Skip to content

benchmarks.md

Latest commit

 

History

History
60 lines (50 loc) · 3.1 KB

benchmarks.md

File metadata and controls

60 lines (50 loc) · 3.1 KB

Benchmark Records

This document records representative JMH runs for Taker. These numbers are not release guarantees; they are sanity-check baselines for spotting large performance or allocation regressions.

Run benchmarks from a clean target directory when comparing normal tests and JMH:

mvn clean
mvn -Pjmh test-compile exec:exec "-Djmh.prof=gc"

After running the JMH profile, run mvn clean before a plain mvn test. Generated JMH classes are written under target/test-classes and can confuse Surefire when JMH is no longer on the test classpath.

2026-06-13 Baseline

Context:

  • Java baseline: 21+
  • Mode: JMH throughput
  • Count: 15
  • Profiler: gc
  • Units: ops/ms for benchmark score, B/op for normalized allocation
  • Notes: recorded after zero-width not(...), standardized escaped literal expectations, and construction-time API guardrails.
Benchmark Score ops/ms Error Allocation B/op Notes
collectCharsLetters 17.618 0.071 72.389 Scanner-level character collection; low allocation.
collectExpectedFailures 11826.679 390.262 328.001 Failure collection path; expected labels are precomputed.
collectStringLetters 3.809 0.042 1007193.812 Repeated parser string collection; allocation-heavy.
countWhileLetters 17.730 0.076 88.387 Scanner-level count; low allocation.
foldRepeatedLettersAndDigits 2.125 0.061 1838051.246 Repeated combinator path; allocation-heavy.
listJoinLetters 3.161 0.006 1176186.184 List materialization plus join; allocation-heavy.
parseCsv 0.355 0.005 6671131.385 Full CSV materialization.
parseCsvCountFields 1.467 0.053 1776084.700 Combinator-based CSV field counting.
parseLocatedNumber 10660.301 706.496 168.001 located() adds about 72 B/op over parseNumber.
parseNumber 12774.003 556.058 96.001 Simple scalar parser; low allocation.
parseNumberAll 12657.793 51.092 96.001 EOF check adds no normalized allocation.
parseRepeatedLettersAndDigits 2.141 0.011 1849043.212 Repeated combinator materialization.
scanCsvCountFields 6.958 0.912 0.998 Hand scanner path; effectively allocation-free.
skipWhileLetters 18.761 0.022 72.365 Scanner-level skip; low allocation.
trimSpacesChar 16181.977 426.908 240.000 Space-only trim around a char parser.
trimWhitespaceChar 16810.068 61.322 240.000 Java-whitespace trim around a char parser.

Interpretation:

  • Scanner primitives (collectChars, countWhile, skipWhile, direct CSV scanning) are fast and low allocation.
  • Simple scalar parsers such as parseNumber are fast with stable low allocation.
  • located() has a small, predictable allocation cost.
  • Failure-heavy paths are cheaper than the previous baseline because exact character expectation labels are computed once during parser construction.
  • Repeated combinator materialization and full CSV parsing are intentionally more allocation-heavy; use scanner primitives for raw character runs and inspection workloads.