fix(scripts): narrow napi targets instead of staging placeholder binaries

[popescuoctavian]

Fixes the Regression benchmark job on main, broken since #1650 bumped
@napi-rs/cli to 3.8.6 (first surfaced on #1648's run, the first to actually
execute the job after the bump):

Internal Error: Release package @nomicfoundation/edr-darwin-arm64 is incomplete: missing edr.darwin-arm64.node

@napi-rs/cli >= 3.8 validates during pre-publish that every target in
napi.targets has its .node binary staged — unconditionally, regardless of
--skip-optional-publish. #1650 fixed this for the release workflow by moving
pnpm artifacts before prepublish.sh, but the Verdaccio flow only ever
builds one binary, so validation fails on the first missing platform.

Fix: stage empty placeholder .node files for the platforms that weren't
built, run prepublish.sh unmodified — so pre-publish sees the exact same
configuration as a real release — then delete the placeholders. They can never
be published: prepublish.sh passes --skip-optional-publish, the script only
publishes the host's platform package, and they're removed right after
pre-publish returns. The existing optionalDependencies prune loop is kept
unchanged. This avoids hardcoding the platform→triple mapping and keeps the
Verdaccio route as close as possible to the real publish flow.

Verified locally against @napi-rs/cli 3.8.6: without placeholders,
pre-publish reproduces the exact CI error; with them it succeeds, and the
resulting package state (versions, optionalDependencies) is identical to
what the script produced before the CLI bump.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 1cb634f

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.01%. Comparing base (617568b) to head (1cb634f).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1664      +/-   ##
==========================================
- Coverage   80.02%   80.01%   -0.01%     
==========================================
  Files         464      464              
  Lines       80392    80392              
  Branches    80392    80392              
==========================================
- Hits        64332    64325       -7     
- Misses      13854    13860       +6     
- Partials     2206     2207       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Wodann]

Is there a fix for this problem that:

1. doesn't duplicate the supported platform "triples" into this script; and
2. keeps the verdaccio publish route as close as possible to the real publish script?

[popescuoctavian]

@Wodann I've pushed a different approach, which adds placeholders for missing binaries. This keeps the implementation closer to real publishing.

[Wodann]

Alternative: narrow napi.targets for this one invocation

Pushed as b9a65f4.

There's no flag to skip the pre-publish validation — prePublish() loops over every target from readNapiConfig and calls validateReleasePackageContents unconditionally (src/api/pre-publish.ts:342), and none of the documented options (--skip-optional-publish, --dry-run, --npm-dir, …) gate it. But --config-path overrides napi.targets for that invocation, so the other platforms are never considered in the first place.

Change

• scripts/write_napi_host_config.ts writes a config naming only this host's target, looked up in napi.targets via napi's exported parseTriple — so adding a platform still needs no change here. It follows the Hardhat repo's precedent of running TypeScript scripts directly with node.
• scripts/publish_to_verdaccio.sh calls it into a mktemp file (removed by an EXIT trap), checks the file is non-empty, and passes --config-path.
• scripts/prepublish.sh forwards "$@" to napi pre-publish. The release workflow calls it with no args, so that path is unchanged.
• scripts/tsconfig.json + tsc:scripts / test:scripts give scripts/ the type-checking, formatting and test coverage Hardhat's scripts have. scripts/package.json marks the directory ESM — the repo root can't be type: module because config/eslint/eslintrc.js and .syncpackrc.js are CommonJS.
• engines.node moves to >=22.18.0, the first release where Node runs TypeScript without a flag. Every CI job is already on 24.

Why this over placeholders

• No fake .node files on disk. Validation currently only does existsSync, but it already reads binary contents for wasm targets, so empty placeholders are betting on that not spreading to node targets.
• The npm pkg delete optionalDependencies.… pruning loop becomes unnecessary — resolveRootOptionalDependencies wires exactly the configured targets, so only this host's is ever added.
• Smaller blast radius on the working tree: 2 tracked files modified (root + host platform package.json) instead of 8 — with the full target list, napi version-syncs all 7 npm/*/package.json even though 6 are never published.

Tests (24, ~0.8s, no Rust build and no Verdaccio needed)

• The suffix→target lookup, and the emitted config's exact bytes.
• The CLI's exit codes and its refusal to write a config when no target matches.
• The emitted config fed to napi's own readNapiConfig, asserting binaryName survives — see caveat 1.
• The invariant that napi.targets and crates/edr_napi/npm/ stay in sync, generated from the real directory listing, so adding a platform to one and not the other fails with a named test.

Two things worth a look in review:

1. napi prints

   Both napi field in crates/edr_napi/package.json and NAPI-RS config file are found, the NAPI-RS config file will be used.

   The message overstates it — the merge is a shallow Object.assign over pkgJson.napi (src/utils/config.ts:283), so binaryName and everything else survive and only targets is overridden. That's load-bearing, so there's a test pinning it: if an upgrade made it a true replacement, binaryName would silently fall back to "index" and pre-publish would look for the wrong .node.

2. scripts/package.json is a 5-line ESM marker, not a workspace package (the pnpm globs don't cover scripts/, and it's private). Without it, node scripts/*.ts warns MODULE_TYPELESS_PACKAGE_JSON on every run.

Unrelated sharp edge found while testing: after running publish_to_verdaccio.sh locally, crates/edr_napi/package.json is left at the local version with an injected optionalDependencies, and pnpm run lint is green on that state — no prettier glob or syncpack rule covers crates/**/package.json. A git commit -a would ship a version downgrade of the published package. The script's header does say to run git checkout -- crates/edr_napi afterwards; might be worth making that harder to miss.

[Wodann]

/bench

[github-actions]

🚀 Starting regression benchmark for 1cb634f10f69 against Hardhat main (benchmarks matching test solidity,test mocha,test vitest).

[github-actions]

✅ Regression benchmark passed for 1cb634f10f69 against Hardhat main.

View workflow run