hive-ttracking.bench — io.github.hive-agi/hive-ttracking 0.1.4

BC-4 Bench — statistical benchmarking for Clojure code.

Wave-2 implementation: warmup + N timed iterations, percentile aggregation (p50/p95/p99/min/max/mean/stddev), :tt/bench-run event emission, and a defbench macro that wraps deftest with a p95 threshold assertion.

Design notes:

• Pure data in, pure data out. BenchResult is a value type.
• Event publishing delegated to hive-ttracking.events/publish, which swallows throwables so a stalled router never fails a bench.
• Warmup iterations run the thunk but are NOT included in samples — they let JIT compile hot paths before measurement.

aggregate-samples clj

(aggregate-samples {:keys [name runs warmup]} samples-ns)

Build a BenchResult from a seq of iteration wall-time samples (nanos).

bench* clj

(bench* {:keys [name runs warmup] :or {runs 30 warmup 3} :as opts} thunk)

Run thunk warmup times (discarded) then runs times (measured). Emits :tt/bench-run and returns a BenchResult.

opts: :name keyword/string — label (required) :runs long — measured iterations (default 30) :warmup long — warmup iterations (default 3)

The returned :samples-ns is a vector of raw nanosecond measurements in execution order — callers computing extra stats can reuse it.

defbench clj macro

(defbench bench-sym opts & body)

Define a clojure.test test that runs a bench and asserts a p95 threshold.

opts: :name keyword — bench label (defaults to test sym) :runs long — measured iterations (default 30) :warmup long — warmup iterations (default 3) :threshold-p95-ms double — fail if p95 exceeds this (required) :on-result ref/ns-qualified fn — side effect on BenchResult

Example: (defbench hot-path-bench {:runs 100 :threshold-p95-ms 5.0} (work!))

Expands to a tagged deftest — run via lein test :bench or equivalent.

IBencher clj protocol

aggregate clj

(aggregate this samples)

compare-benches clj

(compare-benches this base target)

run-iteration clj

(run-iteration this f)