jp-boxed-stream
Claim
Boxed Stream reduce (Integer::sum / similar) is far slower and more allocating than mapToInt/mapToLong on HotSpot.
How measured
- Source:
J32_RuleEvidenceBatch.jsonl - JDK 27 (27+35), JVM args
-XX:+UseCompactObjectHeaders - before:
boxedReduce_n10k - after:
mapToInt_n10k
Runtime (this evidence file)
| Field | Value |
|---|---|
| When | 2026-10-23 Asia/Shanghai |
| before | 745977.453 ns_per_op, 160178.813 B/op |
| after | 20050.141 ns_per_op, 207.420 B/op |
| ratio before/after | ~37.2× |
Lint mapping / policy
Policy error (wall-clock ~37×).
Limits
Microbench on this machine; see scripts/run-static-vs-hotspot.sh. Do not invent ratios. Corpus residual ≠ speedup.