Deterministic profiling

Opening your lesson

A profiler can show that a normalizer ran thousands of times even when each call looked harmless in isolation. Its call counts help you find repeated work. Its timings need more care, because instrumentation changes the program it measures.

You run a real cProfile probe, inspect supplied exclusive-cost evidence and count a call tree. Keep the measurement question narrow. A high total time may belong to a function that mostly waits for its children, while a smaller total can hide expensive work performed directly. The exercises avoid timing assertions so a busy laptop and a quiet server produce the same checked answer. Count empty inputs and repeated calls too.

import json

def process_payload(payload: str) -> str:
    data = json.loads(payload)
    import cProfile
    profile = cProfile.Profile()
    def normalize(value):
        return value.strip().lower()
    profile.enable()
    for value in data:
        normalize(value)
    profile.disable()
    result = sum(entry.callcount for entry in profile.getstats()
                 if getattr(entry.code, 'co_name', '') == 'normalize')
    return json.dumps(result)

Profiler call counts include every invocation, even repeated or empty inputs. Define the named function inside the measured scope, then enable the profiler around the calls. Call counts are deterministic; the profiler's timing overhead makes this unsuitable as a speed benchmark.

Choose your character

Your Work Gym account