Build Samples Solution: transient yarn resolution failure (@swc/core-darwin-x64@1.16.22 registry propagation race) #678

What happened

The Build Samples Solution workflow failed on main at commit 4bf80a546e0b84808e07aae326e14f966856e4ba (merge of #677):

https://github.com/Cratis/Samples/actions/runs/42012017428

Every Frontend — * job failed in the yarn install step with:

➤ YN0082: @swc/core-darwin-x64@npm:1.16.22: No candidates found
➤ YN0000: · Failed with errors in 5s 43ms

Root cause

This repository intentionally carries no committed yarn.lock (**/yarn.lock is gitignored — see conventions.md), so every yarn install re-resolves live against whatever is newest on the npm registry instead of a pinned set of versions.

@swc/core@1.16.22 had been published only moments before the run (same-day nightly tag 1.16.22-nightly-20261105.1 plus the stable 1.16.22). Its optional platform package @swc/core-darwin-x64@1.16.22 had not finished replicating across the registry yet, so Yarn's resolution step couldn't find it — a transient propagation race, not a bad pin or a removed/unpublished version.

Verification

• npm view @swc/core-darwin-x64 versions now lists 1.16.22 as published and resolvable.
• A local yarn install against the same commit now resolves @swc/core@1.16.22 cleanly.
• Re-running the exact same failed run now succeeds: https://github.com/Cratis/Samples/actions/runs/42012017428 (conclusion: success). main is green again; no code or dependency change was needed.

Suggested follow-up (optional hardening)

Since there's no lockfile to pin against by design, this class of registry-propagation race can recur for any transitive dependency with per-platform optional packages. A small retry around the yarn install step in .github/workflows/build.yml (the frontend job) would let a future race resolve itself instead of failing the whole build, following the existing .github/scripts/push-with-retry.sh convention:

#!/usr/bin/env bash
# .github/scripts/yarn-install-with-retry.sh
set -euo pipefail

attempts=${YARN_INSTALL_RETRY_ATTEMPTS:-3}

for ((attempt = 1; attempt <= attempts; attempt++)); do
    if yarn install; then
        echo "yarn install succeeded on attempt ${attempt} of ${attempts}"
        exit 0
    fi

    if [ "${attempt}" -eq "${attempts}" ]; then
        break
    fi

    delay=$((attempt * 30))
    echo "yarn install attempt ${attempt} of ${attempts} failed - retrying in ${delay}s in case this is a registry propagation delay..."
    sleep "${delay}"
done

echo "::error::yarn install failed after ${attempts} attempts"
exit 1

...called from the frontend job instead of run: yarn install.

I'm not opening a PR for this: the GitHub App token available to me is blocked from creating or updating workflow files (refusing to allow a GitHub App to create or update workflow .github/workflows/build.yml without workflows permission), so a maintainer with the right permissions would need to add it.