omnizip 0.4.3 → 0.4.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 34bab2534b75e9ca958c00130c6b5962dbd5422bd6e0f4683c8e756415d9961b
-  data.tar.gz: 9089d88fbe8ce8bda17fecd2a21454db2daa23c86e930b69cf8bb86a495b1ce0
+  metadata.gz: 8cf4c62c7304ef792b48b96486685707c68010f0cc6529b6a7504454b94ea025
+  data.tar.gz: 7c0d5e62fc21a99ccc85520ee0cd3ea32afaae39e76f10826dd56697ffa8376b
 SHA512:
-  metadata.gz: d92441da2c757c65e1e5865abd446cd2a296adafcdf56d5dabfc00f6c792fd6f9b8f3e51dda0231ce5d8829eaf4b81662cfdadc13cdbc9495c73d088968c6c81
-  data.tar.gz: 30591da9a2e0af6922a49596133f98e5f05bf30e927b698e4b413c1f9f0d1907ab6f0f9a0cc32eff385772b7445aff36e271066690c85e3a134dd040e8185591
+  metadata.gz: 539124e9189208423daacdac7cad6735f4a1069358a6ee7e39bb09155524589a2b1a01a6b1c41af6b650648dee327b2c4d36e71ccbbba5ff5e4453168f6ec9ef
+  data.tar.gz: 02ae0fdb9844f006e5289d8bd351b4ac023e8fafd704bd6fc96b6a0489bc27c96d3e7d3e32ef06011a8123b37573d877191e93cdbaa7c0f6e63e139e8485c09e

data/lib/omnizip/algorithms/deflate.rb

@@ -68,8 +68,13 @@ module Omnizip
       # @return [void]
       def compress(input_stream, output_stream, options = nil)
         input_data = input_stream.read
-        encoder = Encoder.new(output_stream, build_encoder_options(options))
-        encoder.encode_stream(input_data)
+        level = build_encoder_options(options)[:level] || 6
+        output_stream.write(
+          Backends.compress("zlib", input_data, level) do
+            encoder = Encoder.new(output_stream, build_encoder_options(options))
+            encoder.encode_stream(input_data)
+          end,
+        )
       end
 
       # Decompress Deflate-compressed data

data/lib/omnizip/algorithms/deflate64.rb

@@ -42,21 +42,22 @@ module Omnizip
       # @param options [Hash] Compression options
       # @option options [Integer] :level Compression level (1-9)
       def compress(input, output, options = {})
-        level = options[:level] || Zlib::DEFAULT_COMPRESSION
-
+        raw_level = options[:level] || Zlib::DEFAULT_COMPRESSION
         data = input.read
         return if data.nil? || data.empty?
 
-        # Use Zlib::Deflate with maximum window size
-        deflater = Zlib::Deflate.new(
-          level,
-          Zlib::MAX_WBITS, # Maximum window size
-          Zlib::MAX_MEM_LEVEL,
-        )
-
-        compressed = deflater.deflate(data, Zlib::FINISH)
-        deflater.close
-
+        # Zlib::DEFAULT_COMPRESSION (-1) maps to the FFI's 6.
+        level = raw_level.negative? ? 6 : raw_level.clamp(0, 9)
+        compressed = Backends.compress("zlib", data, level) do
+          deflater = Zlib::Deflate.new(
+            raw_level,
+            Zlib::MAX_WBITS,
+            Zlib::MAX_MEM_LEVEL,
+          )
+          result = deflater.deflate(data, Zlib::FINISH)
+          deflater.close
+          result
+        end
         output.write(compressed)
       end
 

data/lib/omnizip/algorithms/zstandard.rb

@@ -86,8 +86,13 @@ module Omnizip
       # @return [void]
       def compress(input_stream, output_stream, options = nil)
         input_data = input_stream.read
-        encoder = Encoder.new(output_stream, build_encoder_options(options))
-        encoder.encode_stream(input_data)
+        level = build_encoder_options(options)[:level] || 3
+        output_stream.write(
+          Backends.compress("zstd", input_data, level) do
+            encoder = Encoder.new(output_stream, build_encoder_options(options))
+            encoder.encode_stream(input_data)
+          end,
+        )
       end
 
       # Decompress Zstandard-compressed data

data/lib/omnizip/backends.rb

@@ -27,19 +27,23 @@ module Omnizip
   #     over frame stability.
   module Backends
     # Codecs whose Rust encoder is BYTE-IDENTICAL to the pure-Ruby
-    # encoder (empirical, differential-corpus-proven). 2026-09-19
-    # probe: bzip2 L6 on a 50 KB corpus => Ruby 906 B vs Rust 874 B —
-    # DIFFERENT valid streams. No codec qualifies; encode stays
-    # pure-Ruby under :auto so downstream content addressing never
-    # sees tier-dependent bytes. Force with OMNIZIP_BACKEND=rust.
-    RUST_ENCODE_IDENTICAL = {}.freeze
+    # 2026-09-20 policy flip (owner directive): RUST IS THE AUTHORITY
+    # — its encoders/decoders are the ports validated against the
+    # reference C/C++ implementations (bzip2, xz/liblzma, zstd,
+    # libdeflate), and they are determinism-contracted (same input +
+    # level => same bytes on every machine). :auto therefore routes
+    # BOTH directions through Rust whenever the cdylib loads; encode
+    # falls back to the Ruby core on any Rust error, so :auto is
+    # never worse than pure Ruby. The pure-Ruby cores remain the
+    # portable fallback and the OMNIZIP_BACKEND=ruby escape hatch.
+    RUST_AUTHORITATIVE = true
 
     # usize::MAX — the FFI's "caller does not know the plaintext
     # size" sentinel (streaming decode contract).
     UNKNOWN_LENGTH = 0xFFFF_FFFF_FFFF_FFFF
 
     class << self
-      def for(codec, direction)
+      def for(codec, _direction)
         mode = ENV.fetch("OMNIZIP_BACKEND", "auto")
 
         return backend_rust if mode == "rust"
@@ -48,17 +52,26 @@ module Omnizip
         return backend_ruby unless Implementations::Rust::Library.available?
         return backend_ruby unless Implementations::Rust::Library::CODECS.include?(codec)
 
-        direction == :encode && !RUST_ENCODE_IDENTICAL.fetch(codec, false) ? backend_ruby : backend_rust
+        backend_rust
       end
 
       # Compress through the tier-selected backend. The block is
       # the pure-Ruby core — evaluated ONLY on the Ruby path, so
-      # the accelerated path never pays for it.
+      # the accelerated path never pays for it. A Rust encode error
+      # falls back to the Ruby core (auto is never worse than pure
+      # Ruby); only the forced `rust` mode propagates.
       def compress(codec, data, level, &ruby_core)
         backend = Backends.for(codec, :encode)
         return yield if backend == RubyBackend
 
-        backend.compress(codec, data, level)
+        begin
+          backend.compress(codec, data, level)
+        rescue Implementations::Rust::Error => e
+          raise if ENV.fetch("OMNIZIP_BACKEND", "auto") == "rust"
+
+          warn "omnizip: rust #{codec} encode failed (#{e.message}); using ruby" if ENV["OMNIZIP_BACKEND_DEBUG"]
+          yield
+        end
       end
 
       # Decompress through the tier-selected backend (output-

data/lib/omnizip/implementations/rust/library.rb

@@ -16,9 +16,9 @@ module Omnizip
       # built; compress/decompress are pure C calls).
       class Library
         # Codecs the cdylib dispatches by name.
-        # Names the cdylib dispatches: the codec-level acceleration surface
-        # (decode tier by default; encode only under forced rust mode —
-        # RUST_ENCODE_IDENTICAL gates the auto tier).
+        # Names the cdylib dispatches: the codec-level acceleration
+        # surface (both directions by default — Rust is the
+        # authority; Ruby is the fallback).
         CODECS = %w[
           bzip2 zstd lzma xz lzma-alone lzip
           deflate deflate64 zlib gzip

data/lib/omnizip/implementations/rust.rb

@@ -13,11 +13,11 @@ module Omnizip
     #   1. ENV['OMNIZIP_FFI_DYLIB'] (explicit path)
     #   2. ../target/release/ inside a sibling omnizip-rs checkout
     #
-    # Tier policy (see Omnizip::Backends): decode is always
-    # output-invariant, so `auto` uses Rust for decode whenever
-    # available; encode uses Rust only when byte-identity with the
-    # pure-Ruby encoder has been verified for that codec
-    # (RUST_ENCODE_IDENTICAL), keeping content addressing stable.
+    # Tier policy (see Omnizip::Backends): Rust is the authority —
+    # validated against the reference C/C++ implementations and
+    # determinism-contracted — so `auto` routes BOTH directions
+    # through Rust whenever the library loads, with Ruby-core
+    # fallback on any Rust error.
     module Rust
       autoload :Library, "omnizip/implementations/rust/library"
     end

data/lib/omnizip/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Omnizip
-  VERSION = "0.4.3"
+  VERSION = "0.4.4"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: omnizip
 version: !ruby/object:Gem::Version
-  version: 0.4.3
+  version: 0.4.4
 platform: ruby
 authors:
 - Ribose Inc.