[kamil-tekiela]

Sorting shouldn't fetch the weights at each comparison. It's better to sort the precomputed weights and then map them to the Test objects array. The new approach should theoretically be ~.5% faster, but it should also prevent accidental bugs if, at any point in time, weight is no longer a pure method. And if weight becomes more computationally expensive, the performance won't drop exponentially.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.57%. Comparing base (400384b) to head (b23197e).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@             Coverage Diff              @@
##               13.4    #7035      +/-   ##
============================================
- Coverage     99.57%   99.57%   -0.01%     
- Complexity     9907     9908       +1     
============================================
  Files           964      964              
  Lines         30083    30081       -2     
============================================
- Hits          29956    29954       -2     
  Misses          127      127              

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

[sebastianbergmann]

Thank you for this.

Since PHPUnit 13.2.5, the weight of a test suite is the sum of the durations of the tests it contains. ByDuration computes that sum inside the comparator, so usort() walks the entire subtree of a test suite every time it compares that test suite with another one. On a synthetic test suite with about 37,000 tests, sorting by duration takes 380 ms with the current code and 93 ms with your change.

I consider this a fix for a performance regression and would like to ship it in PHPUnit 13.4.2. Can you please rebase your branch onto 13.4, squash it into a single commit, and change the base branch of this pull request to 13.4? ByDuration.php is identical on 13.4 and main, so the rebase will not run into conflicts. I will merge the change up into main after that.