为什么python timeit结果与shell的time命令相矛盾？

Question

I'm Not quite sure what I'm missing here, but looks like time from shell reports higher values for commands that timeit from python report less values for. And practically looks like the time numbers are correct. What is happening?

time python -m timeit 'd={"i": 1}; d.update({"i": 2, "j": 3, "k": 4})'

leads to:

1000000 loops, best of 3: 0.373 usec per loop (timeit)           
1.54s user 0.00s system 99% cpu 1.545 total (time)

---

time python -m timeit 'd={"i": 1}; d = dict(d, **{"i": 2, "j": 3, "k": 4})'

leads to:

1000000 loops, best of 3: 0.43 usec per loop (timeit)            
1.77s user 0.00s system 99% cpu 1.778 total (time)

---

time python -m timeit 'd={"i": 1}; d["i"] = 2; d["j"] = 3; d["k"] = 4'

leads to:

10000000 loops, best of 3: 0.145 usec per loop (timeit)
5.98s user 0.00s system 99% cpu 5.986 total (time)                                                                                                                                                                                    

Answer 1

• time tells you how long it takes to run the benchmark .
• timeit tells you how long it takes to run the code the benchmark is built to measure .

These are two very different things. time includes all the overhead of starting the interpreter; of collecting the statistics on how long your individual benchmarked runs take; of interpreting those results; of doing any additional runs which got thrown out because they were unstable or otherwise statistically unsuitable; of performing garbage collection between timed invocations; etc.

Answer 2

看起来结果并不矛盾，因为第三种情况在10,000,000次运行10次以上，而情况1和2运行1,000,000次。