAL - ip - MIN_MATCH, 0].max

data/lib/omnizip/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Omnizip
-  VERSION = "0.3.22"
+  VERSION = "0.3.23"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: omnizip
 version: !ruby/object:Gem::Version
-  version: 0.3.22
+  version: 0.3.23
 platform: ruby
 authors:
 - Ribose Inc.