Packages
26
Comparison with upstream
This page compares the plugins in this repository with the upstream markdown-it-* packages they are based on or correspond to: correctness of the rendered output, throughput and complexity scaling, and package size. All data comes from the committed comparison run; select a package to see its detailed benchmark, scaling, reference errors and differences.
Correctness
720 cases
493 exact45 normalized36 reference errors28 differences
Our test suites
961 / 1208passed
247 failed264 feature gaps
Benchmarks
186 benchmark cases
Measurement
2026-11-03
| Package | Correctness & tests | Throughput | Entry gzip |
|---|---|---|---|
| @mdit/plugin-abbr | 22/22vs19/2223/23 exact | 1.02× | 979 Bvs1.6 kB |
| @mdit/plugin-alert | 23/23vs8/234/25 exact4 errors4 diffs7 feature gaps | 1.00× | 1.4 kBvs1.9 kB |
| @mdit/plugin-anchor | 88/88vs70/8875/88 exact1 errors5 diffs4 feature gaps | 1.71× | 2.2 kBvs2.2 kB |
| @mdit/plugin-attrs | 616/616vs610/61619/20 exact1 diffs20 feature gaps | 1.03× | 4.6 kBvs2.6 kB |
| @mdit/plugin-cjk-breaks | 10/10vs10/1022/22 exact | 1.02× | 673 Bvs921 B |
| @mdit/plugin-cjk-friendly | 19/19vs19/1922/22 exact | 1.00× | 797 Bvs1.2 kB |
| @mdit/plugin-container | 32/32vs27/3220/23 exact1 diffs4 feature gaps | 1.00× | 1.1 kBvs1.4 kB |
| @mdit/plugin-dl | 14/14vs11/143/22 exact1 errors1 feature gaps | 1.00× | 978 Bvs1.9 kB |
| @mdit/plugin-emoji | 54/54vs49/5467/75 exact3 errors1 diffs2 feature gaps | 1.02× | 17.8 kBvs19.0 kB |
| @mdit/plugin-figure | 12/12vs5/1218/21 exact1 errors1 diffs27 feature gaps | 1.00× | 1.1 kBvs972 B |
| @mdit/plugin-footnote | 25/25vs5/256/22 exact1 errors | 1.06× | 1.7 kBvs2.7 kB |
| @mdit/plugin-img-size | 18/18vs14/186/25 exact1 errors3 diffs10 feature gaps | 0.98× | 1.8 kBvs1.6 kB |
| @mdit/plugin-ins | 12/12vs12/1223/23 exact | 0.98× | 187 Bvs1.1 kB |
| @mdit/plugin-katex | 17/17vs8/1710/25 exact5 errors2 diffs4 feature gaps | 0.96× | 605 Bvs2.0 kB |
| @mdit/plugin-katex-slim | 17/17vs8/1710/25 exact5 errors2 diffs5 feature gaps | 0.96× | 673 Bvs2.0 kB |
| @mdit/plugin-mark | 9/9vs9/923/23 exact | 0.99× | 185 Bvs1.1 kB |
| @mdit/plugin-mathjax | 36/36vs0/368/25 exact3 errors1 diffs38 feature gaps | 1.14× | 1.7 kBvs1.7 kB |
| @mdit/plugin-mathjax-slim | 36/36vs0/368/25 exact3 errors1 diffs38 feature gaps | 1.15× | 1.7 kBvs1.7 kB |
| @mdit/plugin-plantuml | 5/5vs2/56/24 exact1 errors1 diffs6 feature gaps | 6.96× | 749 Bvs1.5 kB |
| @mdit/plugin-ruby | 11/11vs5/1110/21 exact1 diffs1 feature gaps | 1.03× | 614 Bvs1.9 kB |
| @mdit/plugin-spoiler | 13/13vs6/1322/22 exact3 feature gaps | 0.99× | 235 Bvs1.7 kB |
| @mdit/plugin-sub | 9/9vs9/924/24 exact | 0.98× | 156 Bvs674 B |
| @mdit/plugin-sup | 10/10vs10/1024/24 exact | 0.98× | 155 Bvs674 B |
| @mdit/plugin-tasklist | 14/14vs0/147/21 exact1 errors1 diffs2 feature gaps | 1.18× | 731 Bvs1.4 kB |
| @mdit/plugin-tex | 66/66vs33/6618/25 exact4 errors1 diffs92 feature gaps | 1.01× | 1.7 kBvs2.0 kB |
| @mdit/plugin-uml | 20/20vs12/2015/25 exact2 errors2 diffs | 6.83× | 665 Bvs1.5 kB |
Methodology
- Correctness renders the same markdown with both implementations and compares the output byte for byte after a minimal normalization (attribute order; wrapper markup and generated ids only where both sides emit them by design). For variants without a custom renderer, the parsed token streams are compared as well — token metadata such as source maps is part of the public contract for plugins that build on top of these tokens. Every remaining difference is categorized as a reference error or an acceptable difference, grouped by root cause.
- Benchmark throughput and scaling numbers are medians of interleaved samples, so both sides run in the same process under the same load. Corpora are committed markdown files:
normalis a realistic document with sparse syntax,heavyis dense plugin syntax; each is measured at three sizes. - The test suite comparison runs each compared package's own unit tests against the reference implementation by swapping the built package entry for the reference package. Tests of features the reference does not provide are reported as feature gaps; tests that only assert internal implementation details are skipped silently and only counted in the aggregate; every other failing test is a behavior difference between the two packages.
- Each correctness difference is grouped under one root cause: "reference error" when the reference output or token stream is wrong (wrong source maps, missing or broken nesting, lost or misplaced content, security issues), and "difference" when the implementations merely represent the same result differently (equivalent token shape, alternative but valid output, intentional syntax or option design). Counts are per root cause, not per case.
- Scaling exponents are fitted on log-log axes: ~1 is linear and ~2 is quadratic. The series are built to expose the known superlinear paths of the reference implementations.
- Import cost is the gzipped size of the module graph reachable from the package's main entry (
exports['.']): the entry file plus every relative import it loads, concatenated and gzipped as one stream, i.e. what a consumer downloads and parses when importing the plugin. Bare (peer) imports are not counted for either side. Comparing a single bundled file with an unbundled reference entry would understate the reference badly, so both sides are measured over the whole graph; the unpacked size of the published files is recorded in the committed data but not shown. - Entry gzip is the gzipped size of the ESM entry: the resolved
exportsentry for the current package and themodule/exports.import/mainentry for the reference.
Tips
The comparison harness and the authored specs live in compare/. Run pnpm run compare to refresh the data, pnpm run compare:sizes to refresh the package sizes, and pnpm run compare:docs to regenerate this page's data.