MoonBit Toolchain Performance Dashboard

An English, static GitHub Pages dashboard that tracks MoonBit toolchain performance across Linux, Windows, and macOS. It has no backend service: GitHub Actions produces versioned JSON files under public/data/, and Vite publishes them with the ECharts frontend.

The dashboard is intended for trend detection, not a performance gate. GitHub-hosted runner hardware is shared infrastructure and can be noisy, so each cell reports all 30 raw observations alongside min/median/max.

Benchmark protocol

• At the start of each workflow, the latest commit on moonbitlang/core is resolved once. All three runners then check out that same revision.

• The scheduled workflow runs every day at 02:00 UTC (10:00 CST) and supports manual dispatch.

• Runner labels are fixed to ubuntu-24.04, windows-2025, and macos-15.

• Backends are wasm, wasm-gc, js, and native.

• Stable is installed from latest; pre-release is installed separately. If their parsed moonc -v identities match exactly, nightly becomes the candidate. Otherwise pre-release is the candidate.

• Every OS/backend/toolchain unit runs 30 independent moon processes. There is no warm-up, each process receives a fresh target directory, cleanup happens before timing, and the first toolchain alternates by iteration.

• The measured command is:

  moon check --target <target> --target-dir <fresh-dir> --frozen --quiet
  

• MOONC_RC_CONVENTION=borrow matches core's regular CI convention. Each process has a 120-second timeout.

• Delta is (candidate median - stable median) / stable median × 100%; negative values mean the candidate was faster.

A fully successful run contains 24 benchmark units, 720 raw samples, and 12 stable/candidate comparisons. Failed, timed-out, or unavailable units are retained with a truncated error summary and exit code. They never produce a delta. The data is still committed before the workflow reports failure.

Static data API (schema V1)

The only public API is JSON served with the site:

• public/data/index.json contains newest-first RunSummaryV1 objects. It is sufficient for the latest overview, matrix, and all trend charts.
• public/data/runs/<workflow-run-id>-<attempt>.json contains one complete BenchmarkRunV1: exact toolchains, runner inventory, raw samples, protocol, errors, and workflow metadata.

Every document has schemaVersion: 1. Runtime validators are shared by the collector, publisher, and browser. The publisher validates all three OS artifacts, validates the resulting run and index before writing, sorts history by completion time, and deduplicates by workflow run ID plus attempt.

Unit status is one of ok, failed, timeout, or unavailable. History is retained indefinitely.

Local development

Requirements: Node.js 22 or newer and npm.

npm ci
npm run dev

Verification:

npm run typecheck
npm test
npm run validate:data
npm run build

The production build uses base: "./", so assets and JSON work from the /toolchain-performance-dashboard/ project path as well as local static servers.

Run the collector manually

The collector downloads official MoonBit installers and writes into isolated MOON_HOME directories. It does not modify the current shell's toolchain.

npm run benchmark -- \
  --os ubuntu \
  --core-dir /path/to/core \
  --output artifacts/shard-ubuntu.json

The core directory must be a Git checkout because the collector records its exact HEAD commit. Aggregation expects one valid shard for each OS, all collected from the same core revision:

npm run aggregate -- --artifacts-dir artifacts --data-dir public/data

Repository setup

Before enabling automation:

1. In Settings → Pages, set the source to GitHub Actions.
2. In Settings → Actions → General, allow workflows to read and write repository contents.
3. Ensure branch protection permits the benchmark workflow's GITHUB_TOKEN to push data commits to main, or provide an equivalent approved integration.
4. Run the Benchmark workflow manually once. Confirm 24 units, 720 samples, one data commit, and a Pages deployment at https://moonbit-community.github.io/toolchain-performance-dashboard/.

The project intentionally has no alerting, regression gate, or history retention limit.