Hacktoberfest 2026:メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

perf: reduce per-row and per-access overhead in entity and model casting

オープン
#10,618 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
35/100
issue の種類
リファクタリング
明瞭さ
おおむね明確
活発さ
活発
技術スタック
php

調査の方向性

Start with DataCaster::castAs() for the cast-type parsing, DataConverter::reconstruct() for the duplicate syncOriginal() call after injectRawData(), and Entity::__get(), __set() and toArray() for the getter/setter lookup. Run the benchmark or the DataCaster, DataConverter and Entity tests before and after each of the three commits. Done means the listed tests pass, including the documented behavior change for entities that override injectRawData(), and the proposal's timing targets hold.

索引モデルが issue の本文から書いたものです。

説明

Problem

Loading rows through a model with casts, especially as entities, repeats the same work on every row and every field:

  1. DataCaster::castAs() parses the type string on every cast. For each value it runs str_starts_with, a match, a preg_match(), explode/array_map(trim) and a trim, all to turn something like '?int' or 'json-array' into a handler key and params. The handler itself often costs less than the parsing.
  2. DataConverter::reconstruct() calls syncOriginal() twice for each entity. It calls injectRawData(), which already calls syncOriginal(), and then calls syncOriginal() again. When attributes hold arrays or objects, each sync normalizes them and runs json_encode().
  3. Entity::__get() and __set() rebuild the getter or setter name on every access. They run str_replace/ucwords, then make two method_exists() calls. toArray() repeats this for every key.

Measurements

PHP 8.3 on macOS, OPcache on, production mode, warm. The test entity has 10 columns, 6 casts (int, ?int, bool, json-array, float, int-bool) and the default $dates.

Case Time
1000 rows → arrays (fromDataSource) 3.2 ms
1000 rows → entities (reconstruct) 6.3 ms
1000 rows → entity with no casts 2.0 ms
Step Cost
castAs() for an int field 0.33 µs
IntegerCast::get() called directly 0.07 µs
syncOriginal() with an array attribute 1.2 µs per call
Getter-name build + 2× method_exists() ~0.1–0.2 µs per access
__get() on a plain field 0.46–0.58 µs
toArray() with 10 keys 7.6–11 µs

Proposal

1. Cache parsed cast types (DataCaster)

Add a private static cache keyed by the type string. It stores only the parsed result: base type, params and the nullable flag.

  • Handlers are not cached. Each instance still looks up the handler in its own $castHandlers, so instances with different custom handlers keep working.
  • The non-strict 'nullable' param is still added after the cache lookup.
  • An unknown type still throws on every call.
  • Why static: each entity has its own DataCaster, so a per-instance cache would rarely get a hit. The parse depends only on the string, so the cache is safe across Worker Mode requests. It also can't grow without limit, because type strings come from code, not from input.
2. Sync originals once (DataConverter::reconstruct())

Remove the second syncOriginal() call. injectRawData() already syncs.

3. Per-class getter/setter map (Entity)

On the first access, build a map of each entity class's get*, _get*, set* and _set* methods using ReflectionClass::getMethods(), with lowercase keys. Lookups then replace the two method_exists() calls.

  • Bounded memory: the map depends on the class, not on input. Keys passed to fill() from request data can't make it grow.
  • Same lookup rules: lowercase keys match PHP's case-insensitive method names. Reflection includes private methods, so results match method_exists() exactly.

Estimated gains

Fixes 1 and 2 bring 1000 entities from about 6.3 ms to about 3.6 ms, roughly 40% faster. Array results drop from about 3.2 ms to about 1.7 ms. Fix 3 removes about 25–35% of the cost of each __get()/__set() call, which matters most in toArray() and in views that read many properties.

These are estimates from micro-benchmarks, not from the final code.

Behavior changes (4.8)

Only fix 2 changes behavior. If an app entity overrides injectRawData() without calling syncOriginal(), reconstruct() no longer syncs for it. Such an entity will report every field as changed after a find(), and save() will write all columns. That override is already broken for any other caller of injectRawData(). This will be listed in the v4.8.0 changelog under "Behavior Changes", with a note in upgrade_480.rst saying overrides must call syncOriginal().

Fixes 1 and 3 change no behavior.

Not changing

  • Lazy DataCaster creation in the Entity constructor (~5%). Subclasses can read the protected $dataCaster property directly and would get null.
  • The non-strict is_subclass_of() checks and the toArray() filter closure. Each costs under 2%.

Plan

One PR with three commits, one per fix. Fix 2 also includes the changelog and upgrade-guide entries.

Tests

  • DataCasterTest:
    • Two instances with the same type string but different custom handlers each call their own handler.
    • Non-strict ?type still adds nullable, and strict mode doesn't.
    • An unknown type throws on every call.
  • DataConverterTest:
    • A reconstructed entity reports hasChanged() === false.
    • An entity that overrides injectRawData() without syncing now reports changes. This test documents the behavior change.
  • EntityTest:
    • Getter lookup ignores case.
    • _getX wins over getX.
    • Inherited getters are found.
    • A private getter behaves as it does today.
    • Maps are kept separate for each class.

Related

  • #10615, #10616, #10617
主要言語
PHP
スター
6k
フォーク
2k
平均マージ
1日 22時間
マージ済み PR(30日)
59

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

codeigniter4/CodeIgniter4 のほかの issue

codeigniter4/CodeIgniter4 の issue をすべて見る

似ている issue

PHP の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。