Benchmark

Micro-benchmark comparing sequential await vs Promise.all for the two parallelization patterns in this PR.

This is a synthetic benchmark to demonstrate the theoretical benefit of the patterns, not an end-to-end measurement. In current rstest workloads, the parallelized operations use Rspack's in-memory filesystem (delay ≈ 0), so the improvement is negligible. However, as project count, test file count, or I/O latency increases (e.g., real disk I/O via writeToDisk, network FS, or complex config resolution), the benefit scales accordingly.

Pattern A: two independent awaits → Promise.all

Applies to getAssetFiles + getSourceMaps parallelization (pool/index.ts, runTests.ts, listTests.ts).

• count = number of test files (each triggers one pair of calls)
• delay = per-call I/O latency (0 ≈ in-memory FS, >0 ≈ real disk)

+-------+----------+----------------+----------------+-----------+
| count | delay_ms | sequential_ms  | parallel_ms    | speedup   |
+-------+----------+----------------+----------------+-----------+
|    10 |     0.00 |           0.00 |           0.00 |    0.52x  |
|    10 |     0.05 |          23.13 |          11.48 |    2.01x  |
|    10 |     0.50 |          23.16 |          11.53 |    2.01x  |
|   100 |     0.00 |           0.02 |           0.03 |    0.54x  |
|   100 |     0.05 |         228.08 |         114.39 |    1.99x  |
|   100 |     0.50 |         227.79 |         114.55 |    1.99x  |
|   500 |     0.00 |           0.12 |           0.20 |    0.60x  |
|   500 |     0.05 |        1135.12 |         570.36 |    1.99x  |
|   500 |     0.50 |        1137.39 |         567.81 |    2.00x  |
+-------+----------+----------------+----------------+-----------+

Theoretical ceiling: 2x (two independent operations fully overlapped).

Summary

Pattern B: for loop → Promise.all map

Applies to collectTestFiles, collectProjectEntries, resolveProjects (listTests.ts, hostController.ts, cli/init.ts).

• count = number of projects
• delay = per-project async work (glob, config resolution, etc.)

+-------+----------+----------------+----------------+-----------+
| count | delay_ms | sequential_ms  | parallel_ms    | speedup   |
+-------+----------+----------------+----------------+-----------+
|     3 |     0.00 |           0.00 |           0.00 |    0.31x  |
|     3 |     0.50 |           3.44 |           1.15 |    3.00x  |
|     3 |     2.00 |           6.84 |           2.30 |    2.98x  |
|     3 |    10.00 |          33.14 |          11.04 |    3.00x  |
|    10 |     0.00 |           0.00 |           0.00 |    0.54x  |
|    10 |     0.50 |          11.58 |           1.17 |    9.90x  |
|    10 |     2.00 |          22.81 |           2.30 |    9.91x  |
|    10 |    10.00 |         110.39 |          11.08 |    9.96x  |
|    50 |     0.00 |           0.01 |           0.01 |    0.61x  |
|    50 |     0.50 |          56.56 |           1.18 |   47.93x  |
|    50 |     2.00 |         113.04 |           2.28 |   49.53x  |
|    50 |    10.00 |         548.19 |          11.05 |   49.61x  |
+-------+----------+----------------+----------------+-----------+

Theoretical ceiling: Nx where N = project count (all projects processed concurrently).

Background

Startup and test discovery still serialized several independent async tasks, which added avoidable latency to CLI setup, project discovery, and asset preparation.

Implementation

• Parallelized browser and core project discovery while preserving result order.
• Overlapped asset and source map loading for global setup and worker asset filtering.
• Started independent CLI and coverage module imports earlier so config I/O and module loading can run concurrently.

User Impact

Large multi-project runs, list commands, and startup paths spend less time waiting on serialized async setup.

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).

Validation:

• pnpm format attempted, but Biome hit an existing internal panic in packages/core/src/types/coverage.ts before completing.
• pnpm run lint
• pnpm run typecheck
• pnpm run build
• pnpm run test
• cd e2e && pnpm test