Problem
PR #332's performance claims (unbounded 61→16 ms, rate-limited 163→8.7 ms at 100k items, 444→162 B/item) were measured with a throwaway scratchpad harness. Nothing in the repo can reproduce them or catch a regression.
Proposal
Add a benchmarks/ project using BenchmarkDotNet:
- Scenarios: unbounded parallel, throttled (
maxConcurrency), rate-limited, timed rate-limited, batch, one-at-a-time, result streaming (GetResultsAsyncEnumerable) — each at 1k/10k/100k items with completed-task selectors (pure overhead) and Task.Yield selectors (dispatch cost).
[MemoryDiagnoser] for per-item allocation tracking.- A short doc section on how to run and how to compare against a baseline (
--filter, BenchmarkSwitcher).
- Optional follow-up: a manual CI workflow that runs the suite and posts results, so PRs with perf implications have numbers.
Acceptance criteria
dotnet run -c Release --project benchmarks produces the suite.- README or CONTRIBUTING points at it.
Problem
PR #332's performance claims (unbounded 61→16 ms, rate-limited 163→8.7 ms at 100k items, 444→162 B/item) were measured with a throwaway scratchpad harness. Nothing in the repo can reproduce them or catch a regression.
Proposal
Add a
benchmarks/project using BenchmarkDotNet:maxConcurrency), rate-limited, timed rate-limited, batch, one-at-a-time, result streaming (GetResultsAsyncEnumerable) — each at 1k/10k/100k items with completed-task selectors (pure overhead) andTask.Yieldselectors (dispatch cost).[MemoryDiagnoser]for per-item allocation tracking.--filter,BenchmarkSwitcher).Acceptance criteria
dotnet run -c Release --project benchmarksproduces the suite.