Files
ArtPlayer/scripts/release/README.md
T

74 lines
4.1 KiB
Markdown

# Preparing exact npm candidate files
Run from the repository root with the pinned Node and Yarn:
```sh
yarn release:preflight --packages artplayer,artplayer-plugin-chapter
yarn release:bundle --packages artplayer,artplayer-plugin-chapter --tag next
yarn typecheck:release
yarn test:release-bundle
```
`release:bundle` does not build, pack, install, contact npm or publish. It is a
handoff step for candidates already recorded in `refactor/release-ledger.json`.
Both commands currently refuse release acceptance because the repository still
has unfinished candidate, review, device and other required gates. Synthetic test
reports are not evidence that an ArtPlayer package is ready to publish.
## Responsibilities
- `../prepare-release.mjs` is the Node entry and failure exit-code boundary.
- `prepare.ts` parses the explicit package batch/tag, requires the canonical
Node/Yarn, and checks clean Git state before both ledger reads. Ignored generated
files may exist; staged, unstaged and untracked source changes are rejected.
- `bundle.ts` consumes the repository ledger, checks the batch, copies exact
tarball bytes and writes a hash-bound manifest. Its injected inspector is an
internal test seam; the CLI never trusts a supplied JSON report as approval.
- `../../refactor/scripts/release-ledger.mjs` remains the source of candidate,
input fingerprint, review, risk, source/license, device, rollback and CI checks.
Do not introduce another manually maintained green-status list here.
The narrow report interfaces describe the fields consumed by preparation. The
original report is preserved in full. The candidate assertion after batch
validation is the only non-null assertion; an absent or invalid candidate must
be rejected before any output directory is created.
## Output and failure behavior
An accepted run creates a fresh `refactor/.cache/npm-bundle-*/` directory with
the existing tarballs, `preflight.json` and `manifest.json`. No existing output is
overwritten. The manifest binds package names, versions, per-package source
commits, source fingerprints, SHA-256, SHA-512 integrity, batch source commit,
toolchain/lock information and the complete preflight report's hash.
The second ledger read must equal the first; candidates and copied files are
checked again. The manifest completion marker is written last. On failure, only
the verified temporary directory is removed. Redirected paths are rejected; if
cleanup also fails, both errors and the retained directory are reported. A
leftover directory without a manifest is not a successful handoff.
Allowed tags are explicitly `next`, `alpha`, `beta`, `rc` and `latest`; there is
no implicit default, and prereleases cannot use `latest`. Registry metadata is
fixed to the public npm registry. Site distributions cannot enter this npm bundle.
These are preparation-tool constraints, not changes to any player package API.
## Trust boundary and remaining workflow work
Every output retains `publicationAuthorized: false`. A manifest and its hashes
provide content binding, not authenticity or publishing permission. CI-03 still
needs trusted workflow/run artifact provenance, registry occupancy checks, exact
tarball publication without lifecycle rebuilding, permission/OIDC configuration,
partial-failure recovery and post-publication readback. Neither this script nor
its tests enables a GitHub workflow, creates a token, or authorizes publication.
The future publisher must validate downloaded contents and fresh release evidence;
it must not trust an artifact solely because it has this manifest shape. Do not
add a rebuild fallback to preparation or publishing when an artifact is missing.
`release-bundle.test.mjs` uses small real tar archives with explicitly synthetic
ledger reports for byte-copy/failure tests, and temporary real Git repositories
for the clean-source guard. It covers stale inputs, changed outputs, missing
gates, site confusion, selection/tag errors, outside paths/junctions, partial
cleanup and original exception retention. These tests run in `test:baseline`;
strict TypeScript checking runs in `ci:check`, and the TS source is in root lint.