asciichem 0.27.0 → 0.28.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4602bf349585d9d81483e3561d970efd6bb06f3196d30a0a3017df316699664f
-  data.tar.gz: ea4c38d2af66b75ce2024c5cb1cc866af19584af2edba92b09cc264a7e48037e
+  metadata.gz: ccf91a425d3850a5730481c27c475dd79b219479ad91ee41a755ec72c316c75a
+  data.tar.gz: f306ca9fa6299d78f3cbabcbb51dfbf6d962c228946406b696608b349eb8ec50
 SHA512:
-  metadata.gz: 4dad38021607a88a0f37b90a2320b228d47b6e6197b7102eb48616ea94486ccfc16f953c2c233446ca216eefa1d76aebd0c88b1d40edcb2f582abb56c062e79f
-  data.tar.gz: 4fc1f074a3b182e2d22c263f09044d873a9d96ba1e627e89780e42977a00aedd0e7142590df6aa645e405e7dcbca89076e505abccc3e1e37be99dddfca527372
+  metadata.gz: 3c1823eb43793553108cef0eb28d22c98bbf79dfc275d7181b6b3538a51ea622d8001b3ce8060093ea4dbd4f0e77a33f23ab831bcbaac02accc91d2765c5446c
+  data.tar.gz: 38b55cf589f09cc497b8cfecfff3805d254fb80908269caff589a9703309c30c76a5beb1b014f123ba9fc766266640139c97dc0dd8892a8e0165f4fb191bc05f

data/CHANGELOG.md

@@ -3,6 +3,28 @@
 All notable changes to AsciiChem are documented here.
 This project follows [Semantic Versioning](https://semver.org/).
 
+## [0.28.1] - 2026-09-14
+
+### Fixed
+- Cascade transform is engine-agnostic: CascadeBuilder#canonicalise_hash
+  wraps scalar arrow/products values instead of Array(hash), which
+  enumerates a Hash. The parsanol 1.3.13 re-check (221/221 corpus-green
+  through the unmodified grammar) surfaced the shape; parslet is
+  unaffected. Spec'd for all three engine shapes.
+
+## [0.28.0] - 2026-09-14
+
+### Changed
+- Embedded-math MathML now crosses through the mml gem
+  (plurimath/mml) in both directions: Plurimath output is parsed
+  into a typed Mml::V3 graph and re-serialized by the framework
+  before grafting - replacing xpath surgery on the raw string.
+  mml (~> 2.3) is now a direct dependency (previously transitive
+  via plurimath).
+- New spec contract: every emitted MathML (and every shared-corpus
+  golden) parses back through Mml.parse - output is valid per the
+  typed MathML contract model, not just well-formed XML.
+
 ## [0.27.0] - 2026-09-14
 
 ### Added
@@ -505,3 +527,13 @@ This project follows [Semantic Versioning](https://semver.org/).
 [0.3.0]: https://github.com/asciichem/asciichem-ruby/compare/v0.2.0...v0.3.0
 [0.2.0]: https://github.com/asciichem/asciichem-ruby/compare/v0.1.0...v0.2.0
 [0.1.0]: https://github.com/asciichem/asciichem-ruby/releases/tag/v0.1.0
+    ### Changed
+- Embedded-math MathML now crosses through the mml gem (plurimath/mml)
+  in both directions: Plurimath output is parsed into a typed
+  Mml::V3 graph and re-serialized by the framework before grafting —
+  replacing xpath surgery on the raw string. mml (~> 2.3) is now a
+  direct dependency (previously transitive via plurimath).
+- New spec contract: every emitted MathML (and every shared-corpus
+  golden) parses back through Mml.parse — output is valid per the
+  typed MathML contract model, not just well-formed XML.
+

data/asciichem.gemspec

@@ -36,6 +36,7 @@ Gem::Specification.new do |spec|
   spec.add_dependency "chemicalml", "~> 0.3.0"
   spec.add_dependency "elkrb", "~> 1.0"
   spec.add_dependency "lutaml-model", ">= 0.8", "< 2"
+  spec.add_dependency "mml", "~> 2.3"
   spec.add_dependency "nokogiri", "~> 1.16"
   spec.add_dependency "parslet", "~> 2.0"
   spec.add_dependency "relaton-bib", ">= 0.1", "< 2"

data/benchmarks/README.md

@@ -15,6 +15,7 @@ WORKLOAD = ["H_2O", "Ca^2+", "SO_4^2-", "(R)-CH_3CH(OH)COOH",
 |---|---|---|---|
 | Ruby (parslet), 3.4.8 arm64 | 29.1 ms | ~2.9 ms | `bundle exec ruby benchmarks/engines.rb` |
 | Ruby + parse+text | 35.3 ms | ~3.5 ms | round-trip adds the formatter |
+| Ruby (parsanol compat, :ruby), 1.3.13 | 19.9 ms | ~2.0 ms | `benchmarks/parsanol_recheck.rb`, same session as the 11.8 ms parslet baseline below |
 | TypeScript (peggy), Node 24 | 0.19 ms | ~19 µs | `npm run bench` (asciichem-ts) |
 | Python (RD), 3.10 | 3.94 ms | ~394 µs | `python benchmarks/engines.py` (asciichem-py) |
 
@@ -54,3 +55,33 @@ upstream in parsanol-ruby#25):
 Revisit trigger unchanged: engage the native backend for full
 grammars, fix the repetition-termination bug, and beat parslet on
 this workload — then re-run the corpus against the port.
+
+### Re-check (2026-09-14, parsanol 1.3.13)
+
+The upstream "native by default" + RepetitionTag work landed, so the
+revisit trigger was tested (`benchmarks/parsanol_recheck.rb`):
+
+1. **The repetition-termination bug is fixed.** `SO_4^2-` parses and
+   round-trips, and the unmodified grammar passes the entire shared
+   corpus through the `Parsanol::Parslet` compat layer —
+   **221/221** parse/reject/round-trip cases identical to parslet.
+   One divergence surfaced on our side and is fixed in the gem:
+   parsanol's transform delivers cascade tail segments as scalar
+   hashes, and `CascadeBuilder#canonicalise_hash` used `Array(hash)`
+   (which enumerates a Hash instead of wrapping it) — now wraps
+   explicitly, engine-agnostic.
+2. **Perf (compat, :ruby forced):** 19.9 ms vs 11.8 ms per 10-input
+   pass in the same session — **~1.7x slower than parslet** (down
+   from ~6x in the first investigation). Correct but not a win.
+3. **Native still cannot serialize full parslet grammars** — two
+   upstream bugs (reported in parsanol-ruby#25):
+   `native.rb` never loads `native/dynamic` (NameError silently
+   falls back to :ruby), and `Dynamic.register` never increments
+   `@next_id`, so the second lazily-bound rule panics the Rust core
+   with "callback ID 1000000 is already registered".
+
+**Verdict: still not adopted — but one small upstream fix away from
+a meaningful re-measure.** Corpus correctness is already there; the
+native path is the whole point and remains unmeasurable until
+serialization survives a multi-rule grammar.
+

data/benchmarks/parsanol_recheck.rb

@@ -0,0 +1,79 @@
+# frozen_string_literal: true
+
+# Parsanol re-check (parsanol-ruby 1.3.13, post-issue-25): runs the
+# UNMODIFIED AsciiChem grammar over the Parsanol engine via the
+# Parslet compat shim, then (1) gates on the shared corpus, (2) gates
+# on the issue-25 EOF repro, (3) measures against the parslet path.
+#
+# Native mode cannot serialize this grammar today (two upstream bugs:
+# native.rb never loads native/dynamic; Dynamic.register never
+# increments @next_id so the second callback panics the Rust core —
+# parsanol-ruby#25). The measurement therefore forces :ruby, the
+# only working mode for full parslet grammars via the shim.
+#
+# Run from asciichem-ruby/:
+#   ruby -I /tmp/parsanol_spike -I ../parsanol/parsanol-ruby/lib benchmarks/parsanol_recheck.rb
+require "benchmark/ips"
+require "asciichem"
+require "json"
+
+Parsanol::Native.singleton_class.define_method(:available?) { false } # spike: force :ruby
+
+puts "parsanol #{Parsanol::VERSION} | parslet-compat Parser=#{Parsanol::Parslet::Parser}"
+puts "AsciiChem::Grammar superclass: #{AsciiChem::Grammar.superclass}"
+
+# -- 1. Issue-25 repro: repeat-of-maybe at end of input --------------
+begin
+  formula = AsciiChem.parse("SO_4^2-")
+  puts "issue-25 repro SO_4^2-: PARSES -> #{formula.to_text.inspect}"
+rescue AsciiChem::ParseError => e
+  puts "issue-25 repro SO_4^2-: FAILS -> #{e.message[0, 100]}"
+end
+
+# -- 2. Shared-corpus gate --------------------------------------------
+corpus_dir = File.expand_path("../../asciichem-tests/corpus/fixtures", __dir__)
+cases = Dir[File.join(corpus_dir, "*.json")].sort.flat_map { |p| JSON.parse(File.read(p)) }
+parser_cases = cases.select { |c| c.key?("input") && !c.key?("lint") && !c.key?("convention") }
+
+pass = fail_parse = fail_reject = fail_roundtrip = 0
+parser_cases.each do |fixture|
+  input = fixture.fetch("input")
+  if fixture.fetch("parses")
+    begin
+      formula = AsciiChem.parse(input)
+      if fixture["roundTrip"] && formula.to_text != input
+        fail_roundtrip += 1
+        puts "  ROUNDTRIP DIFF: #{input.inspect} -> #{formula.to_text.inspect}" if fail_roundtrip <= 5
+      end
+      pass += 1
+    rescue AsciiChem::ParseError, Parslet::ParseFailed => e
+      fail_parse += 1
+      puts "  PARSE FAIL: #{input.inspect} -> #{e.message[0, 90]}" if fail_parse <= 8
+    end
+  else
+    begin
+      AsciiChem.parse(input)
+      fail_reject += 1
+      puts "  SHOULD REJECT: #{input.inspect}" if fail_reject <= 8
+    rescue AsciiChem::ParseError, Parslet::ParseFailed
+      pass += 1
+    end
+  end
+end
+total = parser_cases.length
+puts format("corpus gate: %d/%d ok (parse-fails %d, should-reject %d, roundtrip-diffs %d)",
+            pass, total, fail_parse, fail_reject, fail_roundtrip)
+
+# -- 3. Performance ----------------------------------------------------
+WORKLOAD = [
+  "H_2O", "Ca^2+", "SO_4^2-", "(R)-CH_3CH(OH)COOH",
+  "2H_2 + O_2 -> 2H_2O", "N_2 + 3H_2 <=>[Fe][400C] 2NH_3",
+  "C1-C-C-C-C-C1", "CH_3-CH_2-OH",
+  '^14C @name("carbon-14") @cas("14104-86-4")',
+  "A ->[heat] B ->[cool] C",
+].freeze
+
+Benchmark.ips do |x|
+  x.report("parsanol parse x10") { WORKLOAD.each { |s| AsciiChem.parse(s) } }
+  x.compare!
+end

data/lib/asciichem/formatter/mathml.rb

@@ -1,6 +1,7 @@
 # frozen_string_literal: true
 
 require "nokogiri"
+require "mml"
 
 module AsciiChem
   module Formatter
@@ -196,17 +197,27 @@ module AsciiChem
       end
 
       def visit_embedded_math(em)
-        # Strip the outer <math> wrapper so the embedded fragment slots
-        # into our surrounding <mrow>.
-        fragment = em.formula.to_mathml
-        parsed = Nokogiri::XML(fragment)
-        math = parsed.at_xpath("//m:math", m: MATHML_NS)
-        return el("mrow") unless math
-
-        # Detach children into a fresh <mrow>.
+        # Embedded math crosses through the mml gem in both directions:
+        # Plurimath's MathML string is parsed into a typed Mml::V3
+        # graph (the MathML contract model), and each top-level child
+        # is re-serialized by the framework before grafting — never
+        # ad-hoc xpath surgery on the string.
+        math = Mml.parse(em.formula.to_mathml, version: 3)
         mrow = el("mrow")
-        math.children.each { |c| mrow.add_child(c.dup) }
+        math.element_order.each do |entry|
+          next unless entry.respond_to?(:name)
+
+          attr_name = xml_element_attribute(math.class, entry.name)
+          next unless attr_name
+
+          Array(math.public_send(attr_name)).each do |child|
+            fragment = Nokogiri::XML::DocumentFragment.parse(child.to_xml)
+            fragment.children.each { |node| mrow.add_child(node) }
+          end
+        end
         mrow
+      rescue Mml::Error, Lutaml::Model::Error, Nokogiri::XML::SyntaxError
+        el("mrow")
       end
 
       def visit_text(text)
@@ -309,6 +320,14 @@ module AsciiChem
 
       private
 
+      # Symbol of the model attribute an XML child element name maps
+      # to on a Mml class (e.g. "mstyle" => :mstyle_value), via the
+      # lutaml-model XML mapping — no name-munging.
+      def xml_element_attribute(klass, element_name)
+        rule = klass.mappings_for(:xml).elements.find { |r| r.name == element_name }
+        rule&.to
+      end
+
       def render_node(node)
         node.accept(self)
       end

data/lib/asciichem/transform.rb

@@ -830,11 +830,18 @@ module AsciiChem
 
       def canonicalise_hash(hash)
         first = hash[:first]
-        arrows = Array(hash[:arrow])
-        products = Array(hash[:products])
-        tail = arrows.zip(products)
+        # Wrap, never Array(): Array(hash) on a scalar Hash value
+        # (the shape parsanol's transform delivers) would enumerate
+        # key/value pairs instead of wrapping it.
+        tail = wrap(hash[:arrow]).zip(wrap(hash[:products]))
         [first, tail]
       end
+
+      def wrap(value)
+        return [] if value.nil?
+
+        value.is_a?(Array) ? value : [value]
+      end
     end
 
     # Maps the captured stereo letter to the model's stereo symbol.

data/lib/asciichem/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module AsciiChem
-  VERSION = "0.27.0"
+  VERSION = "0.28.1"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: asciichem
 version: !ruby/object:Gem::Version
-  version: 0.27.0
+  version: 0.28.1
 platform: ruby
 authors:
 - Ribose Inc.
@@ -57,6 +57,20 @@ dependencies:
     - - "<"
       - !ruby/object:Gem::Version
         version: '2'
+- !ruby/object:Gem::Dependency
+  name: mml
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '2.3'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '2.3'
 - !ruby/object:Gem::Dependency
   name: nokogiri
   requirement: !ruby/object:Gem::Requirement
@@ -176,6 +190,7 @@ files:
 - benchmarks/RESULTS.md
 - benchmarks/benchmark.rb
 - benchmarks/engines.rb
+- benchmarks/parsanol_recheck.rb
 - exe/asciichem
 - lib/asciichem.rb
 - lib/asciichem/citation.rb