omnizip 0.4.5 → 0.4.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 5a55c37209e986f7e4499e8a485163728e9c8f0dff506e1aa852672a9585653f
-  data.tar.gz: ba0b5437026feed4c1fe4d0ab0146f868386ebe2284ec7b08b45c7260930d582
+  metadata.gz: 7dc6e02021b3915af5f65944e4d734bd3f95a5624b2a01c30100b6bbc36d5856
+  data.tar.gz: 998838c2bd3df94ced6617b9d199a0f2eab552e332d285bdc6819860fef50324
 SHA512:
-  metadata.gz: 13d25032be26ccd096718846ac74126fd14ab3e2729d1e9ed366a3c431815b69c0817e33092f214eea4ed041e41e2eebe070ac92862bd54c70b0379017281637
-  data.tar.gz: 8e68e9a4a0d7f489be9956baf6f2d18bb82d5cc0db123adaf689ab73aefca1dcb44c0172a8e774c0cbcfbcca2927c9f40b62da184678588e084642b3b86cd7eb
+  metadata.gz: f02b277ef180e076b5e0deb3847ed4cdf3f2a83abf829095779cdbfbdbc811891c8d9db9e000fb071c1d3bc164bd3cf4f031b8e19207b7d6e48232a21bf573dc
+  data.tar.gz: 4d20378f2973be45cc1cb691894b9c54916c1f5546d45a64a5143feaf8cabf0e053d8d604b8f30b6534d123ad12084021d9815b82945f78f5bfacd64f4daef96

data/CHANGELOG.md

@@ -25,6 +25,17 @@ 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/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

@@ -86,7 +86,7 @@ module Omnizip
             funcs = function_table.each_with_object({}) do |(key, (name, args, ret)), h|
               h[key] = Fiddle::Function.new(handle[name], args, ret)
             end
-            new(funcs)
+            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
@@ -120,8 +120,9 @@ 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 +137,31 @@ 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 = "0.4.6"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: omnizip
 version: !ruby/object:Gem::Version
-  version: 0.4.5
+  version: 0.4.6
 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