CHANGELOG.md

Changelog

All notable changes to markdown2pdf are documented here. The format follows Keep a Changelog and the project adheres to Semantic Versioning. Each release section below is what ships as the GitHub Release notes.

[1.6.0] - 2026-07-22

A security and robustness release. A crash on malformed HTML is fixed, remote fetching is hardened against SSRF and unbounded downloads, local image reads can now be confined to a directory, three RUSTSEC advisories are closed, and CI gained real gates for formatting, linting, and dependency audits.

Thanks to Alexander Bruegmann (@Brueggus), first-time contributor, for the table header background fix in #121.

• Table header backgrounds render correctly: a header row's background_color was resolved but never drawn, and the fill was missing again on every repeated header after a page or column break. Header and alternating-row fills now share one drawing path. Thanks to Alexander Bruegmann.
• Fixed a crash on malformed HTML: a block such as <p </p>, where the opening tag is never closed, made the wrapper-stripping code compute an inverted byte range and panic, taking down the calling process. Such blocks now fall back to verbatim HTML rendering. Reachable from a seven-character document, so any caller rendering untrusted Markdown was exposed.
• Remote fetching hardened: the renderer now resolves a remote image's host and validates every resolved address before connecting, then pins the connection to the address it checked. Loopback, private, link-local, CGNAT, and IPv6 NAT64 targets are refused, so a document can no longer reach cloud metadata endpoints or internal services. Redirects are capped and each hop is re-validated. Downloads are bounded in both size and total wall-clock time; previously the size cap trusted a server-supplied Content-Length and the timeout only measured idle time, so a slow-drip server could hold a fetch open indefinitely. The CLI's --url input gained the same size and time bounds. Set MARKDOWN2PDF_ALLOW_PRIVATE_NETWORK=1 to opt out on an intranet.
• Local image reads can be confined: a new [security] config block adds image_root, allow_absolute_image_paths, and allow_remote_images. With image_root set, image paths resolve inside it and anything escaping it, including through a symlink, is refused and degrades to alt text. Defaults are unchanged, so existing documents behave exactly as before; callers rendering untrusted Markdown should set image_root. See docs/configuration.md.
• SVG parsing no longer loads external resources: an untrusted SVG could previously reference local files through <image href="..."> during parsing. Data URIs still work.
• Crate documentation now matches the real schema: the configuration examples on the docs.rs landing page used keys that no longer exist ([heading.1], size, textcolor, fontfamily) and could not parse against the deny_unknown_fields schema. parse_into_file also had its documentation attached to the wrong function. A test now parses the documented examples so they cannot drift again.
• Three advisories closed: RUSTSEC-2026-0187 (lopdf, stack overflow), RUSTSEC-2026-0185 (quinn-proto, memory exhaustion), and RUSTSEC-2026-0009 (time, denial of service). cargo audit reports zero vulnerabilities.
• printpdf 0.9 to 0.11: collapses lopdf to a single 0.44 copy, where two versions previously coexisted, and drops the unmaintained proc-macro-error from the tree entirely.
• CI covers the optional features and gates on quality: fetch and svg gate real code that CI had never compiled. Builds and tests now run across a four-way feature matrix, and the fmt, clippy, and cargo audit jobs block a merge rather than reporting into the void. A weekly scheduled run catches newly published advisories.
• Renderer internals reorganised: network policy and image-path policy moved out of the layout engine into net_guard, net_read, and image_policy. No behavior change.
• Whole tree reformatted: cargo fmt --all applied crate-wide for the first time, with the development toolchain pinned so the result is reproducible. Purely mechanical.
• Breaking: ResolveError is now #[non_exhaustive], so exhaustive matches need a wildcard arm, and ResolveError::BadToml.source changed from toml::de::Error to Box<toml::de::Error>, which shrinks the error type from 160 to 80 bytes. Code that constructs or exhaustively matches this type needs updating. The MSRV is now 1.88, which the existing lopdf dependency already required in practice.

[1.5.1] - 2026-07-17

Two rendering fixes: math \text{…} now renders non-Latin scripts, and documents with many headings no longer produce corrupt PDFs.

• Math \text{…} renders non-Latin scripts — characters STIX Two Math lacks (CJK, Arabic, Devanagari, Cyrillic, …) inside \text{}/\operatorname{} and as bare symbols are now outlined from the document's body font and [defaults].fallback_fonts faces instead of silently vanishing. RTL runs reorder to visual order (ordering only — no joining/shaping); zero-width format characters (ZWNJ / ZWJ / bidi marks) stay invisible; a character no configured font covers renders as a placeholder box with a warning instead of disappearing. Covered-Latin math output is byte-identical.
• Corrupt PDFs with many headings — bumped lopdf to 0.44, fixing an object-stream defect that could corrupt PDFs containing many headings.

Resolves #115 and #119.

[1.5.0] - 2026-05-26

Layout polish — multi-column flow, plus targeted fixes to tables, defs, admonitions, HTML, math, page breaks, and Unicode rendering.

• Multi-column layout — [page] columns = N (1..=4), column_gap_mm; default columns = 1 byte-identical to pre-feature output. Headers, footers, title page, and TOC stay single-column.
• Admonitions — nested-body footnotes resolve against doc-wide numbering; caution/important keep their authored label with canonical kind styling; auto-emitted strings seed the external-font subset; a page-bottom admonition keeps its label with its first body line.
• GFM tables in list items / blockquotes / admonitions — a 4-space-indented table inside any container body tokenises as a table instead of leaking literal pipes.
• Definition lists — bodies capture 4-space-indented continuations and re-lex in block context (code, table, blockquote, nested list, multiple paragraphs); multi-term groups (Alpha\nBeta\n: shared) and a second : block after a blank line are recognised.
• HTML & links — inline <span>/<strong>/<em>/<code>/<sup>/<sub> render semantically; <div class="…">…</div> unwraps; <br/>/<hr/> interpreted; unresolved wikilinks render in a dead-link colour; long URLs wrap at /?&# boundaries; #dup-2 links resolve.
• Math engine — $$ … = … $$ no longer eaten as a setext H1 underline; \lvert/\rvert/\lVert/\rVert/\omicron added; tall paren/bracket assemblies for 4+ row matrices stack in the right order; long display equations scale to fit; inline $…$ no longer dropped when it is the only content in its paragraph.
• Widow / orphan — headings drag their first follow-on chunk across page breaks; a bullet glyph no longer renders alone at the page bottom.
• Unicode by default — when no font is configured, auto-pick an installed Unicode body font (Helvetica Neue / Geneva on macOS, Segoe UI / Arial / Tahoma on Windows, DejaVu Sans / Liberation Sans / Noto Sans on Linux) so café, em-dashes, smart quotes, and math symbols outside $..$ stop becoming ?. Same fallback fires when the configured name is a built-in alias whose only on-disk copy is a .ttc collection the loader skips. CJK / Arabic / Devanagari / emoji still need explicit [defaults].fallback_fonts.

Resolves #102, #105, #106, #107, #108, #109, and #111.

1.4.0 - 2026-05-23

Fallback font chain, end-to-end style-config fidelity, transliteration-aware measurement, and automatic config-file discovery.

• Fallback font chain — new fallback_fonts list on [defaults] (plus FontConfig::with_fallback_fonts(...)); each codepoint matches the primary first, then each fallback in declaration order. A codepoint covered by nothing degrades visibly (? with built-in, missing-glyph box with external) instead of failing. Names resolve like font_family (built-in aliases, system names, .ttf/.otf paths); missing fonts log and skip. Wrapping stays consistent across font boundaries.
• Built-in measurement matches emission — the transliterating built-in path used to price source codepoints (•, —, …, smart quotes, NBSP) at the font's average width while emitting their ASCII expansion. The cursor drifted and underlines/link rects landed beside their text on links separated by • or —. Measurement now routes through the same transliteration the emit path uses; Unicode-font documents unaffected.
• Config file discovery — with no -c flag, the CLI checks MARKDOWN2PDF_CONFIG, then ./markdown2pdf.toml, then ~/.config/markdown2pdf/config.toml before the default theme.
• Style-config fidelity — a long tail of style fields the renderer was silently dropping is now honoured: body font (system lookup returning a bold face as the regular weight is fixed, [defaults].font_family actually loads, heading bold from the style block embeds its variant faces, [link].underline = false suppresses link underline); per-block (letter_spacing_pt, indent_pt, underline/strikethrough/text_align, table cell_padding/alternating_row_background); inline (every code_inline and [mark] field including font_family/padding); layout (list indent_per_level_pt, new bullet_gap_pt on [list.common], image caption styling, title-page cover image, admonition body inheritance). Center- and right-aligned paragraphs no longer collapse wrapped lines onto line one. Only page.columns still deferred — tracked as #102.

Resolves #81, #97, and #100.

1.3.0 - 2026-05-20

Merged GFM table cells, inline HTML anchors and structural wrappers, and admonition callout boxes — all additive, render path unchanged for docs that don't use them.

• Merged table cells — > extends the cell before it by one column, ^ continues the cell above by one row; markers match raw cell source so \>/\^ stay literal. They chain and combine; layout keeps a spanned group whole across page breaks. Plain tables byte-identical.
• Inline HTML anchors & structural wrappers — <a href="…" title="…">…</a> rewrites to a real link token (clickable PDF annotation with tooltip, same path as [text](url "title")). <div>/<section>/<figure>/<figcaption> (joining existing <p>/<center>) unwrap. Malformed input falls through; unknown tags still render verbatim; no scripting introduced.
• Admonition / callout boxes — MkDocs !!! note "Title" (4-space-indented body) and GitHub > [!WARNING] both render as tinted boxes with accent border, bold header, and a vector icon (● ⓘ 💡 ⚠ ⊗, ≡ for unknown kinds). Five first-class kinds absorb aliases (caution/error → danger, important → info, warn/attention → warning, hint → tip); unknown kinds get a grey box with the raw label uppercased. New [admonition] config block plus per-kind [admonition.<kind>] overlays; every bundled theme ships its own palette.

Resolves #83, #84, and #86.

1.2.0 - 2026-05-18

LaTeX math is now parsed and typeset in-tree — $…$ (inline) and $$…$$ (display) stop leaking through as literal text.

• In-tree TeX math engine — fractions, radicals, sub/superscript stacks, big operators with limits, growing delimiters, matrices, accents, and the blackboard/script/fraktur alphabets, laid out over STIX Two Math's OpenType MATH metrics.
• Vector outlines, not embedded font — math draws as filled outlines, so no font is embedded and the equation isn't selectable (behaves like a figure). Matches LaTeX/MathJax/KaTeX-to-PDF pipelines and keeps PDFs small.
• Pandoc-style delimiters — $ opener needs a non-space after it, closer needs a non-space before and no trailing digit; \$ is literal; unterminated $/$$ degrades to text.
• Coverage — \frac/\binom, \sqrt[n]{}, coupled sub/superscript stacks, big operators with \limits/\nolimits, growing \left…\right/\big delimiters, pmatrix/cases/aligned environments, accents (incl. stretchy \widehat), \mathbb/\mathbf/\mathcal/… alphabets, operator names. Unknown commands degrade to literal text; input depth bounded so adversarial markup can't blow up layout.
• [math] config block — align (center/left/right), scale, color, and block margins. See docs/configuration.md.

Resolves #78.

1.1.0 - 2026-05-18

Note-vault syntax: inline highlight, WikiLinks, and Pandoc inline footnotes — Obsidian/Pandoc/MediaWiki-style docs export to PDF without leaking syntax as literal text. Plus two lexer/validation edge cases surfaced during the work.

• Inline footnotes (text^[note body]) (#80) — Pandoc-style footnotes inline, no separate [^id]: definition. Lexer scans a balanced bracket pair (nesting and \-escapes respected), sub-lexes the body as inline content, and assigns a doc-unique internal label from a counter shared across nested sub-lexers. Inline and [^id] footnotes share one numbering sequence in first-reference order, the marker is the same superscript linking to #footnote-N, and bodies collect into the single tail Footnotes section without splitting their host paragraph. Unbalanced ^[ or empty ^[] degrades to literal text.
• WikiLinks ([[Target]], [[Target|Label]]) (#82) — target slugifies to an in-document anchor and flows through the same resolution path as [text](#slug): match becomes a clickable internal jump; unmatched logs a warning and falls back to styled text so a partial export isn't broken. Unclosed [[ or escaped \[\[…\]\] renders literally.
• Inline highlight (==text==) (#79) — new Highlight token painting a configurable background; nestable with other inline styles (==**bold**== is bold and highlighted). New [mark] config block with background_color (default #FFF59D). Dispatch runs after Setext detection so ===/--- lines still underline the paragraph above. Unterminated == degrades.
• Lexer fix — unclosed [… spanning a line break no longer emits a .notdef box. parse_link's no-closing-bracket fallback used to flatten the multi-line body into one Text token with an embedded \n, which the renderer had no glyph for. Multi-line bodies now route through the pending-token path so each break survives as a real Token::Newline. Covered by an invariant regression test (no Token::Text ever carries a literal newline).
• Validation fix — footnote syntax no longer trips the "unmatched square brackets" warning. The crude [ vs ] tally falsely flagged [^id] refs/defs, inline ^[…], and the intentional ^[ degradation; footnote brackets are now neutralized before the tally and a genuine [unclosed link still warns.
• Docs — new syntax documented in docs/configuration.md; doc filenames lowercased (docs/cli.md, docs/configuration.md, docs/library.md) and all inbound references updated.

Resolves #79, #80, and #82.

1.0.0 - 2026-05-15

First stable release — the genpdfi dependency is removed and the old pub mod pdf is replaced by a fully rewritten in-tree rendering engine built directly on printpdf 0.9. The library now owns the entire path from the lexer's token stream to PDF bytes, with a complete TOML configuration system, six themes, a redesigned public API, and frontmatter support.

• In-tree rendering engine (src/lib/render/) — four-stage pipeline: token stream → block IR → positioned page op-streams → printpdf 0.9 → an lopdf post-process pass for features printpdf doesn't expose (link tooltips, Catalog /Lang).
• Markdown rendering — headings 1–6 with anchors and PDF bookmarks; glyph-accurate line wrapping; inline bold/italic/monospace/strikethrough/underline/superscript/subscript/small-caps/small; ordered/unordered/GFM-task lists with arbitrary nesting and loose-vs-tight spacing; GFM tables with per-column alignment and header repeat across pages; blockquotes with configurable borders and cross-page backgrounds; fenced and indented code blocks; images (local PNG/JPEG, URL fetch and SVG rasterization behind features, HTML <img>, alignment, max-width, captions); GFM footnotes with bidirectional anchors and multi-line bodies; definition lists; cross-references ([text](#slug)); inline HTML mapped to run styles (<sup> <sub> <u> <s> <del> <small> <kbd>); hyphenation for overflow words; NBSP (U+00A0/U+202F/U+2007) respected by the wrapper.
• Document features — configurable page size/orientation and manual page breaks; headers/footers with page-number substitution and per-piece gap control; auto-generated TOC with clickable entries and convergent page numbering; title page (title/subtitle/author/date); PDF metadata (title/author/subject/keywords/creator) and Catalog /Lang for accessibility; inline link tooltips.
• Configuration system (styling/) — serde-derived deny_unknown_fields TOML schema with typo suggestions; cascade resolver (default theme → inherits chain → [defaults] → per-block → --theme); six bundled themes (default, github, academic, minimal, compact, modern); every visual choice user-overridable. Prose configuration guide and annotated reference config ship under docs/.
• Frontmatter — YAML / TOML parsed into document metadata.
• Public API — parse_into_file, parse_into_bytes, and symmetric parse_into_file_with_style / parse_into_bytes_with_style; path args accept impl AsRef<Path>; ConfigSource (Default / Theme(&str) / File / Embedded); MdpError::ParseError carries structured line + column. Page size configurable from TOML (#12); images embedded into the PDF (#13).
• Robustness — hostile/mistaken numeric config (zero/negative/NaN font size, line height, margins, custom page size) is clamped; bounded parser recursion returns a typed error instead of overflowing; linear-time table-start scan; U+0000 normalised to U+FFFD per CommonMark; real line/column in config errors; frontmatter tolerates a leading UTF-8 BOM. List bullets and task checkboxes are font-independent vector paths; underline/strike decorations darkened for legibility.
• Breaking — pub mod pdf removed (rendering is internal behind parse_into_* / render_to_*); MdpError::ParseError raw byte position became structured line + column; the fetch feature no longer needs a separate TLS feature (rustls-tls / native-tls removed) and svg is now opt-in; crate modernised to edition 2024, MSRV 1.85.
• Security — removing genpdfi (#31) drops encoding (RUSTSEC-2021-0153), lzw (RUSTSEC-2020-0144), rusttype (RUSTSEC-2021-0140) and stb_truetype (RUSTSEC-2020-0020) from the dep graph; font handling now uses ttf-parser. Downstream consumers can drop the corresponding cargo-deny exclusions.

Resolves #12, #13, and #31.

0.4.0 - 2026-05-13

Major lexer overhaul toward full CommonMark + GFM compliance.

• HTML handling — raw inline HTML, block-element blocks (<div>/<table>/<p>), standalone-tag blocks, raw-content blocks (<script>/<pre>/<style>/<textarea>), block comments, processing instructions, CDATA, and declarations.
• AST & resolution — stack-based emphasis resolution; wider Link/Image AST with parsed inline content and preserved titles; loose-vs-tight list detection; blockquote lazy continuation; entity decoding in link text / URL / alt and reference labels; full HTML5 named-entity table.
• Performance & correctness — linear-time emphasis; BOM stripping; tab-aware block-marker detection; many CommonMark spec gaps closed (60.6% → 89.0% spec pass rate).
• Packaging — missing build.rs / entities.json added to the package.
• Tests — reorganised under tests/markdown/ (~250 new cases) with a CommonMark spec runner and a stress suite; CI runs cargo test on every PR.

0.3.0 - 2026-05-10

Renderer breadth + CommonMark/GFM gap-closing.

• Renderer — inline images as links, real strikethrough, HTML-tag styling, soft-break newlines, blockquotes, and task-list checkboxes.
• Lexer coverage — stricter heading rules, lazy continuation, escapes, blockquote blocks, CRLF, hard breaks, entities, titles, reference links, mid-paragraph #, intra-word _.

0.2.2 - 2026-02-27

Built-in metric fallback + error-handling polish.

• Font metrics — embedded minimal TrueType font for built-in metric fallback; FontSource::bytes() constructor with documented priority.
• Errors — Pdf::new errors propagated through parse_into_file / parse_into_bytes; graceful fallback (no panics) on font-source loading failures.

0.2.1 - 2026-01-27

Bold / italic / underline / strikethrough styles applied to links.

0.2.0 - 2026-01-27

Tables, plus a font-system simplification.

• Tables — GFM table parsing and rendering with per-cell/header styling.
• Internals — font system simplified (removed fontdb) with a global cached font database; logging moved to the log crate; context-aware parsing via ParseContext; rendering and font-loading performance improved.

0.1.9 - 2025-11-14

Custom fonts and a pre-flight validation pass.

• Fonts — custom font loading with configuration options, alias/variant fallback chains, .ttc handling, and font subsetting (smaller PDFs); optional per-call font configuration on parse_into_*.
• Diagnostics — markdown-conversion validation system; richer MdpError diagnostics; CLI verbosity levels and --dry-run.

0.1.8 - 2025-10-07

Improved list parsing; JSON token-dump for debugging.

0.1.7 - 2025-09-21

Release automation (cargo-dist) and project housekeeping.

0.1.6 - 2025-07-21

ConfigSource enum with default / file / embedded configuration sources.

0.1.5 - 2025-07-16

parse_into_bytes for in-memory PDF generation; parse renamed to parse_into_file.

0.1.4 - 2025-07-09

Font discovery + OpenSSL polish.

• Fonts — comprehensive built-in and system font loading with caching and variant analysis.
• Build & UX — OpenSSL made optional; clearer CLI error guidance; assorted advisory cleanups.

0.1.3 - 2025-03-31

before_spacing text style; graceful handling of invalid image syntax; broader lexer/styling/config test coverage.

0.1.2 - 2024-12-01

URL input support for remote markdown files; CLI restructured.

0.1.1 - 2024-11-29

Hierarchical list rendering with proper indentation and mixed ordered/unordered nesting; embedded asset fonts.

0.1.0 - 2024-11-17

Initial release: a Markdown lexer and a genpdfi-backed PDF converter with basic styling, configuration via mdprc, code blocks, emphasis, links, and nested tokens.