canon 0.3.46 → 0.3.48

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 2dc57b150c1a0a2b9179f4b88a78c22aeb1529aa74d30857d8ca71a6159d6639
-  data.tar.gz: 65db8a7cadb8718b11a475858d3bfd5d4adcf0a4c738a7c940a6dfa60a6bfd3e
+  metadata.gz: fcfa53c5436e38941e6c789536bf9c268fb5b587930aeaca6020c22993def0a2
+  data.tar.gz: a814221d4dd934e35f8f6eb1169f8694fadf2e2fd60e49a9f8f61e6355ff9853
 SHA512:
-  metadata.gz: d411dc92e7978a904f303a95397f301286e9a12a3fc3732bb91919c2564375221b12cb13db64198f41942edeb5bbe2dfc8512b173cbfbdab44d27409d5a009a9
-  data.tar.gz: eaa339eae29ca147aa6ee9a5cecf9a328e69fc9b1f64b3acb742d11f4d451bef89efc11cde012e06e15f21b10ec09f95fbf36e0456c718ba925c015e47dba0ce
+  metadata.gz: 1ac2772108ab1d9cae92a479a98c94cbc02bff2a431a64e3dc9052ff4df5ec595a38c81849c3d9ddc1c1329b6a86bdfbb7d3aeae6b5bbacfa2d6d8d283acc0a6
+  data.tar.gz: 7794b9a9392eaa18bb43b85ec7a2e830d3c0f172bfd827e13b7e3f6290c82cb6b3491312090f8230671c1a7442e722a7c1b3d66a44398965b89ab4664ec11a4d

data/CLAUDE.md

@@ -145,7 +145,7 @@ Engine A/B testing: `CANON_XML_BACKEND=nokogiri bundle exec rspec` (or `=moxml`
 
 ### C14N Engines
 
-`Canon::Xml::C14n.canonicalize` stays on canon's Ruby C14N 1.1 processor by default. leptris' native C-side C14N 1.1 is ~50x faster on large documents and byte-identical on the main corpus, but diverges from the spec on edge cases (attribute ordering, prefix preservation, `>`/tab escaping, document-level PIs — leptris#1015, all pinned in `spec/canon/xml/c14n_engine_parity_spec.rb`). `CANON_C14N_BACKEND=leptris` opts in; when the parity gate runs clean the default flips. `canonicalize_subset` and `with_comments: true` always use the Ruby processor.
+`Canon::Xml::C14n.canonicalize` uses leptris' native C-side C14N 1.1 by DEFAULT (since libleptris 1.9.178 / gem 1.9.178.0) — 23x faster than the Ruby processor through canon's API (1MB document). The native parse runs `noblanks: true` (compact bytes, matching the Ruby lane's product), relative namespace URIs raise identically, and document-level PIs serialize in spec-correct document order (the Ruby processor's root-first order was non-conformant for prolog PIs). `CANON_C14N_BACKEND=ruby` forces the Ruby processor. `canonicalize_subset` and `with_comments: true` always use the Ruby processor. The parity spec (`spec/canon/xml/c14n_engine_parity_spec.rb`) pins byte-identity; its pending cases track intentional divergences (document-order PIs) and the canon-side CR-reference finding.
 
 ### YAML Engines
 

data/lib/canon/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Canon
-  VERSION = "0.3.46"
+  VERSION = "0.3.48"
 end

data/lib/canon/xml/c14n.rb

@@ -23,32 +23,37 @@ module Canon
       end
 
       # leptris' C-side C14N 1.1 — 23x faster than the Ruby processor
-      # through canon's own API (1MB document). libleptris 1.9.164
-      # closed four of the five leptris#1015 families (attribute
-      # ordering, prefix loss and rebinding, `>` and TAB escaping —
-      # unpinned in the parity spec since gem 1.9.174.0). A local
-      # default-flip attempt surfaced three MORE divergences the
-      # edge corpus did not cover, so the lane stays opt-in
-      # (CANON_C14N_BACKEND=leptris) until they close (leptris#1096):
-      # redundant namespace redeclarations are kept (Ruby removes
-      # them), xmlns:xml with the standard URI is kept (Ruby omits
-      # it), and document-level processing instructions are dropped.
-      # The Ruby lane additionally validates relative namespace URIs
-      # at parse time — a future flip must carry that check into the
-      # native wrapper. Comments mode keeps the Ruby path regardless.
+      # through canon's own API (1MB document) and the DEFAULT lane
+      # since libleptris 1.9.178 (gem 1.9.178.0): the leptris#1117
+      # families are closed by the native parse-policy knob
+      # (`noblanks: true` — same compact bytes as the Ruby lane's
+      # parse) and by the canonicalizer itself (whitespace-only PI
+      # data dropped, document-level "\n" separators inserted). The
+      # one residual byte difference vs the Ruby lane is intentional
+      # and a correctness improvement: document-level PIs now
+      # serialize in document order (REC-xml-c14n 2.1) instead of
+      # the Ruby lane's non-conformant root-first reorder. Relative
+      # namespace URIs raise exactly as the Ruby lane does.
+      # Comments mode keeps the Ruby path regardless (the native
+      # seam exposes no with-comments form). CANON_C14N_BACKEND=ruby
+      # forces the stdlib lane.
       def self.native_canonicalize(xml, with_comments)
         return nil if with_comments
         return nil if RUBY_ENGINE == "opal"
-        return nil unless ENV["CANON_C14N_BACKEND"].to_s.casecmp("leptris").zero?
+        return nil if ENV["CANON_C14N_BACKEND"].to_s.casecmp("ruby").zero?
         return nil unless Canon::XmlBackend.moxml? &&
           Canon::XmlParsing.moxml_adapter_name == :leptris
         return nil unless native_c14n_available?
 
-        doc = Canon::XmlParsing.moxml_context.parse(xml, readonly: true,
-                                                         strict: false)
+        validate_relative_namespaces!(xml)
+
+        doc = Canon::XmlParsing.moxml_context.parse(xml,
+                                                    readonly: true,
+                                                    strict: false,
+                                                    noblanks: true)
         begin
-          doc.native.canonicalize(::Leptris::XML::FFI::C14N_1_1, nil,
-                                  mode: ::Leptris::XML::FFI::C14N_MODE_CANONICAL)
+          doc.native.c14n(::Leptris::XML::FFI::C14N_1_1, nil,
+                          mode: ::Leptris::XML::FFI::C14N_MODE_CANONICAL)
         ensure
           doc.free
         end
@@ -58,6 +63,25 @@ module Canon
         nil
       end
 
+      # Canon::Error parity with the Ruby lane, whose parse rejects
+      # relative namespace URIs. A declaration-shaped scan; a
+      # relative-URI-shaped string inside CDATA is the only false
+      # positive this can produce.
+      RELATIVE_NS_DECL = /xmlns(?::[\w.-]+)?="([^"]*)"/
+      URI_SCHEME = %r{\A[a-zA-Z][a-zA-Z0-9+.-]*:}
+
+      def self.validate_relative_namespaces!(xml)
+        return unless xml.is_a?(String)
+
+        xml.scan(RELATIVE_NS_DECL) do |(uri)|
+          next if uri.nil? || uri.empty?
+
+          unless uri.match?(URI_SCHEME)
+            raise Canon::Error, "Relative namespace URI not allowed: #{uri}"
+          end
+        end
+      end
+
       def self.native_c14n_available?
         return @native_c14n_available unless @native_c14n_available.nil?
 

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: canon
 version: !ruby/object:Gem::Version
-  version: 0.3.46
+  version: 0.3.48
 platform: ruby
 authors:
 - Ribose Inc.