Mark: StrictMark to HTML, with enforced limits

Mark is the wiki dog: its purview is rendering StrictMark to HTML and linting it; it owns no persistent data, only the .html it emits. The goal is a lightweight, greppable wiki whose structure and size limits are enforced at build time, no host renderer. The method: drive MKDT for structure and inline tokens, generate HTML via u8b feed helpers, sanitize text and decompose links through two ragel machines, and check each page against the WikiWeb budgets — warn by default, fail under --strict.

Each page is parsed by MKDT, not by mark: line classifiers give block structure, MKDTInlineLexer gives inline G tokens.
Generation is u8b feed only: MARKu8bLit for tags, MARKu8bFeedEsc for text — mark never reparses Markdown by hand.
The MARKE ragel machine escapes & < > " to entities; MARKG splits a G token into kind, text, and label.
A link target ending .mkd is emitted as .html; reference defs are collected in a first pass and never rendered.
Budgets in codepoints: header 64, bullet 128, summary 256, opener 512, with exactly one H1 opener per page.

API

Mark's public surface is small: one render entry, one inline decomposer, and the two u8b emit helpers the renderer feeds through.

MARKRenderDoc — render a whole StrictMark document to HTML, enforcing the WikiWeb budgets as it goes.
MARKDecomposeG(token) — split a G inline token into kind, text, and label (strong/emph/del/link/image).
MARKu8bLit — feed a literal HTML tag run into the output buffer verbatim.
MARKu8bFeedEsc — feed text into the output buffer with the four HTML-special bytes escaped.

CLI

mark takes files or a directory; a file renders to a sibling .html, a directory renders every .mkd in it, so a whole wiki builds in one call.

    mark page.mkd           # -> page.html beside it
    mark --strict wiki/     # build a dir, fail on any budget breach
    mark a.mkd b.mkd        # several files at once

Sanitizing and links

All text reaches HTML through ragel, never a hand-rolled byte scan, and inter-page links are rewritten so the rendered site is self-contained.

MARKG decomposes *strong*, _emph_, ~~del~~, the two link cases [text][l] / [page], and ![alt][l].
A shortcut [page] keys on its bracket text; an explicit [text][l] uses a one-symbol label l.
A shortcut [Page] resolves against the collected definitions, then a .mkd target becomes .html.

Enforced limits

mark is also the wiki linter: it checks the WikiWeb page shape and the 64-codepoint-line budgets, reporting any breach. --strict turns the first breach into a non-zero exit for CI.

Budgets in codepoints: header 64, bullet 128, summary 256, opener summary 512.
Structure: exactly one H1 opener per page, since each page covers one concept.
Default warns to stderr and still renders; --strict fails with MARKLIMIT.

Build and test

mark is a standard Dog: a static library plus a CLI, with table-driven tests. The two ragel units are pre-generated and committed alongside their sources.

Sources: MARK.c driver, MARKE.c.rl escaper, MARKG.c.rl decomposer, MARK.cli.c entry.
Regenerate ragel with ragel -C MARKE.c.rl -o MARKE.rl.c -L (likewise MARKG).
Tests in test/MARK.c render snippets and assert HTML; repros for past lexer gaps live there.