[experiment] Benchmark pnpm vs Bun on CircleCI - #428

reviews:
  auto_review:
    drafts: true

_{Comment @coderabbitai help to get the list of available commands.}

Bun-only runtime benchmark result

Completed three real CircleCI Windows win/default medium runs with Node, npm, Corepack, and pnpm explicitly removed from PATH before the JS workload.

Every run independently asserted node, npm, corepack, and pnpm were unavailable.

Conclusion: Bun 1.4.2 can replace pnpm for install and can drive the production frontend build + Tauri CLI without Node, but Win-CodexBar cannot remove Node yet because the current Vitest/jsdom test contract is not compatible with Bun's runtime.

Additional disposable migration probe using Bun-native bun test + Happy DOM + a small vi compatibility shim improved discovery to 265 tests, but still left 79 failures + 10 errors. Concrete blockers include 40 vi.hoisted uses, 5 partial vi.mock(... importOriginal ...) factories, missing/incompatible vi.setSystemTime and other mock/timer/global helpers, plus accessor-spy differences.

This remains an experiment-only draft and should not be merged as-is.

Thermo-nuclear review: KEEP AS EXPERIMENT / DO NOT MERGE

That matches the PR's own stated intent. The benchmark harness is isolated from product runtime, but the current branch does not reproduce the full three-lane experiment described in the body:

• current head: experiment/bun-pnpm-circleci-20260906
• bun-only-runtime is filtered to that branch
• the three pm-benchmark jobs are filtered to experiment/bun-pnpm-circleci-completed

So a fresh run of this PR head exercises the Bun-only lane, not all three equal pnpm-11.24 / pnpm-11.25 / Bun lanes. If this PR remains as evidence, document the completed-branch runs/results explicitly. Do not merge the experiment-only bun.lock, bunfig.toml, benchmark scripts, or CircleCI routing into the product branch from this PR.

1k/CUA: no production code threshold or UI proof requirement applies.