[lisachenko]

What and why

ClassFieldAccess already creates its ReflectionProperty only when someone asks for it. This PR does the same for the method and function joinpoints, so their constructors no longer build reflection:

• DynamicTraitAliasMethodInvocation: the ReflectionFunction + ReflectionMethod pair that finds the original method body is now resolved on the first call that reaches the body (??= in proceed()), not in the constructor. getMethod() was already lazy.
• ReflectionFunctionInvocation: stores the function name and creates the ReflectionFunction in getFunction(). The original function is called through its first-class callable, so it never needs reflection to run.

If an advice never needs reflection (for example an around advice that returns a cached value without calling proceed()), no reflection is created at all.

ReflectionConstructorInvocation was tried too and then reverted (second commit). Creating the object always needs the class and constructor reflection, and the lazy version measured 20–25% slower on the first call when the constructor runs (8 of 8 interleaved processes), plus 3% on repeated calls. It stays eager, and src/Aop/AGENTS.md notes why.

Behaviour change on internal classes: a closure without a scope class, or an unknown function name, now fails on first use instead of at construction. The existing test was updated to match, and new tests check that reflection stays lazy.

Measurements

These numbers come from a sandbox, with master and the PR loaded in one process and timed interleaved.

First call (joinpoint creation + call), around advice:

Case	master	PR	Change
Method, around returns early	822 ns	512 ns	−38%
Function, around returns early	604 ns	510 ns	−16%
Method, around proceeds	971 ns	1034 ns	+6% (within noise)
Function, around proceeds	656 ns	561 ns	−14%

Repeated calls (composer test:performance, 5 alternating runs, median ratio against a plain call): every case matches master within ±0.3 and stays well within budget, e.g. method 10.6 → 10.9, function 10.9 → 11.1, method + around 14.2 → 14.2, function + around 16.2 → 16.3.

Checklist

• The title follows Conventional Commits (type(scope): summary)
• Tests cover the change
• composer check passes: phpstan and cs are clean, and tests/Aop/Framework passes. In this sandbox, functional/console tests fail and ConstructorExecutionTransformerTest segfaults the same way on master, so these are sandbox problems. CI runs the full suite.
• User-facing or breaking changes are in CHANGELOG.md

🤖 Generated with Claude Code

https://claude.ai/code/session_01YAJm7FNJjQyTtSYtFtJdMR

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.19%. Comparing base (0456abc) to head (fd4acdf).

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #772      +/-   ##
============================================
- Coverage     98.19%   98.19%   -0.01%     
  Complexity     1910     1910              
============================================
  Files           113      113              
  Lines          4999     4995       -4     
============================================
- Hits           4909     4905       -4     
  Misses           90       90              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.