arrolio 0.1.4 → 0.1.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: f0b6af09c229c5ee869a2e052c61933ccf441cccf02ace5b965acaf2dcdfa277
-  data.tar.gz: 650539bee9ae880208f9986e26ccc4a9ffe38acbf773a47a1471a54cfb43f688
+  metadata.gz: 057f22d28911ebd5fb250d0a3ee3aae065a493c2c8aee0a5ff329a138da38674
+  data.tar.gz: 8477b14736601c87ee24c690aa867ef175de94568ef866b48d260948c3d98116
 SHA512:
-  metadata.gz: c8ea9131445f66cdcae37550aebb04036bcbedc995e713220f3e2c76391b2c35e6daa6f71b6e2d32a51f6ad33b526c36e6d2fdc5fac5dd92d0440b31bb7c2a7c
-  data.tar.gz: 19a06bdb5726413dcea2de7022eab8ede88c9a10133ad78545211eb294e1ec5884bd47c7753fdaed9ddc6ef5afac1e71b177fdc30b6061f5af39a03ff894b30e
+  metadata.gz: 2aef88dd6df44e8aef2be218f34f53de0e332ecfb94bc2ef5aa8a87f0221fab634ac05ad0a900997fc1ebb870529626a464eeaa302c7f3e19d1b05664bfd7e90
+  data.tar.gz: b984b3dae46396305d96e5c6977a24992c4ddae317be8a4682fafed9de46a35a7e2442dbc418147c77afb943ba7981c6b8be28b562b85723fe5624b3242b6c7c

data/CLAUDE.md

@@ -53,16 +53,17 @@ PDF bytes (via Pdfrb)
 
 Each layer is independently testable. Keep layer boundaries airtight: Content knows nothing about pages or PDF; Engine knows nothing about PDF bytes; Renderer knows nothing about layout algorithms.
 
-## Project state (as of 2026-08-03)
+## Project state (as of 2026-09-01)
 
-Generic config-driven pipeline works end-to-end: the OIML r060/1 fixture renders to a 25-page A4 PDF (~194 KB) via `Arroolio::ConfigDrivenPipeline` pointed at `flavors/oiml/`. 232 specs pass.
+The generic config-driven pipeline renders the OIML r060/1 fixture to a 28-page A4 PDF matching the mn2pdf v2.55 reference at **68.92% text parity, 28/28 pages** (`bundle exec rake parity:check`). 452 specs, 0 rubocop offenses, 214 files.
 
 What's in the gem (core, flavor-agnostic):
 - `lib/arrolio/{content,style,layout_spec,font,font_metrics,glyph_measurer,inline_run,text_layout,flowable,flowables,frame,flow_context,engine,output,renderer,harness,flavor}/`
-- `lib/arrolio/generic_adapter.rb` — reads `adapter_rules.yml`, parses any flavor's XML into `Content::Document`.
-- `lib/arrolio/generic_flow_builder.rb` — reads `flow_rules.yml`, builds flowables for any flavor.
-- `lib/arrolio/config_driven_pipeline.rb` — orchestrates adapter + flow builder + engine + renderer + ToC.
-- `lib/arrolio/asset_resolver.rb`, `lib/arrolio/toc_builder.rb` — generic helpers.
+- `lib/arroolio/generic_adapter.rb` — dispatch + document glue; per-concern conversion lives in included seams under `lib/arroolio/generic_adapter/` (FootnoteExtraction, TableConversion, InlineRunCollection, HeadingExtraction, ListConversion, MetadataExtraction, DocumentExtraction). Adding an element family = a module + one include.
+- `lib/arroolio/generic_flow_builder.rb` — `build()` + dispatch; per-family emission seams under `lib/arroolio/generic_flow_builder/` (Sequences, Terms, Notes, Figures, Lists, Bibliography, Tables).
+- `lib/arroolio/config_driven_pipeline.rb` — orchestrates adapter + flow builder + engine + renderer + ToC; resolves fonts via fontist (`FontScanner`).
+- `lib/arroolio/renderer/pdf.rb` — page emission; Pdfrb coupling points in `lib/arroolio/renderer/pdf/` (FontEmbedding, Assets, Metadata). The renderer never measures glyph widths — `Line::PlacedRun` carries `chunk_widths` from the breakers.
+- `lib/arroolio/asset_resolver.rb`, `lib/arroolio/toc_builder.rb` — generic helpers (both with direct specs).
 - `exe/arrolio2pdf` — generic CLI taking a flavor directory argument.
 - `scripts/xsl_to_config.rb` — generates all three flavor YAML files from an XSL stylesheet.
 - `data/arrolio/` — AFM metrics + glyphlist for the 14 PDF standard fonts (core data, not flavor data).
@@ -71,18 +72,13 @@ What's outside the gem (flavor artifacts):
 - `flavors/oiml/{layout_spec,adapter_rules,flow_rules}.yml` — generated by `xsl_to_config.rb` from the authoritative XSL at `~/src/mn/metanorma-taste/data/oiml/oiml.xsl`.
 - `spec/fixtures/flavors/sample/` — synthetic flavor-neutral fixture for the gem's tests.
 
-What's deferred (intentionally — see `TODO.complete/`):
-- TODO 53: second flavor spike (proves the generic design works beyond OIML).
-- TODO 54: `flavors/<name>/manifest.yml` contract.
-- TODO 55: header/footer style in `layout_spec.yml` (currently hardcoded in renderer).
-- TODO 56: split `GenericAdapter` into MECE per-element converters.
-- TODO 57: strict `Content::Document#sections` typing (only `Section[]`).
-- TODO 58: flavor packaging convention (`arrolio-<flavor>` gems).
-- TODO 59: XSL converter profiles (metanorma, dita, docbook).
-- TODO 60: footnote / note-hanging-indent / example-body-margin fidelity.
-- TODO 61: typed error model + strict mode.
-- TODO 62: font resolution via fontist (currently absolute paths in YAML).
-- Plus the older TODO.implementation items (Knuth-Plass, PageSequenceMaster, table layout, etc.).
+What's open (intentionally — see `TODO.complete/`, frontmatter `status:` is authoritative):
+- TODO 87: odd/even page templates beyond header parity (headers mirror; templates don't).
+- TODO 88: hyphenation support.
+- TODO 90: MathML mfrac true stacking (the open piece).
+- TODO 93: figure text searchability (figures render as images; the reference's figure text extracts).
+- TODO 96: pagination drift parity — the master ledger: remaining region deficits, keep-rules negative results (the reference does NOT enforce widows; term entries are NOT atomic), the heading-gap coupled-landing recipe, and the C2 margin-unification decode (a dedicated re-tuning session, not a release-day change).
+- TODOs 53-62 are DONE (second flavor spike, manifest contract, header/footer config, adapter MECE, strict typing, packaging docs, XSL profiles, footnote fidelity, typed errors, fontist).
 
 The TODO files in `TODO.complete/` and `TODO.implementation/` are the roadmap. `TODO.complete/README.md` covers the parity roadmap; `TODO.implementation/README.md` covers the foundation phase plan.
 
@@ -137,8 +133,9 @@ When extending any layer, apply OCP/DRY/MECE strictly:
 ## Renderer/Pdfrb coupling notes
 
 - `Renderer::Pdf#render_page` deliberately points each pdfrb page's `Resources` Hash at the catalog's. Pdfrb's `Fonts#add` attaches fonts to the *catalog* Resources, but each Page shadows with its own empty Hash — without this workaround, fonts are invisible to readers.
-- `Renderer::FontRegistry` memoises font family name → Pdfrb resource Symbol per render pass. Falls back to Helvetica for unknown families so the render never crashes mid-document. TODO 62 makes this strict.
-- TODO 55 moves the header/footer visual style (Arial 9pt + 0.5pt border, currently hardcoded in the renderer) into `layout_spec.yml` so each flavor controls its own header/footer look.
+- Margin contract (see `Flowable`'s doc): TextFlowable (and HeadingFlowable) count `space_before/after` inside height/emit; List/Image/Table-based flowables do not — spacing around them travels as explicit Spacers. The engine collapses via `Engine::MarginCollapse`; do NOT unify without a dedicated re-tuning session (decoded in TODO 96).
+- Header/footer styling comes from `layout_spec.yml` (`header_footer:` block); font resolution goes through fontist (`FontScanner`) with AFM fallback for the 14 standard fonts.
+- Parity workflow: `bundle exec rake parity:check` / `parity:diff PAGE=N`; drift-map tooling and diagnostic cautions live in TODO 96.
 
 ## Key files to read first
 

data/TODO.complete/96-pagination-drift-parity.md

@@ -512,6 +512,28 @@ or the layout differs from the pipeline.
   gaps reproduce (p8 69 vs 76) but nets -6pp; still blocked by the
   p9-p15 fill mismatch.
 
+## C2 margin unification — attempt executed and reverted (2026-08-31)
+
+Round-5 attempt with the prepared MarginCollapse seam. The
+implementation (margins inside every List/Image/Note/Table
+flowable, CSS inset = max(pa,b) - pa + b, page-top exception,
+Spacer workarounds deleted, block-style neutralization for the
+dead note/list margins) measured 31.7% / 30pp vs the committed
+68.92% / 28-28 - outside the correction budget, reverted clean.
+
+ROOT CAUSE DECODED (the load-bearing finding): the OLD model
+under-charges every counting flowable's space_before - ink is
+drawn ABOVE the cursor, inside the previous flowable's consumed
+space_after zone. The reference's calibrated constants ride that
+under-charging; a geometrically correct model shifts every b>0
+charge. Page-top behavior also differs from paper analysis
+(heading ink at landings implies the old fresh-page path consumes
+mt INSIDE the page - re-derive from HeadingFlowable's emit before
+the next attempt, and note TableFlowable still needs the
+convention added). Unification = dedicated re-tuning session, not
+a release-day change. Full convention map now on Flowable's
+contract doc.
+
 ## Measurement
 
 `bundle exec rake parity:check` — 64.07% (2026-08-17).

data/TODO.complete/README.md

@@ -1,11 +1,13 @@
 # TODO.complete — Path to 100% mn2pdf parity
 
-## Current baseline (2026-08-09)
+## Current baseline (2026-09-01)
 
-`bundle exec rake parity:check` reports **62.7%** overall similarity
-on the OIML r060/1 fixture, **28 pages vs 28 reference** (all pages PARTIAL+).
+`bundle exec rake parity:check` reports **68.92%** overall similarity
+on the OIML r060/1 fixture, **28 pages vs 28 reference**.
 
-376 specs pass, 0 failures, 0 rubocop offenses across 174 files.
+452 specs pass, 0 failures, 0 rubocop offenses across 214 files.
+TODO 96 is the live ledger (region deficits, keep-rule negative
+results, the C2 margin decode).
 
 ## Status snapshot
 

data/lib/arrolio/flowable.rb

@@ -16,6 +16,20 @@
 #
 # A flowable's height() and emit() must agree on whether margins
 # are included, or fit checks and drawing diverge.
+#
+# Convention map: TextFlowable (and HeadingFlowable, its
+# subclass) count margins. ListFlowable + its subclasses (lists,
+# notes), ImageFlowable, and TableFlowable do not — spacing
+# around them travels as explicit Spacers.
+#
+# UNIFICATION DECODED (2026-08-31 attempt, reverted): the old
+# model UNDER-CHARGES every counting flowable's space_before —
+# its ink is drawn ABOVE the cursor, inside the previous
+# flowable's already-consumed space_after zone. The reference's
+# calibrated constants ride that under-charging. A correct CSS
+# model (inset = max(pa,b) - pa + b, margins inside every
+# flowable) measures 31.7% / 30pp: every b>0 charge shifts
+# geometry. Unify only with a full re-tuning session.
 module Arrolio
   class Flowable
     attr_reader :style

data/lib/arrolio/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Arrolio
-  VERSION = '0.1.4'
+  VERSION = '0.1.6'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: arrolio
 version: !ruby/object:Gem::Version
-  version: 0.1.4
+  version: 0.1.6
 platform: ruby
 authors:
 - Ribose Inc.