A deep-dive comparison of two Python time-mocking libraries — time-machine and freezegun — explaining why time-machine is O(1) while freezegun is O(n). Benchmarks show freezegun's mock/unmock cycle grows linearly with the number of loaded modules (about 2.5 µs per module), while time-machine stays constant at ~1.5 µs regardless of project size. At 16,000 modules, time-machine is 27,300x faster. The difference comes down to approach: freezegun scans all sys.modules to rebind references, while time-machine directly overwrites C function pointers, affecting every reference at the C layer. The post also covers freezegun's leakiness (missed class attributes, closures, changed types) and introduces a migration CLI to automate switching from freezegun to time-machine.
