mark2haru

mark2haru is a small standalone Markdown-to-PDF renderer for neutral brief-style documents. It is derived from the Markdown/PDF path used in briefutil, but it deliberately removes letter-generation semantics, profile handling, Qt UI code, and branding.

The name is a legacy nod to the original briefutil PDF path, which started life on top of libHaru. This project no longer depends on libHaru (or any external PDF library) — it contains its own minimal PDF writer and TrueType parser, and links only the bundled miniz for Flate compression.

What it supports

• paragraphs with source-line reflow (soft line breaks join as spaces)
• ATX headings (# through ######)
• bulleted and numbered lists with continuation lines
• fenced code blocks
• simple pipe tables that split across pages (headers repeat)
• standalone PNG image blocks (![alt](path.png))
• inline **bold**, *italic*, ***bold italic***, and `code`
• links rendered as text (url) so the destination is not lost
• thematic breaks (---, ***, ___, also spaced forms like - - -)
• explicit page breaks via <!-- pagebreak -->

What it does not try to do

• inline images inside paragraph text
• table column alignment from :---, ---:, or :---: separator markers
• full CommonMark compatibility
• font subsetting (see below)

The renderer embeds TrueType fonts into the PDF and emits Unicode text using Identity-H composite fonts. Input stays UTF-8, and Western European characters plus wider Unicode covered by the bundled fonts render correctly without depending on the viewer's installed fonts.

Only the font variants actually used by a document are embedded — a plain brief that uses only regular text does not pay for bold, italic, or mono font data. Each variant that is used embeds the full TrueType file, not a per-glyph subset: the /W widths array is trimmed to the used glyph set, but the glyf/loca/cmap tables are unchanged. True subsetting is a possible follow-up; it would significantly shrink the output for short documents.

Bundled fonts:

• DejaVuSans.ttf
• DejaVuSans-Bold.ttf
• DejaVuSans-Oblique.ttf
• DejaVuSans-BoldOblique.ttf
• DejaVuSansMono.ttf

These are copied next to the executable at build time and embedded into each generated PDF when that face is used. If you want a different font family, replace the files in mark2haru/fonts/ with compatible TrueType faces using the same names.

If the renderer cannot resolve or load the configured fonts, rendering fails instead of silently falling back to viewer-installed fonts. The library overload that accepts an error string reports that reason to the caller.

PDF content streams, font streams, and Unicode CMaps are Flate-compressed when compression makes the object smaller. The project uses the lightweight bundled miniz implementation, so it does not depend on system zlib.

CI Status

Platform	Build + test	Consumer install smoke	Notes
Linux	GitHub Actions	GitHub Actions	Configures with CMake, builds in Release, runs ctest, and validates an installed downstream consumer.
macOS	GitHub Actions	GitHub Actions	Configures with CMake, builds in Release, runs ctest, and validates an installed downstream consumer.
Windows	GitHub Actions	GitHub Actions	Uses the same CMake/CTest flow with the Release config on the hosted MSVC runner.

Build

From the repository root:

cmake -S . -B build
cmake --build build --config Release

On Windows the same two commands work from a Developer Command Prompt or PowerShell.

Any C++17-capable compiler and CMake 3.20+ should work.

Test

After building, the test suite covers:

• smoke rendering of the bundled examples, with PDF-header verification
• Markdown parser cases for escapes, links, lists, code blocks, tables, and explicit page breaks
• negative-path checks for malformed font tables and render/write failure reporting

Run the full suite with:

ctest --test-dir build --output-on-failure

Run

Render the sample brief:

./build/mark2haru examples/neutral_brief.md build/neutral_brief.pdf

Or render your own file:

./build/mark2haru input.md output.pdf

On Windows the binary is build\mark2haru.exe.

Library Use

The simplest library entry point is:

bool render_markdown_to_pdf(
    const std::string& markdown,
    const std::filesystem::path& output_path,
    const Render_options& options = {});

If you want the failure reason, use the overload with an output string:

std::string error;
if (!mark2haru::render_markdown_to_pdf(markdown, output_path, options, error)) {
    // inspect `error`
}

Render_options lets you control page geometry, body text sizing, line spacing, the asset base directory used for relative image paths, and the font root directory used during font resolution.

Installed CMake packages export mark2haru_FONT_ROOT, which points at the installed share/mark2haru data directory. Pass that value to Render_options::font_root_dir when using the bundled installed fonts from a CMake consumer.

Example

examples/neutral_brief.md shows a neutral EBICS-style brief with headings, paragraphs, lists, and a table.

examples/western_european.md is a focused sample with umlauts and accents for verification. It also includes Greek and Cyrillic to exercise Unicode text beyond Windows-1252.

Assumptions

• The output is intended for short-to-medium briefs rather than long books.
• The layout engine uses simple greedy wrapping and basic pagination.
• Tables use content-aware column widths. Long tables split across pages and repeat the header row.
• Standalone Markdown image blocks render PNGs. Relative image paths resolve against Render_options::asset_base_dir; the CLI sets that directory from the input Markdown file.
• Font coverage is defined by the bundled DejaVu family. Code points with no glyph in the selected face fall back to ? if the font has it, otherwise to the font's .notdef glyph.
• Stream compression is size-aware rather than forced; very small objects may remain uncompressed if that produces a smaller file overall.

Contributing

Contributions should follow the Varinomics coding style guideline. The canonical guideline is published by Varinomics at https://github.com/Varinomics/varinomics-standards; general project information is available at https://varinomics.com/.

Attribution

• Bundled DejaVu fonts are distributed under the DejaVu / Bitstream Vera license. See fonts/LICENSE.DejaVu for the local license text.
• third_party/miniz is vendored from https://github.com/richgel999/miniz. See third_party/miniz/LICENSE for the local license notices.