makiri 0.6.0-aarch64-linux → 0.7.0-aarch64-linux

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 6c0a020c0e52029940d64a91b53ae459d08b1d53e3106c3b312b6c86601bc29a
-  data.tar.gz: 2d5bf5107d4de3892b70684aaad11b67f0f7eec74496e01132f79703db87262e
+  metadata.gz: 1117686dde2fd5e8a6e863971b3220a7bfffd621e3dea8097d0b0b0b14f75275
+  data.tar.gz: ee2fbfb3b456a86f9444bf9faa9ccdf051b224e44ffc951e2b84ef49333e6907
 SHA512:
-  metadata.gz: 6e0eda64911e48477808cf4b033b8f24fdbcb978558e27db9ea570bcd4364eb45e81e0a39b0c0e3ebe357aff622fb479e0c230b6c218f4f23ab760d3cd163cc8
-  data.tar.gz: 96f2da4e924ade762e57887c128bd70583330917a18f09abafd2721c2fdc61f647e5d1eb64544b8af753182c2e83d320b9dd2e02dcd384e3187ac27b3b27a352
+  metadata.gz: 5fa4fce173089263057caf5f8f13ab305d60c6d07f7df4a2d5b921d7025a25be5ace35c7797e9664938f7537a7ff3c6870cfacd75b6689f3e692f3e9c9d003a8
+  data.tar.gz: 97694793f4875a0378582e504b2df12baa4902d18c2b28ea9d3fc9ed78de0da9310af78f67a7a7a5e5298a321477175343fcc34320ea49787450e0951b56a371

data/.github/workflows/ci.yml

@@ -12,7 +12,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, macos-latest]
+        os: [ubuntu-latest, macos-latest, windows-latest]
         ruby: ["3.2", "3.3", "3.4", "4.0"]
 
     steps:
@@ -42,6 +42,7 @@ jobs:
           THREADING: ${{ (matrix.os == 'ubuntu-latest' && matrix.ruby == '3.4') && '1' || '' }}
 
       - name: Smoke-load the gem
+        shell: bash
         run: bundle exec ruby -Ilib -r makiri -e 'p Makiri::VERSION'
 
   sanitize:

data/.github/workflows/release.yml

@@ -62,10 +62,12 @@ jobs:
         - x86_64-linux
         - aarch64-linux
         - arm64-darwin
+        - x64-mingw-ucrt
         include:
-          - { platform: x86_64-linux,  os: ubuntu-latest }
-          - { platform: aarch64-linux, os: ubuntu-24.04-arm }
-          - { platform: arm64-darwin,  os: macos-14 }
+          - { platform: x86_64-linux,   os: ubuntu-latest }
+          - { platform: aarch64-linux,  os: ubuntu-24.04-arm }
+          - { platform: arm64-darwin,   os: macos-14 }
+          - { platform: x64-mingw-ucrt, os: windows-latest }
     steps:
       - uses: actions/checkout@v6
         with:
@@ -79,6 +81,7 @@ jobs:
       - name: Compile (builds vendored Lexbor, then the ext)
         run: bundle exec rake compile
       - name: Smoke-load
+        shell: bash
         run: bundle exec ruby -Ilib -r makiri -e 'p Makiri::VERSION'
       - name: Stage the compiled binary under its ABI subdir
         shell: bash
@@ -97,16 +100,25 @@ jobs:
         run: |
           dlext="$(ruby -e 'print RbConfig::CONFIG["DLEXT"]')"
           bin="lib/makiri/makiri.${dlext}"
-          if [ "$(uname)" = "Darwin" ]; then
-            if otool -L "$bin" | grep -i 'libruby'; then
-              echo "::error::$bin hard-links libruby; precompiled gem would not load on other Ruby installs"; exit 1
-            fi
-          else
-            if readelf -d "$bin" 2>/dev/null | grep -iE 'NEEDED.*libruby'; then
-              echo "::error::$bin has a libruby NEEDED entry; precompiled gem may not load on a static Ruby"; exit 1
-            fi
-          fi
-          echo "OK: $bin has no libruby dependency"
+          case "$(uname -s)" in
+            Darwin)
+              if otool -L "$bin" | grep -i 'libruby'; then
+                echo "::error::$bin hard-links libruby; precompiled gem would not load on other Ruby installs"; exit 1
+              fi
+              ;;
+            MINGW*|MSYS*|CYGWIN*)
+              # Windows PE DLLs MUST import the Ruby DLL via the import library
+              # (no dynamic_lookup on PE/COFF). A libruby import here is by design,
+              # not a portability bug, so this assertion does not apply.
+              echo "Windows (x64-mingw-ucrt): Ruby import-library link is by design; skipping libruby assertion"
+              ;;
+            *)
+              if readelf -d "$bin" 2>/dev/null | grep -iE 'NEEDED.*libruby'; then
+                echo "::error::$bin has a libruby NEEDED entry; precompiled gem may not load on a static Ruby"; exit 1
+              fi
+              ;;
+          esac
+          echo "OK: portability check complete for $bin"
       - uses: actions/upload-artifact@v7
         with:
           name: bin-${{ matrix.platform }}-${{ matrix.ruby }}
@@ -128,6 +140,8 @@ jobs:
           os: ubuntu-24.04-arm
         - platform: arm64-darwin
           os: macos-14
+        - platform: x64-mingw-ucrt
+          os: windows-latest
     steps:
       - uses: actions/checkout@v6
         with:
@@ -157,6 +171,7 @@ jobs:
       - name: Build the native gem
         run: ruby script/build_native_gem.rb "${{ matrix.platform }}"
       - name: Verify it installs without recompiling
+        shell: bash
         run: |
           gem install "makiri-"*"-${{ matrix.platform }}.gem" --local
           ruby -e 'require "makiri"; puts Makiri::VERSION'

data/CHANGELOG.md

@@ -1,5 +1,45 @@
 # Changelog
 
+## [0.7.0] - 2026-07-11
+
+### Added
+
+* `Document#create_document_type(name, public_id = "", system_id = "")` on both
+  `Makiri::HTML::Document` and `Makiri::XML::Document` (DOM
+  `DOMImplementation.createDocumentType`): creates a detached `DocumentType`
+  node, to be placed before the document element (e.g. with
+  `root.add_previous_sibling(doctype)`). The name is case-preserving; an omitted
+  or empty public/system id is treated as absent; an invalid name (or a `"` in an
+  id) raises. Inserting a doctype is validated against the WHATWG rules — it may
+  only be a child of the document, must precede the document element, and a
+  document holds at most one — so a misplaced or duplicate doctype raises rather
+  than producing a malformed tree.
+
+* The XML `DocumentType` is now part of the tree: it appears in
+  `Document#children` (before the root, like a browser DOM) and carries
+  `parent`/sibling links, for both a parsed `<!DOCTYPE>` and the factory above.
+  `Document#internal_subset` still returns it. XPath is unaffected — its 1.0 data
+  model excludes the doctype (`//node()` / `/node()` never surface it), as in
+  Nokogiri/libxml2. This supports building a document with
+  `DOMImplementation.createDocument(namespace, qualifiedName, doctype)`.
+
+* `Makiri::XML::Document#create_loose_dom_element(qualified_name, prefix,
+  local_name, namespace_uri)`, an escape hatch for browser-DOM interop that
+  creates an element whose name follows WHATWG DOM element-name rules rather
+  than the stricter XML QName rules (e.g. `"f:o:o"`, `":foo"`, `"0:a"`). The
+  regular factories stay XML-strict. Such an element is intentionally not
+  XML-serializable: `#to_xml` and `#canonicalize` raise `Makiri::Error` if the
+  tree contains one. Renaming it to a valid XML name (`#name=`) returns it to
+  strict mode.
+
+### Fixed
+
+* Replacing an XML node with a `DocumentFragment` (`node.replace(fragment)`) is
+  now all-or-nothing: if the replacement is rejected (e.g. it would leave the
+  document with two root elements), the target node and the fragment are left
+  untouched, instead of dropping the target and splicing in only part of the
+  fragment. Replacing the root element with a single-element fragment now works.
+
 ## [0.6.0] - 2026-07-05
 
 ### Changed

data/Rakefile

@@ -290,7 +290,15 @@ end
 desc "CBMC proofs over the Ruby/Lexbor-free carve-out (core + XML + XPath front; " \
      "needs cbmc; see docs/formal_verification.ja.md and verify/Makefile)"
 task :verify do
-  sh "make -C verify smoke selftest cbmc"
+  # The 10 CBMC proofs are independent processes with a lopsided cost
+  # distribution (a few ~1-2min solves dominate, the rest are trivial), so run
+  # them in parallel: wall-clock drops toward the single slowest harness with no
+  # loss of coverage. Cap -j at the CPU count (Etc.nprocessors; the runner has
+  # RAM for that many concurrent core-set solves). smoke/selftest are quick and
+  # come first as ordered goals.
+  require "etc"
+  jobs = Integer(ENV.fetch("VERIFY_JOBS", Etc.nprocessors))
+  sh "make", "-C", "verify", "-j#{jobs}", "smoke", "selftest", "cbmc"
 end
 
 desc "Run the performance benchmark (Makiri vs Nokogiri reference)"
@@ -309,13 +317,6 @@ task "bench:xml" => :compile do
   end
 end
 
-desc "Run the CSS selector-query benchmark on a note.com-style SPA page (set BENCH_CARDS)"
-task "bench:css" => :compile do
-  Bundler.with_unbundled_env do
-    sh "#{FileUtils::RUBY} -Ilib bench/bench_css_query.rb"
-  end
-end
-
 namespace :conformance do
   desc "WHATWG HTML5 parsing conformance: run html5lib-tests through Makiri"
   task html5: :compile do

data/lib/makiri/attr.rb

@@ -2,9 +2,6 @@
 
 module Makiri
   # An attribute node. Most of the API lives in C (#name, #value).
-  #
-  # An attribute knows its owning element through the document's lazily-built
-  # attr->owner index, exposed as both #parent and #element.
   class Attr < Node
     # The element this attribute belongs to, or nil if detached. Defined as a
     # method (not an alias) so it resolves #parent dynamically on the per-kind

data/lib/makiri/comment.rb

@@ -2,9 +2,7 @@
 
 module Makiri
   class Comment < Node
-    # Create a detached comment owned by +document+ (Nokogiri-style constructor;
-    # the document comes FIRST for Comment / CDATASection / ProcessingInstruction, unlike
-    # Element / Text). Delegates to {Document#create_comment}.
+    # Create a detached comment owned by +document+ (Nokogiri-style constructor)
     #
     # @param document [Makiri::Document]
     # @param content [String]

data/lib/makiri/compat_aliases.rb

@@ -3,18 +3,7 @@
 module Makiri
   # Nokogiri-compatible class-name aliases.
   #
-  # Makiri's canonical node-class names are the WHATWG DOM interface names
-  # (Element, Attr, Text, Comment, CDATASection, ProcessingInstruction,
-  # DocumentType, Document, DocumentFragment). Two of those differ from the
-  # libxml2/Nokogiri spelling; we expose the Nokogiri names as aliases so a
-  # snippet ported from Nokogiri (or an is_a?/case check) resolves unchanged:
-  #
-  #   CDATASection  <- CDATA   (Nokogiri::XML::CDATA)
-  #   DocumentType  <- DTD     (Nokogiri::XML::DTD)
-  #
-  # An alias is the same class object, so #is_a? works under either name; only
-  # #class.name (and inspect) report the canonical DOM name. Defined at all three
-  # levels (the abstract base and the HTML::/XML:: leaves).
+  # Makiri's canonical node-class names are the WHATWG DOM interface names.
   CDATA = CDATASection
   DTD = DocumentType
 

data/lib/makiri/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Makiri
-  VERSION = "0.6.0"
+  VERSION = "0.7.0"
 end

data/script/check_c_safety_allowlist.yml

@@ -96,19 +96,13 @@ ignore_paths:
   - path: ext/makiri/glue/ruby_node_set.c
     rule: node_raw_boundary
     reason: NodeSet.new stores representation-opaque pointers; mkr_node_raw takes a seed node's pointer only after same-document validation guarantees its kind matches the set.
-
-# --- parser-TU reader discipline (raw_scan_call / raw_cursor_member) ---
-allowlist:
-  # --- verified_text_forge: the Ruby-free test entry points that mint over a
-  #     runtime buffer (not a compile-time literal, which uses
-  #     mkr_verified_text_lit). Neither can use mkr_verified_text_from_view
-  #     (bridge = Ruby boundary), so each mints itself and must supply the
-  #     contract by construction. ---
+  # verified_text_forge in fuzz/*.c
   - path: ext/makiri/fuzz/xpath_fuzz.c
     rule: verified_text_forge
-    max: 1
     reason: "The libFuzzer harness mints over an owned mkr_strndup copy, which supplies NUL-termination and no-interior-NUL; UTF-8 validity is deliberately left to the lexer's strict decoder (its rejection path is fuzz-target behavior)."
   - path: ext/makiri/fuzz/xml_xpath_fuzz.c
     rule: verified_text_forge
-    max: 1
     reason: "The expression mint is over an owned mkr_strndup copy, which supplies NUL-termination and no-interior-NUL; UTF-8 validity is pre-checked (the harness filters invalid UTF-8 before parsing). Mirrors xpath_fuzz.c."
+
+# --- parser-TU reader discipline (raw_scan_call / raw_cursor_member) ---
+allowlist: []

data/verify/Makefile

@@ -36,7 +36,7 @@ LDLIBS ?= -lm
 INC    := -I $(EXT) -I $(LXB) -I .
 
 CORE_SRC  := $(EXT)/core/mkr_alloc.c $(EXT)/core/mkr_buf.c \
-             $(EXT)/core/mkr_core.c $(EXT)/core/mkr_utf8.c
+             $(EXT)/core/mkr_core_selftest.c $(EXT)/core/mkr_utf8.c
 XML_SRC   := $(EXT)/xml/mkr_xml_chars.c $(EXT)/xml/mkr_xml_index.c \
              $(EXT)/xml/mkr_xml_mutate.c $(EXT)/xml/mkr_xml_node.c \
              $(EXT)/xml/mkr_xml_tree.c

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: makiri
 version: !ruby/object:Gem::Version
-  version: 0.6.0
+  version: 0.7.0
 platform: aarch64-linux
 authors:
 - takahashim