Performance Code Fixes

This document outlines performance-related issues found in test/property/foundation/beam/processes_properties_test.exs and provides concise recommendations for each.

1. Brittle Memory Assertion The test asserts a fixed 50% memory reduction, which is unreliable. Replace this with a baseline comparison that allows for minor, stable memory overhead.

2. Inaccurate Global Memory Measurement The test uses :erlang.memory(:total), which is noisy. It should measure process-specific memory with Process.info(pid, :memory) for accuracy.

3. Manual Latency Timing The message response test uses manual timing, which is imprecise. It should be replaced with a statistical benchmarking library like Benchee.

4. Inefficient Polling in Helpers TestHelpers.assert_eventually likely uses an inefficient polling loop. Replace its usage with event-driven checks using monitors and assert_receive.

5. Arbitrary Process.sleep(1) for GC A 1ms sleep is not guaranteed to be sufficient for GC to complete. Use a more reliable mechanism or a longer, configurable sleep time.

6. Global async: false The entire test file runs sequentially, slowing down the suite. Isolate tests that require sequential execution and enable async: true for all others.

7. Ineffective Fault-Tolerance Timing The crash-test property checks for survival but not recovery time. It should assert that the system is fully recovered within a defined SLO.

8. Unrealistic Worker Load The test workers perform no meaningful work, measuring only process overhead. They should simulate a more realistic workload for valuable metrics.

9. Arbitrary Scaling Factor The creation time test uses a lenient * 10 scaling factor. This should be a tighter, more statistically justified bound (e.g., * 2 for near-linear).

10. No Back-Pressure Testing The message response test sends a single burst of messages. It should also test system performance under sustained load and with growing message queues.

11. Hardcoded max_runs The number of runs for each property is hardcoded. This should be configurable to allow for quick developer runs and more thorough CI checks.

12. Inefficient Cleanup Fallback The cleanup_ecosystem function has a slow manual fallback. The primary shutdown_ecosystem function should be made robust enough to not need it.

13. Overly Lenient Timeouts Many tests use generous, fixed timeouts (e.g., 2000ms). These should be tightened to the maximum expected response time to fail faster.