perf(autoloader): optimize package list accumulation in loop

[gr8man]

Description
In Autoloader::loadComposerNamespaces(), array_merge() was being used inside a foreach loop to accumulate composer packages into $packageList:

foreach ($allData as $list) {
    $packageList = array_merge($packageList, $list['versions']);
}

Complexity Analysis:

• Before (array_merge inside the loop):
  On each iteration $i$, array_merge allocates a completely new array and copies all existing elements from $packageList (which has size $i \times S$, where $S$ is the size of each package version list) plus the new ones.

  • Time Complexity: $\sum_{i=1}^{K} O(i \times S) = O(K^2 \cdot S) = O(N^2)$ (quadratic) where $N$ is the total number of packages across all lists.
  • Memory Complexity: $O(N)$ allocations and copies at each step, generating excessive garbage collection overhead.

• After (Nested foreach loop with direct assignment):

  foreach ($allData as $list) {
      foreach ($list['versions'] as $packageName => $data) {
          $packageList[$packageName] = $data;
      }
  }

  The nested loop directly assigns elements to $packageList one by one. This preserves the exact behavior of array_merge() where duplicate string keys (package names) in subsequent array elements overwrite previous ones, while completely avoiding intermediate array copying.

  • Time Complexity: $\sum_{i=1}^{K} O(S) = O(K \cdot S) = O(N)$ (linear).
  • Memory Complexity: $O(1)$ additional overhead per element, performing in-place addition/overwrite of keys.

This PR shifts the package list building process from quadratic time $O(N^2)$ to linear time $O(N)$ while maintaining 100% behavioral compatibility with the original array_merge logic.

Additionally, a unit test was added to verify the package list auto-discovery flow inside loadComposerNamespaces.

Checklist:

• Securely signed commits
• Component(s) with PHPDoc blocks, only if necessary or adds value (without duplication)
• Unit testing, with >80% coverage
• User guide updated
• Conforms to style guide
• We don't add entries for refactoring, as there are no real changes for end users. (No changelog entry needed)

[memleakd]

Thanks for the PR. I have to say I respect the ambition here. Not every foreach gets a TED Talk.

The implementation itself looks fine to me, but I think the performance framing is doing more work than the code change needs. I'd trim the PR body quite a bit and frame this as a small internal cleanup.

I'd also consider dropping the new test. It reads a bit like assertComposerStillComposers(), while existing Composer discovery tests already cover that observable behavior.

[michalsn]

Thanks, but I don't think this optimization is justified.

InstalledVersions::getAllRawData() typically returns a single dataset and returns more only when multiple independent Composer autoloaders are registered in the same process. Therefore, array_merge() is normally called only once and is likely faster than a PHP-level nested loop.

A benchmark showing a real-world improvement would be needed to justify this change.