Expanding on this approach, I was interested in how the numbers scaled to the somewhat more meaningful case of emptying a large list, and ended up substantially extending the benchmark script to make all the key parameters easily visible and tunable, and to be useful in the future to run arbitrary benchmarks on a set of alternatives. For reference, these are my results on an old Windows laptop running the original script:
22.9 ± 0.0 ns pass # as baseline
83.1 ± 0.0 ns lst.clear()
108.4 ± 0.1 ns lst *= 0
138.4 ± 0.3 ns del lst[:]
153.2 ± 0.3 ns lst[:] = ()
173.0 ± 0.2 ns lst[:] = []
Python: 3.11.0 | packaged by conda-forge | (main, Oct 25 2022, 06:12:32) [MSC v.1929 64 bit (AMD64)]
and my extended version:
22.9 ± 0.0 ns pass # as baseline
83.1 ± 0.1 ns lst.clear()
108.6 ± 0.2 ns lst *= 0
138.4 ± 0.3 ns del lst[:]
153.1 ± 0.5 ns lst[:] = ()
173.3 ± 0.3 ns lst[:] = []
Python: 3.11.0 | packaged by conda-forge | (main, Oct 25 2022, 06:12:32) [MSC v.1929 64 bit (AMD64)]
Total runtime: 2.56 s
with parameters set to match the original:
OUTER_LOOPS = 100
INNER_LOOPS = 10**4
SAMPLES = 10
UNIT = "ns"
CUSTOM_PARAMS = {"list_len": 5}
With the parameters modified to resemble a more realistic situation where this is more likely to actually matter, i.e. a list with 10 000 items:
OUTER_LOOPS = 100
INNER_LOOPS = 10**2
SAMPLES = 10
UNIT = "us"
CUSTOM_PARAMS = {"list_len": 10_000}
we can see that the difference is entirely fixed per-call overhead, and does not scale at all with list size:
0.3 ± 0.0 us pass # as baseline
30.0 ± 0.1 us lst.clear()
30.0 ± 0.1 us lst *= 0
30.1 ± 0.1 us del lst[:]
30.2 ± 0.0 us lst[:] = []
30.2 ± 0.1 us lst[:] = ()
Python: 3.10.11 | packaged by conda-forge | (main, May 10 2023, 18:51:25) [MSC v.1934 64 bit (AMD64)]
Total runtime: 4.89 s
So, this might matter if you’re clearing the same small list thousands to millions of times in a very tight inner loop, but will be lost in the noise on larger lists or when not in a tight inner loop.
Script
import random
import sys
import time
from statistics import mean, stdev
from timeit import timeit
OUTER_LOOPS = 100
INNER_LOOPS = 10**2
SAMPLES = 10
UNIT = "us"
CUSTOM_PARAMS = {"list_len": 10_000}
FUNCTIONS = '''\
lst.clear()
del lst[:]
lst[:] = []
lst[:] = ()
lst *= 0
pass # as baseline
'''.splitlines()
TEST_CODE = 'for lst in lists: {function}'
SETUP_CODE = 'lst = list(range({list_len})); lists = [lst[:] for _ in range({INNER_LOOPS})]'
UNITS = {
"s": 1,
"ms": 1e-3,
"us": 1e-6,
"ns": 1e-9,
}
PARAMS_TO_FILL = {
'OUTER_LOOPS': OUTER_LOOPS,
'INNER_LOOPS': INNER_LOOPS,
**CUSTOM_PARAMS}
def generate_stats(f, times):
ts = [t / UNITS[UNIT] for t in sorted(times[f])[:SAMPLES]]
return f'{mean(ts):5.1f} ± {stdev(ts):3.1f} {UNIT} '
def time_functions():
start_time = time.monotonic()
times = {f: [] for f in FUNCTIONS}
for __ in range(OUTER_LOOPS):
random.shuffle(FUNCTIONS)
for f in FUNCTIONS:
t = timeit(
TEST_CODE.format(function=f, **PARAMS_TO_FILL),
SETUP_CODE.format(function=f, **PARAMS_TO_FILL),
number=1,
) / INNER_LOOPS
times[f].append(t)
stats = {f: generate_stats(f, times) for f in FUNCTIONS}
for f in sorted(FUNCTIONS, key=lambda f: stats[f]):
print(stats[f], f)
print('\nPython:', sys.version)
print(f'Total runtime: {time.monotonic() - start_time:.2f} s')
if __name__ == '__main__':
time_functions()
