Investigate slow array-backed object access in jperl

已關閉
  • 哪裡不對？
#1,854 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

維護者通常 1 天內回覆

已經有一個關聯 PR 被合併了。

  • #1876 來自 @fglock —— 已合併

評估

難度
5/5
預估耗時
一週以上
新手友好度
38/100
Issue 類型
缺陷
描述清晰度
基本清楚
活躍度
活躍
技術堆疊
java, perl
領域
performance

研究方向

Start with dev/bench/benchmark_method.pl and run the hash-backed and array-backed variants in fresh PerlOnJava processes, as described. Then profile the method-call workload and trace generated JVM code and runtime helpers for field reads and updates; compare JVM and interpreter backends. Done means the performance gap is localized, with a focused benchmark or regression check added and an optimization that preserves Perl semantics.

由索引模型根據 Issue 內容生成。

描述

Summary

Investigate why array-backed objects are substantially slower than hash-backed objects in PerlOnJava for repeated method calls and field updates. The same benchmark shows a small array advantage in system Perl, so this appears to be a PerlOnJava implementation-path difference rather than an inherent cost of Perl array objects.

Reproduction

Run dev/bench/benchmark_method.pl, which performs 500 Benchmark iterations of a method called 20,000 times per iteration. Foo::add reads and increments $self->{x} and $self->{y}, then returns their sum.

For the array-backed variant, replace construction with bless [ 1, 2 ], $class, and change the field accesses to $self->[0] and $self->[1].

Both variants completed successfully and printed the same final value: done 399600080000.

Observed performance

Reported by Perl's Benchmark::timethis (iterations/second; larger is faster):

Runtime Hash-backed Array-backed Difference
System Perl 329–330/s 359–360/s array ~9% faster
PerlOnJava (jperl) 264–268/s 89–91/s array ~66% slower

The PerlOnJava result was repeated twice after the initial run and was consistent. Each PerlOnJava run was executed in a fresh process.

Investigation goals

  • Profile the benchmark to separate method dispatch, argument setup, scalar arithmetic, hash access, and array access costs.
  • Trace generated JVM code and runtime helpers for $self->{key} versus $self->[index], including lvalue fetch/store paths.
  • Determine whether array access allocates or copies wrappers, performs redundant bounds/type checks, or takes a slower generic path.
  • Compare JVM and interpreter backends to localize the issue.
  • Add a focused benchmark or regression performance check once the hot path is understood, then optimize without changing Perl semantics.

Caveats

This is a microbenchmark of the complete method-call path, not an isolated container benchmark. The result does not prove hash access is optimal; it shows that this workload currently favors hash-backed objects in PerlOnJava. More controlled access-only benchmarks and profiling are needed before attributing the gap to a specific implementation component.

主要語言
Perl
星號
65
分支
8
平均合併
5 小時 45 分鐘
30 天內合併 PR
184

環境準備

  • 提供 Dockerfile 或 Docker Compose 檔案
  • 沒有 Pull Request 範本
  • 閱讀貢獻指南

從這裡開始

  1. 先讀完整個 Issue，再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫，在一個分支上完成修改。
  4. 送出 Pull Request，並在描述裡引用這個 Issue 編號。

fglock/PerlOnJava 的其他 Issue

查看 fglock/PerlOnJava 的全部 Issue

相似的 Issue

更多 Perl Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。