moxml 0.5.81 → 0.5.83

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 957697bcb855cea040ac0d8b2f202a7cd8fd230482137602fcf7d50fc2712ebb
-  data.tar.gz: 22815ac3e9920a8939b0076f3b4222a154e0420d4fe1a21d9b7e56bd9b0818b8
+  metadata.gz: 47eeb00aed4df2a6ea9474713ed65ba577889375e2ea7a49a8b36ee7fde86090
+  data.tar.gz: 947e4a25bd867ce42a52015ceea6fe576ded8b1ccf9a232cc5b1a69e1c22b166
 SHA512:
-  metadata.gz: 422b076bb191e01b27693af02a66d4e985f574dbcc49451c9af395c1905f9cabfdb2faf0e1b15303a5444c17a8fba98e70e5bea7d5c61f1f152d9fa09a550329
-  data.tar.gz: 2d75b66d322414ea8c99b79fe1ed78c7aa758b90b7f5b7f56a685e9f861bca12d6f641803aa04ced575a812677a085619bd40ee435963b035cdbe005a36d1b55
+  metadata.gz: daaf3e2d077a2fee6b710d61b512a1e8c89922727cbf147aca2fb69147602aa3c50369bdc990cb385e3522296151102fa5498f0df5e11955d1b16424fa18c2b1
+  data.tar.gz: 55f51e85aa61abcb1bc9ab735cd1e863152c46f284a85ff78355306b0698305772b968c1690d7c0c991715afdd433911ac70fce6ce0fde5f133fa27001748ffc

data/Rakefile

@@ -263,6 +263,14 @@ namespace :benchmark do
     sh "bundle exec ruby benchmark/plan_bench.rb"
   end
 
+  # SAX bench row (leptris#1298 / #279): the bulk record drain vs
+  # the callback Recorder. With LEPTRIS_BENCH_LOCK=1 (CI) it exits 1
+  # unless records-walk allocations stay under half the recorder's.
+  desc "SAX parse benchmark: record drain vs callback recorder"
+  task :sax do
+    sh "bundle exec ruby benchmark/sax_bench.rb"
+  end
+
   desc "Generate adapter benchmark report"
   task :report do
     ruby "benchmarks/generate_report.rb"

data/benchmark/sax_bench.rb

@@ -0,0 +1,99 @@
+# frozen_string_literal: true
+
+# SAX bench row (leptris#1298 follow-up / moxml#279 item 4): the
+# bulk record drain vs the callback Recorder on the canon parse-lane
+# shape. Run via `rake benchmark:sax`; with LEPTRIS_BENCH_LOCK=1
+# (CI) it exits 1 unless the drain stays materially cheaper than the
+# recorder — records-walk allocations must land under half the
+# recorder's, so the win stays visible and a regression re-opens
+# the lane.
+require "moxml"
+
+item = "<item n=\"1\" kind=\"k\" id=\"i1\">value text</item>"
+DOC = "<r>#{item * 225}<t>tail</t></r>".freeze # rubocop:disable Lint/ConstantDefinitionInBlock
+DOCS = [DOC, DOC].freeze # rubocop:disable Lint/ConstantDefinitionInBlock
+
+module SaxBenchHandlers
+  class NullHandler < Moxml::SAX::Handler
+    def on_start_element(_name, _attrs = {}, _namespaces = {}); end
+
+    def on_end_element(_name); end
+
+    def on_characters(_text); end
+  end
+
+  class RecordWalkHandler < Moxml::SAX::Handler
+    attr_reader :elements
+
+    def initialize
+      super
+      @elements = 0
+    end
+
+    # The hot-loop shape: walk the records, materialize one name per
+    # element (the #1298 target — consumer-chosen allocations).
+    def on_sax_records(table)
+      i = 0
+      while i < table.count
+        @elements += 1 if table.kind(i).zero? && table.view(i)
+        i += 1
+      end
+      :claimed
+    end
+  end
+end
+
+CTX = Moxml.new(:leptris).freeze # rubocop:disable Lint/ConstantDefinitionInBlock
+BRIDGE = Moxml::Adapter::Leptris::LeptrisSAXBridge # rubocop:disable Lint/ConstantDefinitionInBlock
+
+def allocations
+  GC.start
+  before = GC.stat(:total_allocated_objects)
+  GC.disable
+  yield
+  GC.enable
+  GC.stat(:total_allocated_objects) - before
+end
+
+def cpu_micros
+  GC.start
+  t0 = Process.clock_gettime(Process::CLOCK_PROCESS_CPUTIME_ID, :microsecond)
+  yield
+  Process.clock_gettime(Process::CLOCK_PROCESS_CPUTIME_ID, :microsecond) - t0
+end
+
+# warm both lanes
+handler = SaxBenchHandlers::NullHandler.new
+3.times { DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, handler) } }
+3.times { DOCS.each { |doc| Leptris::XML::SAX::Parser.new(BRIDGE.new(handler)).parse(doc) } }
+walker = SaxBenchHandlers::RecordWalkHandler.new
+3.times { DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, walker) } }
+
+recorder_allocs = allocations do
+  DOCS.each { |doc| Leptris::XML::SAX::Parser.new(BRIDGE.new(handler)).parse(doc) }
+end
+replay_allocs = allocations do
+  DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, handler) }
+end
+walk_allocs = allocations do
+  DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, walker) }
+end
+
+replay_time = cpu_micros { 5.times { DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, handler) } } }
+recorder_time = cpu_micros { 5.times { DOCS.each { |doc| Leptris::XML::SAX::Parser.new(BRIDGE.new(handler)).parse(doc) } } }
+walk_time = cpu_micros { 5.times { DOCS.each { |doc| Moxml::Adapter::Leptris.sax_parse(doc, walker) } } }
+
+puts format("recorder           %<allocs>6d allocs  %<time>6dµs", allocs: recorder_allocs, time: recorder_time / 5)
+puts format("drain replay       %<allocs>6d allocs  %<time>6dµs", allocs: replay_allocs, time: replay_time / 5)
+puts format("drain records-walk %<allocs>6d allocs  %<time>6dµs", allocs: walk_allocs, time: walk_time / 5)
+puts "(#{walker.elements} elements walked cumulatively; 452 per rep)"
+
+if ENV["LEPTRIS_BENCH_LOCK"]
+  if walk_allocs >= recorder_allocs / 2
+    warn "LEPTRIS_BENCH_LOCK: records-walk allocs (#{walk_allocs}) " \
+         "not under half the recorder's (#{recorder_allocs}) — the " \
+         "#1298 win regressed"
+    exit 1
+  end
+  puts "lock: records-walk under half the recorder's allocations ✓"
+end

data/lib/moxml/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Moxml
-  VERSION = "0.5.81"
+  VERSION = "0.5.83"
 end

data/spec/moxml/adapter/leptris_sax_dup_attr_spec.rb

@@ -0,0 +1,50 @@
+# frozen_string_literal: true
+
+require "spec_helper"
+
+# Duplicate attributes are a fatal XML 1.0 §3.1 error (moxml#281,
+# issue #130 banner semantics): the SAX lane must surface them via
+# on_error exactly like the DOM lane's parse_diagnostics. On
+# bindings whose drain claims dup-containing documents (engine <
+# 1.9.242 — leptris#1374), the drain swallows the error; those
+# bindings skip this spec until the ride lands.
+module SaxDupAttrHandlers
+  class ErrorCollector < Moxml::SAX::Handler
+    attr_reader :errors, :elements
+
+    def initialize
+      super
+      @errors = []
+      @elements = 0
+    end
+
+    def on_error(error)
+      @errors << error.message
+    end
+
+    def on_start_element(_name, _attrs = {}, _namespaces = {})
+      @elements += 1
+    end
+  end
+end
+
+RSpec.describe "SAX duplicate-attribute error surfacing" do
+  let(:context) { Moxml.new(:leptris) }
+  let(:dup_xml) do
+    '<body lang="en" xml:lang="en" xml:lang="en"><div>x</div></body>'
+  end
+
+  it "surfaces the redefined-attribute error through sax_parse" do
+    if defined?(Leptris::XML::SAX::Records) &&
+        !Leptris::XML::SAX::Records.open(dup_xml).nil?
+      skip "engine < 1.9.242 — the drain claims dup-attr documents " \
+           "(leptris#1374)"
+    end
+
+    handler = SaxDupAttrHandlers::ErrorCollector.new
+    context.sax_parse(dup_xml, handler)
+
+    expect(handler.errors.join).to include("xml:lang redefined")
+    expect(handler.elements).to eq(2)
+  end
+end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: moxml
 version: !ruby/object:Gem::Version
-  version: 0.5.81
+  version: 0.5.83
 platform: ruby
 authors:
 - Ribose Inc.
@@ -41,6 +41,7 @@ files:
 - Rakefile
 - benchmark/pipeline_bench.rb
 - benchmark/plan_bench.rb
+- benchmark/sax_bench.rb
 - benchmarks/.gitignore
 - benchmarks/generate_report.rb
 - bin/console
@@ -437,6 +438,7 @@ files:
 - spec/moxml/adapter/entity_restoration_spec.rb
 - spec/moxml/adapter/headed_ox_spec.rb
 - spec/moxml/adapter/leptris_doc_parts_spec.rb
+- spec/moxml/adapter/leptris_sax_dup_attr_spec.rb
 - spec/moxml/adapter/leptris_sax_records_spec.rb
 - spec/moxml/adapter/leptris_spec.rb
 - spec/moxml/adapter/libxml_internals_spec.rb