markast

markast is a Rust Markdown renderer distributed through npm.

By default it outputs HTML fragments like marked. It can also output a full HTML document with built-in or custom styles.

Install

npm i -g markast

Usage

Render Markdown to HTML fragment (default):

markast README.md > out.html
cat README.md | markast

Render full HTML document with built-in theme:

markast --document --theme github README.md > page.html
markast --document --theme dracula README.md > page.html
markast --document --theme paper README.md > page.html

Apply custom style definition (JSON):

markast --document --theme-file theme.json README.md > page.html

theme.json format:

{
  "variables": {
    "--markast-bg": "#0f1115",
    "--markast-fg": "#f2f5f9",
    "--markast-link": "#65c1ff"
  },
  "css": ".markast h1 { letter-spacing: 0.02em; }"
}

Append extra CSS file:

markast --document --css ./extra.css README.md > page.html

Development

npm run check
npm run check:strict
npm run test:own
npm run test:compat:snapshot
npm run test:compat:runtime
npm run test:compat
npm run test:compat:report
npm run build

Parser engine: Current default and only production parser is the in-house markdown module (new parser pipeline), with no external markdown engine dependency in the main crate.

Development docs:

• docs/README.md
• docs/architecture.md
• docs/testing-and-compat.md
• docs/performance.md
• docs/requirements.md

Compatibility fixtures are synced under third_party/marked/test/specs.

Compat now has two layers:

• npm run check:strict: runs Rust compile/test gates with warnings denied.
• npm run test:compat:snapshot: gated comparison against vendored marked fixture/spec snapshots.
• npm run test:compat:runtime: gated comparison against the current vendored marked npm runtime.
• npm run test:compat:runtime-drift: auxiliary audit that checks whether snapshot-xfailed vendored fixtures still match the current runtime.
• npm run test:compat: runs both in sequence.

Known snapshot gaps are tracked in tests/compat/xfail.yaml. Known runtime gaps are tracked in tests/compat/runtime_xfail.yaml.

Refresh the snapshot xfail baseline after intentional parser behavior changes:

npm run test:compat:snapshot:update-xfail

Refresh the runtime xfail baseline after intentional parser behavior changes:

npm run test:compat:runtime:update-xfail

Compatibility Report

Current report date: 2026-03-16

This table compares the same parser-output cases from the official marked corpus under third_party/marked/test/specs.

Included in the same-case comparison:

• new + original fixture pairs: 153
• CommonMark JSON examples: 652
• GFM CommonMark mirror examples: 652
• GFM spec examples: 28
• Total comparable cases: 1485

Excluded from this table:

• third_party/marked/test/unit/*.test.js: 158 JS unit cases. These exercise Marked's JS API surface such as hooks, lexer/parser classes, CLI integration, and instance behavior, so there is no 1:1 Rust-side case mapping in markast yet.
• third_party/marked/test/specs/redos: 7 ReDoS fixtures. These are security/performance-oriented fixtures and are not currently part of the markast compat gates.

| Target | Case source | Passed | Gaps | Pass rate | | --- | --- | ---: | ---: | ---: | | marked self-spec result | vendored marked fixture/spec corpus | 1485 | 0 | 100.0% | | markast snapshot compat | vendored fixture/spec snapshots | 1327 | 158 | 89.4% | | markast runtime compat | current [email protected] runtime | 1479 | 6 | 99.6% |

How to refresh:

• npm run test:compat
• npm run test:compat:report

Benchmark

Reproduce locally:

npm install
npm run bench

The harness benchmarks shared Markdown corpora against five engines:

• markast through an in-process Rust benchmark binary
• pulldown-cmark through a benchmark-only comparison runner
• marked through marked.parse(...)
• markdown-it through markdown-it.render(...)
• remark through remark + remark-gfm + remark-html

CommonMark Core is the fairest suite for pulldown-cmark, because it runs the official CommonMark examples with gfm=false.

Raw data is written to bench/results/latest.json.

Performance strategy and optimization batches live in docs/performance.md.

pulldown-cmark is included as a throughput ceiling reference. markast is not expected to match its architecture or semantics in Phase 1. The pulldown-cmark comparator is kept outside the main markast crate so release/runtime dependencies stay focused on the in-house parser.

Benchmark date: 2026-03-13

Method: in-process render throughput on the same default-GFM corpus for all engines. Outputs are not normalized for semantic equality; this report only measures rendering speed on shared inputs. Trimmed mean ms drops one run from each side for 6-9 samples, or 10% from each side for 10+ samples.

Environment: Apple M4 | darwin 24.6.0 (arm64) | Node 22.22.1 | Rust rustc 1.93.0 (254b59607 2026-01-19)

| Suite | Docs | Input size | Warmup | Measured | Source | | --- | ---: | ---: | ---: | ---: | --- | | README.md | 1 | 7.2 KiB | 10 | 30 | Project README rendered as a single document | | CommonMark Core | 652 | 14.6 KiB | 4 | 10 | Official CommonMark 0.31.2 JSON examples rendered in non-GFM mode | | Marked Fixtures | 153 | 58.3 KiB | 4 | 12 | new + original fixture pairs from vendored marked specs | | Comparable Corpus | 1485 | 88.9 KiB | 4 | 12 | All 1485 comparable parser-output cases from vendored marked specs |

| Suite | Engine | Trimmed mean ms | Median ms | Docs/s | MiB/s | vs marked | | --- | --- | ---: | ---: | ---: | ---: | ---: | | README.md | markast (Rust) | 0.24 | 0.24 | 4116.9 | 29.02 | 1.64x | | README.md | pulldown-cmark (Rust) | 0.05 | 0.05 | 21819.0 | 153.79 | 8.71x | | README.md | marked (JS) | 0.40 | 0.39 | 2506.1 | 17.66 | 1.00x | | README.md | markdown-it (JS) | 0.47 | 0.46 | 2133.9 | 15.04 | 0.85x | | README.md | remark + gfm + html | 5.40 | 5.20 | 185.0 | 1.30 | 0.07x | | CommonMark Core | markast (Rust) | 1.23 | 1.20 | 529408.8 | 11.55 | 1.62x | | CommonMark Core | pulldown-cmark (Rust) | 0.52 | 0.51 | 1258143.2 | 27.46 | 3.85x | | CommonMark Core | marked (JS) | 1.99 | 1.93 | 327005.7 | 7.14 | 1.00x | | CommonMark Core | markdown-it (JS) | 2.33 | 2.26 | 280410.0 | 6.12 | 0.86x | | CommonMark Core | remark + gfm + html | 29.32 | 28.71 | 22234.4 | 0.49 | 0.07x | | Marked Fixtures | markast (Rust) | 3.24 | 3.17 | 47214.0 | 17.56 | 1.41x | | Marked Fixtures | pulldown-cmark (Rust) | 0.59 | 0.59 | 259931.5 | 96.69 | 7.77x | | Marked Fixtures | marked (JS) | 4.57 | 4.57 | 33466.2 | 12.45 | 1.00x | | Marked Fixtures | markdown-it (JS) | 3.69 | 3.64 | 41492.2 | 15.43 | 1.24x | | Marked Fixtures | remark + gfm + html | 47.08 | 46.84 | 3250.1 | 1.21 | 0.10x | | Comparable Corpus | markast (Rust) | 4.41 | 4.41 | 336770.0 | 19.69 | 1.90x | | Comparable Corpus | pulldown-cmark (Rust) | 1.23 | 1.20 | 1206282.9 | 70.52 | 6.80x | | Comparable Corpus | marked (JS) | 8.37 | 8.37 | 177508.4 | 10.38 | 1.00x | | Comparable Corpus | markdown-it (JS) | 6.94 | 6.90 | 213845.0 | 12.50 | 1.20x | | Comparable Corpus | remark + gfm + html | 140.03 | 140.38 | 10604.6 | 0.62 | 0.06x |

Raw benchmark data: bench/results/latest.json

Release

Push a semver tag like v0.1.0.

GitHub Actions workflow .github/workflows/release.yml will:

1. Build each platform binary.
2. Pack and publish platform npm packages.
3. Publish the main package markast.