md2any 0.3.0 — math rendering overhaul, rendering robustness, docs & crate polish

.gitignore

@@ -5,7 +5,6 @@ Cargo.lock.bak
 # IDE
 .idea/
 .vscode/
-.claude/
 *.iml
 
 # LibreOffice lock files (created when opening a doc for testing)

CHANGELOG.md

@@ -7,6 +7,66 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.3.0] — 2026-06-14
+
+A math-rendering and robustness release. The native math layout engine gained
+font-aware metrics, real bold, and properly-centred accents; custom OTF/CFF
+fonts now embed correctly in PDF; and a broad rendering sweep fixed several
+overflow / silent-drop bugs.
+
+### Added
+
+- **Font-aware math metrics.** The math layout engine now measures glyph
+  advances against the face that will actually render them (the embedded PDF
+  font, or bundled DejaVu for SVG/PNG) via a `GlyphMetrics` provider, so
+  reserved width matches drawn glyphs — equations stay aligned under
+  `--pdf-font`, not just with the default font.
+- **Math bold.** `\mathbf`, `\textbf`, `\boldsymbol`, `\bm`, `\pmb`, `\mathbfit`
+  now render in real bold weight (PDF/SVG/PNG); previously a no-op, and
+  `\boldsymbol` leaked literal source.
+- **`\dfrac` / `\tfrac` / `\cfrac`** and **`\mid` / `\vert` / `\Vert`** are now
+  recognised (previously leaked literal source).
+- **GFM table column alignment** (`:---`, `:---:`, `---:`) is honoured in PDF,
+  SVG, PNG, and HTML.
+- **Inline `<svg>…</svg>`** blocks are now rasterised and embedded instead of
+  being silently dropped.
+
+### Fixed
+
+- **OTF/CFF font embedding.** PDF now emits `CIDFontType0` + `FontFile3` for
+  CFF/OpenType faces (e.g. STIX Two Math) instead of mislabelling them as
+  TrueType, which strict readers rejected.
+- **Math accents** (`\bar`, `\hat`, `\vec`, `\tilde`, `\dot`, `\ddot`) are drawn
+  as centred geometry rather than zero-advance combining marks, which had
+  landed at the glyph's right edge (and differently per font — e.g. `\bar{d}`
+  rendered as "đ").
+- **Long titles no longer overflow.** Title slides (all layouts) and section
+  dividers auto-shrink to fit; wrapped content-slide headings push the body and
+  underline down instead of overprinting them.
+- **Nested blockquotes** (`>`/`>>`/`>>>`) keep every level instead of rendering
+  only the deepest.
+- **Long unbreakable tokens** (URLs, hashes, CamelCase) hard-break to wrap
+  instead of overflowing the slide edge (PDF/SVG); HTML uses `overflow-wrap`.
+- **Tabs in code** expand to 4-column tab stops instead of rendering as
+  missing-glyph boxes.
+- **`{width=N%}` > 100** clamps instead of leaking the literal attribute text.
+- **`:::` columns** handle leading/trailing/duplicate dividers (the fence form
+  `::: … ::: … :::`) instead of dumping all content into the right column.
+- **Task-list items** show only their checkbox, not a checkbox *and* a bullet
+  (PDF/SVG/HTML/PPTX).
+- **Display-math pagination** weights generated equation images by their real
+  aspect ratio, so a one-line equation no longer splits onto a near-empty
+  continuation slide.
+
+### Known limitations
+
+- Right-to-left text uses `direction: rtl` for right-alignment but does not
+  perform full Unicode bidi reordering of mixed LTR/RTL runs.
+- CJK and colour-emoji glyphs require `--cjk <font>` for PDF (the bundled DejaVu
+  face does not cover them).
+- Table column alignment is not yet applied to the editable Office/ODF outputs
+  (pptx/odp/odt/docx).
+
 ## [0.2.0] — 2026-05-24
 
 A feature-and-fix-heavy follow-up to the initial release. New CLI surface

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "md2any"
-version = "0.2.0"
+version = "0.3.0"
 edition = "2021"
 rust-version = "1.74"
 description = "Markdown → PowerPoint, OpenDocument Impress, PDF, Word, Writer, HTML, SVG, and PNG. One markdown source, one small Rust binary."
@@ -22,6 +22,11 @@ exclude = [
     "/examples/*.odp",
     "/examples/*.docx",
     "/examples/*.odt",
+    # The manual in binary office formats (~1.2 MB) is redundant with HELP.md
+    # for a source crate; keep it in the repo, out of the published package.
+    "/docs/manual",
+    # Large pre-rendered sample images not referenced by any shipped example.
+    "/examples/assets/*.png",
 ]
 
 [[bin]]

README.md

@@ -9,6 +9,10 @@
 binary that converts markdown into editable PowerPoint, OpenDocument Impress,
 PDF, Word, LibreOffice Writer, standalone HTML, and per-slide SVG/PNG files.
 
+<p align="center">
+  <img src="docs/img/slides.png" alt="Content, code, and table slides rendered by md2any" width="100%">
+</p>
+
 ```bash
 md2any talk.md                    # → talk.pptx     (PowerPoint)
 md2any talk.md -o talk.odp        # → talk.odp      (LibreOffice Impress)
@@ -44,6 +48,24 @@ md2any is the only tool in this list that produces every common Office /
 OpenDocument format **natively** from a single small binary — no headless
 browser, no LaTeX stack, no language runtime.
 
+## Native math — no LaTeX, no KaTeX, no browser
+
+A built-in layout engine renders display math (fractions, radicals, scripts,
+matrices, cases, accents, bold) as **selectable text and vector geometry** —
+not images, and without a TeX or JavaScript stack. Bring any math font with
+`--pdf-font` (here, STIX Two Math); metrics adapt to the face automatically.
+
+<p align="center">
+  <img src="docs/img/equations.png" alt="Euler's identity, the Lorentz factor, the Dirac equation, and Bayes' theorem" width="100%">
+</p>
+
+It scales all the way up to the full **Standard Model Lagrangian** on a single
+A4 page ([source](examples/standard-model-lagrangian-a4.md)):
+
+<p align="center">
+  <img src="docs/img/lagrangian.png" alt="The Standard Model Lagrangian rendered on one A4 page" width="62%">
+</p>
+
 ## Install
 
 ### Prebuilt binaries
@@ -138,13 +160,13 @@ md2any hello.md --serve         # live preview at http://localhost:8421
 | **Themes** | Light + dark built-in, full custom palettes via YAML overlay |
 | **Layouts** | Clean / studio / frame / bold |
 | **Aspect ratios** | 16:9, 4:3, 9:16, A3/A4/A5, Letter, Legal, Tabloid + landscape variants + custom `WxH[unit]` |
-| **Markdown** | Lists (9 deep nesting), tables, dark-by-default code blocks (40+ tags incl. mainframe/IBM i, Haskell/BCPL, web/API/config, and `bf`), block quotes, footnotes, links (clickable), strike, inline code |
+| **Markdown** | Lists (9 deep nesting), task lists, tables with column alignment (`:---:` / `---:`), dark-by-default code blocks (40+ tags incl. mainframe/IBM i, Haskell/BCPL, web/API/config, and `bf`), block quotes, footnotes, links (clickable), strike, inline code, inline `<svg>` |
 | **Layout extras** | Side-by-side columns (`:::`), landscape code two-up (`--code-columns` / `columns=2`), image/text full-page hints, per-slide background (`<!-- bg: -->`), TOC injection, hand-tuned pagination with widow/orphan control |
-| **Math** | `$inline$` and `$$display$$` LaTeX subset -> Unicode, including accents/text/math alphabets; `--math source` preserves source; `--math svg` uses a built-in box layout for display fractions, radicals, scripts, matrices/cases/arrays, scalable delimiters, and aligned equations |
+| **Math** | `$inline$` and `$$display$$` LaTeX subset -> Unicode, including accents/text/math alphabets; `--math source` preserves source; `--math svg` uses a built-in box layout for display fractions, radicals, scripts, matrices/cases/arrays, scalable delimiters, aligned equations, **bold** (`\mathbf`/`\boldsymbol`), and drawn accents (`\bar`/`\hat`/`\vec`). Font-aware metrics keep equations aligned under `--pdf-font` (incl. OTF/CFF math fonts like STIX) |
 | **Diagrams** | Embedded `dot`, `mermaid`, `plantuml` (shells out if installed) |
 | **Speaker notes** | `<!-- notes: -->` HTML comments |
 | **Transitions** | `fade`, `push`, `wipe`, `cover`, `split` (PPTX/ODP/PDF) |
-| **RTL / CJK** | `direction: rtl` for Arabic/Hebrew; CJK via the consumer's font |
+| **RTL / CJK** | `direction: rtl` right-aligns Arabic/Hebrew (no full bidi reordering of mixed runs yet); CJK & emoji need `--cjk <font>` for PDF |
 | **Workflow** | `--watch` for rebuild-on-save, `--serve` for live preview with hot reload, `--check` for lint mode |
 | **Handouts** | `--handout 2|4|6` for N-up A4 printable PDF |
 | **Multi-file** | Concat multiple `.md` files; `-` reads stdin |
@@ -371,6 +393,10 @@ std::fs::write("talk.pdf", bytes)?;
   `--font-fallback`
 - DejaVu does not include CJK glyphs; for PDF Chinese/Japanese/Korean decks,
   pass `--cjk PATH` or include a CJK face in `--font-fallback`
+- `direction: rtl` right-aligns Arabic/Hebrew but does not yet perform full
+  Unicode bidirectional reordering of mixed LTR/RTL runs
+- GFM table column alignment (`:---:`, `---:`) is applied in PDF / SVG / PNG /
+  HTML; the editable Office/ODF outputs (pptx/odp/odt/docx) don't carry it yet
 - Speaker notes attach natively in PPTX / ODP, can be emitted as a PDF
   companion with `--with-notes`, can be appended to DOCX/ODT via
   `--doc-style speaker-notes`, and can be bundled with deck + handout via

docs/torture-test-findings.md

@@ -0,0 +1,97 @@
+# md2any torture-test findings
+
+Broad stress test of the renderer (2026-06-14). Six domains — tables/lists/columns,
+code/syntax, typography/unicode, images/SVG, themes/layouts/aspects, and multi-format
+robustness — each rendered to PDF/PNG (and cross-checked in HTML/Office formats) and
+visually inspected.
+
+**Multi-format output is solid:** 72 artifacts across all 8 formats (pptx, odp, pdf,
+docx, odt, html, svg, png) — no crashes, no corrupt archives, no dropped content.
+All defects below are in the slide layout/render layer.
+
+Status legend: ☐ open · ☑ fixed · ◐ in progress · ⊘ won't-fix / larger effort
+
+---
+
+## HIGH
+
+- ☑ **H1 — Long titles overflow & overprint the footer.** Fixed across **all**
+  title layouts (Clean/Studio/Frame/Bold): `fit_hero_size` shrinks the hero title
+  to fit above the subtitle/footer; Bold no longer drops the title off the top of
+  its accent block (`src/pdf.rs`). Subtitle/author stay visible.
+
+- ☑ **H2 — Wrapped content headings collide with body.** Fixed: `render_content_slide`
+  measures the wrapped title line count and pushes the underline + body down by the
+  extra lines (`src/pdf.rs`). Was: fixed-height title band.
+
+- ☑ **H3 — Section-divider long titles clipped off the bottom.** Fixed: section
+  divider (Clean/Frame) now auto-shrinks via `fit_hero_size` (`src/pdf.rs`).
+
+- ☑ **H4 — Inline `<svg>…</svg>` silently dropped.** Fixed: the parser accumulates
+  inline SVG across HTML events and routes it through the image pipeline as a
+  base64 `data:image/svg+xml` block, rasterised like external SVG (`src/parser.rs`).
+
+- ☑ **H5 — Table column alignment ignored** (`:---:`, `---:`). Fixed: added
+  `ir::ColumnAlign` + `Table.aligns`, captured from the GFM delimiter row; honoured
+  in **PDF / SVG / PNG / HTML**. (Office formats — pptx/odp/odt/docx — thread the
+  data but don't yet apply native cell alignment; see L7.)
+
+- ☑ **H6 — Tabs in code → tofu boxes + broken indentation.** Fixed: code lines
+  expand tabs to 4-column tab stops at parse time, fixing every backend
+  (`src/parser.rs`).
+
+## Known-limitation class (larger efforts)
+
+- ⊘ **H7 — RTL (Arabic/Hebrew) not reordered.** No bidi; Hebrew reversed, Arabic
+  left-aligned. Needs a bidi pass — larger effort.
+
+- ⊘ **H8 — CJK + emoji → tofu in PDF.** Only DejaVu embedded; no CJK/emoji fallback
+  font. HTML preserves them. Needs a fallback font pipeline — larger effort.
+
+## MEDIUM
+
+- ☑ **M1 — Nested blockquotes collapse.** Fixed: the quote accumulator only
+  resets on the outermost `>`, so nested levels keep the outer paragraphs
+  (`src/parser.rs`).
+
+- ☑ **M2 — Long unbreakable tokens overflow the right edge.** Fixed: PDF/SVG
+  wrappers hard-break a token wider than the line into character chunks; HTML
+  uses `overflow-wrap: anywhere` (`src/pdf.rs`, `src/svg.rs`, `src/html.rs`).
+
+- ☑ **M3 — `{width=N%}` > 100 leaks literal `{width=…}` as body text** (+ spurious
+  slide). Fixed: `parse_width_pct` now clamps to 1..=100 instead of returning `None`
+  (`src/parser.rs`).
+
+- ☑ **M4 — `:::` columns: a leading/extra marker dumps everything into the right
+  column** (empty left). Fixed: `coalesce_columns` now splits on all dividers and
+  drops empty segments, so the fence form `::: … ::: … :::` works (`src/paginate.rs`).
+
+- ☑ **M5 — Task-list items render a bullet AND a checkbox** (`● ☐ task`). Fixed:
+  added `ListItem::is_task()`; bullet suppressed in pdf/svg/html/pptx render_list.
+  ODT/DOCX use native list styles and still show both — needs a no-bullet list
+  style declaration (follow-up, see L7).
+
+- ☐ **M6 — Pagination over-splits in 16:9** — image+caption and tables split with
+  slides ~60% empty (a4 fits them). Same family as the math-image weight fix already
+  landed; the height budget is too conservative for several block types in 16:9.
+
+## LOW / polish
+
+- ☐ L1 — Code: leading/trailing blank lines inside a fence kept verbatim.
+- ☐ L2 — Rust `'static` lifetime mis-highlighted as a char/string literal.
+- ☐ L3 — `&nbsp;` not decoded (renders literal `nbsp`).
+- ☐ L4 — `₿` (U+20BF) tofu; other currency symbols fine.
+- ☐ L5 — Misleading "aspect too small" error for ratios like `16:10`/`1:1` (only
+  `16:9`/`4:3`/`9:16` presets accepted).
+- ☐ L6 — TOC lists only H1 sections.
+
+---
+
+## Fix order
+
+Cheap/safe first, then the big UX wins:
+M3 (width clamp) → M5 (task-list bullet) → M4 (`:::` leading marker) →
+H1/H2/H3 (title & heading auto-fit + body push-down) → H4 (inline SVG) →
+H5 (table alignment, needs IR field) → M6 (16:9 pagination budget) →
+M1 (nested blockquotes) → M2 (token overflow) → low/polish.
+H7/H8 (RTL, CJK/emoji) are larger and tracked separately.

examples/famous-equations.md

@@ -0,0 +1,78 @@
+---
+title: A Wall of Famous Equations
+subtitle: The greatest hits, set in STIX Two Math
+author: md2any
+date: auto
+theme: light
+aspect: 16:9
+layout: clean
+math: svg
+math_block_align: center
+math_scale: 1.2
+math_max_height: 230
+toc: false
+---
+
+# A Wall of Famous Equations
+
+A greatest-hits tour — from a patent clerk's β to the geometry of spacetime.
+
+## Euler's Identity
+
+The most beautiful equation in mathematics — five constants, one line.
+
+$$ e^{i\pi} + 1 = 0 $$
+
+## Mass–Energy Equivalence
+
+Einstein's 1905 sequel to *On the Electrodynamics of Moving Bodies*.
+
+$$ E = mc^2 $$
+
+## The Lorentz Factor
+
+Straight out of §6 of that 1905 paper — the β that warps space and time.
+
+$$ \gamma = \frac{1}{\sqrt{1 - \dfrac{v^2}{c^2}}} $$
+
+## The Schrödinger Equation
+
+How a quantum state evolves in time.
+
+$$ i\hbar\frac{\partial}{\partial t}\Psi = \hat{H}\Psi $$
+
+## The Dirac Equation
+
+Relativity meets the quantum — and predicts antimatter.
+
+$$ \left(i\hbar\gamma^\mu\partial_\mu - mc\right)\psi = 0 $$
+
+## Einstein's Field Equations
+
+Matter tells spacetime how to curve; spacetime tells matter how to move.
+
+$$ G_{\mu\nu} + \Lambda g_{\mu\nu} = \frac{8\pi G}{c^4}\,T_{\mu\nu} $$
+
+## The Gaussian Integral
+
+The bell curve's deceptively simple area.
+
+$$ \int_{-\infty}^{\infty} e^{-x^2}\,dx = \sqrt{\pi} $$
+
+## Bayes' Theorem
+
+How evidence updates belief.
+
+$$ P(A \mid B) = \frac{P(B \mid A)\,P(A)}{P(B)} $$
+
+## Boltzmann's Entropy
+
+Carved on his gravestone in Vienna.
+
+$$ S = k_B \ln W $$
+
+## The Basel Problem
+
+Euler summed the reciprocal squares — and π showed up.
+
+$$ \sum_{n=1}^{\infty} \frac{1}{n^2} = \frac{\pi^2}{6} $$