omnizip 0.4.5 → 0.4.7

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5a55c37209e986f7e4499e8a485163728e9c8f0dff506e1aa852672a9585653f
-  data.tar.gz: ba0b5437026feed4c1fe4d0ab0146f868386ebe2284ec7b08b45c7260930d582
+  metadata.gz: f958a6bde5f52729231c66fcc680ecaf30bd3bb3e3a04f59bd843a63f5d664ec
+  data.tar.gz: c35f940965b7776bbab14f41258917bf4ae444899ec70d4c0713a36a12d5c05b
 SHA512:
-  metadata.gz: 13d25032be26ccd096718846ac74126fd14ab3e2729d1e9ed366a3c431815b69c0817e33092f214eea4ed041e41e2eebe070ac92862bd54c70b0379017281637
-  data.tar.gz: 8e68e9a4a0d7f489be9956baf6f2d18bb82d5cc0db123adaf689ab73aefca1dcb44c0172a8e774c0cbcfbcca2927c9f40b62da184678588e084642b3b86cd7eb
+  metadata.gz: 6e219d9d7052b667c112709d0ebaa4a4e3b6925e374694f1fd646114e6ab7d2157e255a07a8e05b9a42a9234b4e61841617ce4682b6e79727ee6a4381a4e2b35
+  data.tar.gz: f6921e7de8606e19dd28e62d4a00a8bec55b2f0bc6f4a4478805df6299f88b912aa6b8ea56fbf774f6df42a7f29cc707d709c4dc250c702a5d3b4ca91d785601

data/CHANGELOG.md

@@ -7,6 +7,25 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+- Prebuilt-binary distribution (parsanol model): platform gems
+  (x86_64/aarch64/arm Linux glibc+musl, arm64/x86_64 macOS,
+  x64/aarch64-mingw-ucrt, x64-mingw32) now vendor the prebuilt
+  `omnizip-ffi` cdylib — `gem install omnizip` gives Rust
+  acceleration with zero compilation on those platforms. The plain
+  `ruby` gem and every platform without a binary (JRuby,
+  TruffleRuby, anything else) stay fully functional on the pure-Ruby
+  core. The cdylibs are cross-built by omnizip-rs's release-binary
+  workflow and attached to each `omnizip-ffi-v*` release;
+  `rake platform_gems:build` packages them and `gem-build.yml`
+  smoke-tests each installed gem (tier active + pure-Ruby path with
+  `OMNIZIP_NO_RUST=1`).
+- `OMNIZIP_NO_RUST=1` kill switch forces the pure-Ruby core
+  everywhere; `Library.resolve_path` (public now) prefers the
+  vendored cdylib over dev checkouts; `Library.dylib_version`
+  reports the Rust release a loaded cdylib was built from
+  (`ozip_version`, new in omnizip-rs 0.21.109).
+
 ### Fixed
 - PPMd works for the first time: the pure-Ruby ppmd7 core cannot
   decode its own output (root-only decoder, 100-symbol cap — proven)
@@ -25,6 +44,17 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   route decode through the Rust backend when it loads, each pinned by
   a cross-implementation differential. A Rust failure falls back to
   the Ruby core — auto mode is never worse than pure Ruby.
+- Archive-level acceleration: `Implementations::Rust::Archive` binds
+  the cdylib's whole-archive surface (open, list entries, read entry)
+  over every multi-entry format — zip, tar, cpio, 7z, rar3/4/5, iso,
+  rpm, xar, ole, encrypted archives included. `ZipHandler#list` and
+  `#read_entry` route through it first (auto mode; any Rust error or
+  a missing cdylib falls back to the pure-Ruby path). The tier's
+  600-mutant fuzz gate also caught a Rust decompression bomb
+  (truncated deflate streams decoded an endless phantom bit-stream;
+  fixed upstream in omnizip 0.21.108).
+- `Library#bind` / `Library#take_buffer` seams the archive surface
+  rides (Fiddle symbol binding + copy-out-and-free).
 
 ### Fixed
 - RAR recovery-record repair fails loudly instead of pretending:

data/README.adoc

@@ -28,6 +28,28 @@ It provides both a command-line interface and a programmatic Ruby API, making it
 
 See link:readme-docs/compression-algorithms.adoc[Compression Algorithms Guide] for detailed information.
 
+==== Rust Acceleration (prebuilt, zero compilation)
+
+Where the platform gem exists, `gem install omnizip` ships a prebuilt
+`omnizip-ffi` cdylib and the codec + archive tiers route through it
+automatically — no Rust toolchain, no compilation:
+
+* x86_64-linux, aarch64-linux, arm-linux (glibc and musl)
+* arm64-darwin, x86_64-darwin
+* x64-mingw-ucrt, aarch64-mingw-ucrt, x64-mingw32
+
+Everywhere else — including JRuby/TruffleRuby and any platform without
+a binary — the gem stays fully functional on its pure-Ruby core (the
+acceleration is strictly optional). Three escape hatches:
+
+* `OMNIZIP_NO_RUST=1` — force the pure-Ruby core everywhere.
+* `OMNIZIP_FFI_DYLIB=/path/to/libomnizip_ffi` — load a specific build.
+* `rake rust:build` — build and vendor the cdylib yourself from a
+  checkout (developers).
+
+Any Rust-side failure also falls back to the Ruby core mid-operation,
+so `auto` mode is never worse than pure Ruby.
+
 ==== SDK-Compatible LZMA Mode (v0.2.0)
 
 Omnizip now provides SDK-compatible LZMA encoding and decoding for full XZ/LZMA tool compatibility.

data/lib/omnizip/archive_handlers/zip_handler.rb

@@ -48,6 +48,14 @@ module Omnizip
       end
 
       def list(path, details: false, **_)
+        # Whole-archive tier (names-only path): the Rust archive
+        # surface lists any supported format; the Ruby reader stays
+        # the fallback and the details path's source.
+        unless details
+          tier_names = Omnizip::Backends.archive_entry_names(path)
+          return tier_names if tier_names
+        end
+
         reader = Formats::Zip::Reader.new(path)
         return reader.entries.map(&:filename) unless details
 
@@ -65,6 +73,9 @@ module Omnizip
       end
 
       def read_entry(path, entry_name)
+        tier = Omnizip::Backends.archive_read_entry(path, entry_name)
+        return tier unless tier.nil?
+
         Formats::Zip::Reader.new(path).read_entry(entry_name)
       end
 

data/lib/omnizip/backends.rb

@@ -94,6 +94,26 @@ module Omnizip
         end
       end
 
+      # Whole-archive tier helpers: nil means "Rust unavailable or
+      # entry not found — caller falls back to the Ruby reader".
+      def archive_entry_names(path)
+        Omnizip::Implementations::Rust::Archive.open(File.binread(path)) { |a| return a.entry_names }
+      rescue Omnizip::Implementations::Rust::Archive::LibraryMissing, Omnizip::Implementations::Rust::Error, StandardError
+        nil
+      end
+
+      # rubocop:disable-next Lint/ReturnInVoidContext
+      def archive_read_entry(path, entry_name)
+        Omnizip::Implementations::Rust::Archive.open(File.binread(path)) do |a|
+          idx = a.entry_names.index(entry_name)
+          return nil if idx.nil?
+
+          return a.read_entry(idx)
+        end
+      rescue Omnizip::Implementations::Rust::Archive::LibraryMissing, Omnizip::Implementations::Rust::Error, StandardError
+        nil
+      end
+
       private
 
       def backend_rust

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

@@ -0,0 +1,93 @@
+# frozen_string_literal: true
+
+module Omnizip
+  module Implementations
+    module Rust
+      # Whole-archive acceleration over the cdylib: open any supported
+      # format (auto-detected), list entries, read entry bytes. This is
+      # the archive-level tier surface — the Ruby core remains the
+      # fallback (the cdylib unavailable ⇒ these raise LibraryMissing
+      # and callers fall back).
+      class Archive
+        class LibraryMissing < StandardError; end
+
+        # Built lazily inside #initialize: the Fiddle::TYPE_* constants
+        # only resolve after a successful require, and Ruby 4.0 moved
+        # fiddle out of the default gems (same reason
+        # Library.function_table is a method, not a constant).
+        def initialize(data, password: nil)
+          lib = Library.instance
+          raise LibraryMissing, "omnizip-ffi cdylib not loaded" unless lib
+
+          @lib = lib
+          funcs = function_table
+          @funcs = funcs.each_with_object({}) do |(key, (name, args, ret)), h|
+            h[key] = lib.bind(name, args, ret)
+          end
+
+          pw = password&.to_s
+          pw_ptr = pw && !pw.empty? ? Fiddle::Pointer[pw] : nil
+          @handle = @funcs[:arch_open].call(Fiddle::Pointer[data], data.bytesize, pw_ptr)
+          return unless @handle.null?
+
+          raise Error, @lib.last_error
+        end
+
+        def entry_count
+          @funcs[:arch_count].call(@handle)
+        end
+
+        def entry_names
+          (0...entry_count).map do |i|
+            ptr = @funcs[:arch_entry_name].call(@handle, i)
+            next nil if ptr.null?
+
+            ptr.to_s
+          end
+        end
+
+        def entry_size(index)
+          @funcs[:arch_entry_size].call(@handle, index)
+        end
+
+        # Read one entry's uncompressed bytes.
+        #
+        # @return [String] binary
+        def read_entry(index)
+          out_len = [0].pack("Q")
+          ptr = @funcs[:arch_read_entry].call(@handle, index, out_len)
+          raise Error, @lib.last_error if ptr.null?
+
+          @lib.take_buffer(ptr, out_len.unpack1("Q"))
+        end
+
+        def close
+          return unless @handle
+
+          @funcs[:arch_close].call(@handle)
+          @handle = nil
+        end
+
+        def self.open(data, password: nil)
+          archive = new(data, password: password)
+          yield archive
+        ensure
+          archive&.close
+        end
+
+        private
+
+        def function_table
+          {
+            arch_open: ["ozip_arch_open", %i[voidp size_t voidp], Fiddle::TYPE_VOIDP],
+            arch_count: ["ozip_arch_count", [:voidp], Fiddle::TYPE_SIZE_T],
+            arch_entry_name: ["ozip_arch_entry_name", %i[voidp size_t], Fiddle::TYPE_VOIDP],
+            arch_entry_size: ["ozip_arch_entry_size", %i[voidp size_t], Fiddle::TYPE_VOIDP],
+            arch_read_entry: ["ozip_arch_read_entry", %i[voidp size_t voidp], Fiddle::TYPE_VOIDP],
+            arch_close: ["ozip_arch_close", [:voidp], Fiddle::TYPE_VOID],
+          }
+        end
+      end
+    end
+  end
+end

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

@@ -76,30 +76,47 @@ module Omnizip
             }
           end
 
-          private
-
-          def open
-            path = resolve_path
-            return nil if path.nil?
-
-            handle = Fiddle.dlopen(path.to_s)
-            funcs = function_table.each_with_object({}) do |(key, (name, args, ret)), h|
-              h[key] = Fiddle::Function.new(handle[name], args, ret)
-            end
-            new(funcs)
-          rescue StandardError => e # includes Fiddle::DLError
-            warn "omnizip: rust backend unavailable (#{e.message}); using pure Ruby" if ENV["OMNIZIP_BACKEND"] == "rust"
+          # The Rust release the loaded cdylib was built from (e.g.
+          # "0.21.108"), read through the ozip_version symbol — the
+          # platform-gem smoke gate asserts it matches. nil when the
+          # library is absent or predates the symbol.
+          def dylib_version
+            handle = instance&.instance_handle
+            return nil if handle.nil?
+
+            func = Fiddle::Function.new(handle["ozip_version"], [], Fiddle::TYPE_VOIDP)
+            ptr = func.call
+            ptr.null? ? nil : ptr.to_s
+          rescue StandardError
             nil
           end
 
-          def resolve_path
+          # Resolve the cdylib path. Precedence:
+          #
+          #   1. OMNIZIP_NO_RUST=1 — hard kill switch (always nil, so
+          #      the pure-Ruby core handles everything; also the smoke
+          #      gate for "works without Rust").
+          #   2. OMNIZIP_FFI_DYLIB — explicit path (developer override).
+          #   3. vendor/ — the prebuilt cdylib. Platform gems ship it
+          #      here and `rake rust:build` writes here; rubygems
+          #      already matched the platform at install time, so no
+          #      runtime architecture probing is needed and the wrong
+          #      arch cannot be picked.
+          #   4. ext/ (legacy location).
+          #   5. A sibling omnizip-rs checkout.
+          #
+          # `dirs` overrides 3-5 (spec seam: staged candidate paths).
+          def resolve_path(dirs = nil)
+            return nil if ENV.fetch("OMNIZIP_NO_RUST", nil) == "1"
+
             explicit = ENV.fetch("OMNIZIP_FFI_DYLIB", nil)
             return Pathname.new(explicit) if explicit && File.file?(explicit)
 
-            candidates = [
+            candidates = dirs || [
+              Pathname.new(__dir__).join("../../../../vendor"),
               Pathname.new(__dir__).join("../../../../ext/libomnizip_ffi"),
-              sibling_checkout,
-            ].compact.flatten
+              *sibling_checkout,
+            ]
 
             %w[.dylib .so .dll].each do |ext|
               candidates.each do |dir|
@@ -110,6 +127,22 @@ module Omnizip
             nil
           end
 
+          private
+
+          def open
+            path = resolve_path
+            return nil if path.nil?
+
+            handle = Fiddle.dlopen(path.to_s)
+            funcs = function_table.each_with_object({}) do |(key, (name, args, ret)), h|
+              h[key] = Fiddle::Function.new(handle[name], args, ret)
+            end
+            new(funcs, handle)
+          rescue StandardError => e # includes Fiddle::DLError
+            warn "omnizip: rust backend unavailable (#{e.message}); using pure Ruby" if ENV["OMNIZIP_BACKEND"] == "rust"
+            nil
+          end
+
           # A sibling omnizip-rs checkout: ../omnizip-rs relative to
           # this gem's repository root.
           def sibling_checkout
@@ -120,8 +153,9 @@ module Omnizip
           end
         end
 
-        def initialize(functions)
+        def initialize(functions, handle = nil)
           @functions = functions.freeze
+          @handle = handle
         end
 
         def compress(codec, data, level)
@@ -136,11 +170,31 @@ module Omnizip
           take(ptr, out_len.unpack1("Q"))
         end
 
+        # Bind one extra C symbol from the loaded cdylib (the archive
+        # surface uses this; Fiddle::Function is immutable once built).
+        def bind(symbol, args, ret)
+          handle = instance_handle
+          raise Error, "cdylib not loaded" if handle.nil?
+
+          Fiddle::Function.new(handle[symbol.to_s], args, ret)
+        end
+
         def last_error
           ptr = @functions[:last_error].call
           ptr.null? ? "unknown error" : ptr.to_s
         end
 
+        # The raw Fiddle::Handle (for extra symbol binds).
+        def instance_handle
+          @handle
+        end
+
+        # Copy out + free one returned buffer (public seam for the
+        # archive surface).
+        def take_buffer(ptr, len)
+          take(ptr, len)
+        end
+
         private
 
         def call(name, codec, input, input_len, a, out_len)

data/lib/omnizip/implementations/rust.rb

@@ -20,6 +20,7 @@ module Omnizip
     # fallback on any Rust error.
     module Rust
       autoload :Library, "omnizip/implementations/rust/library"
+      autoload :Archive, "omnizip/implementations/rust/archive"
     end
   end
 end

data/lib/omnizip/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Omnizip
-  VERSION = "0.4.5"
+  VERSION = "0.4.7"
 end

data/lib/tasks/platform_gems.rake

@@ -0,0 +1,107 @@
+# frozen_string_literal: true
+
+# Build per-platform gems that vendor a prebuilt omnizip-ffi cdylib,
+# so `gem install omnizip` gives Rust acceleration with zero
+# compilation (parsanol distribution model, adopted 2026-09-21).
+#
+#   # after downloading the release-binary artifacts:
+#   gh release download omnizip-ffi-v0.21.108 -R omnizip/omnizip-rs \
+#     -p 'libomnizip_ffi-*.tar.gz' -D artifacts
+#   bundle exec rake platform_gems:build DIR=artifacts
+#
+# Each target tarball maps to one rubygems platform; the built gems
+# land in pkg/. The plain `ruby` platform gem keeps shipping without
+# a binary (pure-Ruby core; OMNIZIP_NO_RUST=1 disables the tier).
+
+# rust target → rubygems platform (keep in sync with
+# omnizip-rs .github/workflows/release-binary.yml)
+PLATFORM_GEM_TARGETS = {
+  "x86_64-unknown-linux-gnu" => "x86_64-linux",
+  "aarch64-unknown-linux-gnu" => "aarch64-linux",
+  "arm-unknown-linux-gnueabihf" => "arm-linux",
+  "x86_64-unknown-linux-musl" => "x86_64-linux-musl",
+  "aarch64-unknown-linux-musl" => "aarch64-linux-musl",
+  "arm-unknown-linux-musleabihf" => "arm-linux-musl",
+  "aarch64-apple-darwin" => "arm64-darwin",
+  "x86_64-apple-darwin" => "x86_64-darwin",
+  "x86_64-pc-windows-msvc" => "x64-mingw-ucrt",
+  "aarch64-pc-windows-msvc" => "aarch64-mingw-ucrt",
+  "x86_64-pc-windows-gnu" => "x64-mingw32",
+}.freeze
+
+namespace :platform_gems do
+  desc "Build one platform gem: TARGET=<rust target> DYLIB=<cdylib path>"
+  task :one do
+    target = ENV.fetch("TARGET") { abort "TARGET=<rust target> required" }
+    dylib = ENV.fetch("DYLIB") { abort "DYLIB=<cdylib path> required" }
+    platform = PLATFORM_GEM_TARGETS[target] or abort "unknown target #{target} (see lib/tasks/platform_gems.rake)"
+    out = build_platform_gem(platform, dylib)
+    puts "built: #{out}"
+  end
+
+  desc "Build platform gems for every tarball in DIR=libomnizip_ffi-<target>.tar.gz"
+  task :build do
+    dir = ENV.fetch("DIR") { abort "DIR=<artifacts dir> required" }
+    Dir.glob(File.join(dir, "libomnizip_ffi-*.tar.gz")).each do |tarball|
+      target = File.basename(tarball, ".tar.gz").delete_prefix("libomnizip_ffi-")
+      platform = PLATFORM_GEM_TARGETS[target] or abort "no rubygems platform mapping for #{target}"
+      require "tmpdir"
+      Dir.mktmpdir do |tmp|
+        binary = extract_binary(tarball, tmp)
+        out = build_platform_gem(platform, binary)
+        puts "built: #{out} (#{File.size(out)} bytes)"
+      end
+    end
+  end
+
+  def extract_binary(tarball, tmp)
+    system("tar", "-xzf", tarball, "-C", tmp) or abort "tar failed for #{tarball}"
+    Dir.glob(File.join(tmp, "*")).first or abort "empty tarball #{tarball}"
+  end
+
+  # Stage the tracked gem files plus the vendored cdylib, flip the
+  # gemspec platform, and build.
+  def build_platform_gem(platform, dylib)
+    require "fileutils"
+    require "rubygems/package"
+    require "tmpdir"
+
+    gem_root = File.expand_path("../..", __dir__)
+    spec = Gem::Specification.load(File.join(gem_root, "omnizip.gemspec"))
+    abort "omnizip.gemspec failed to load" unless spec
+
+    Dir.mktmpdir do |tmp|
+      stage = File.join(tmp, "stage")
+      spec.files.each do |rel|
+        src = File.join(gem_root, rel)
+        next unless File.file?(src)
+
+        dst = File.join(stage, rel)
+        FileUtils.mkdir_p(File.dirname(dst))
+        FileUtils.cp(src, dst)
+      end
+
+      vendored = "vendor/libomnizip_ffi#{File.extname(dylib)}"
+      FileUtils.mkdir_p(File.join(stage, "vendor"))
+      FileUtils.cp(dylib, File.join(stage, vendored))
+
+      spec.platform = Gem::Platform.new(platform)
+      spec.files += [vendored]
+      spec.test_files = []
+
+      gemspec_path = File.join(stage, "omnizip.gemspec")
+      File.write(gemspec_path, spec.to_ruby)
+
+      pkg = File.join(tmp, "pkg")
+      FileUtils.mkdir_p(pkg)
+      Dir.chdir(stage) do
+        Gem::Package.build(spec)
+      end
+      built = Dir.glob(File.join(stage, "omnizip-*.gem")).first or
+        abort "gem build produced nothing for #{platform}"
+      FileUtils.mkdir_p(File.join(gem_root, "pkg"))
+      FileUtils.mv(built, File.join(gem_root, "pkg", File.basename(built)))
+      File.join(gem_root, "pkg", File.basename(built))
+    end
+  end
+end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: omnizip
 version: !ruby/object:Gem::Version
-  version: 0.4.5
+  version: 0.4.7
 platform: ruby
 authors:
 - Ribose Inc.
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-09-20 00:00:00.000000000 Z
+date: 2026-09-21 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: base64
@@ -601,6 +601,7 @@ files:
 - lib/omnizip/implementations/base/lzma_encoder_base.rb
 - lib/omnizip/implementations/base/state_machine_base.rb
 - lib/omnizip/implementations/rust.rb
+- lib/omnizip/implementations/rust/archive.rb
 - lib/omnizip/implementations/rust/library.rb
 - lib/omnizip/implementations/seven_zip.rb
 - lib/omnizip/implementations/seven_zip/lzma.rb
@@ -723,6 +724,7 @@ files:
 - lib/omnizip/zip/input_stream.rb
 - lib/omnizip/zip/output_stream.rb
 - lib/tasks/msi_test_files.rake
+- lib/tasks/platform_gems.rake
 - lib/tasks/rust_backend.rake
 - readme-docs/advanced-features.adoc
 - readme-docs/api-usage.adoc