Recently I had a comment in a merge request in which I was suggested to copy the Nth-1 last elements of a list, a[1:], instead of popping its first element, pop(0).
I was curious about its performance implications and decided to do some simple tests:
t1 = timeit.Timer('a=50*[\'a\'];a.pop(0)')
t2 = timeit.Timer('b=50*[\'b\'];b[1:]')
t1.timeit(10000)/10000
6.973965995712205e-07
t2.timeit(10000)/10000
8.281046990305186e-07As shown in the results above, the difference is insignificant at least for small lists.
Checking for bigger lists:
t1 = timeit.Timer('a=5000*[\'a\'];a.pop(0)')
t2 = timeit.Timer('b=5000*[\'b\'];b[1:]')
t1.timeit(10000)/10000
2.4743527399550657e-05
t2.timeit(10000)/10000
4.327827289962443e-05shows that there’s a ~53% performance penalty of slicing versus popping.
Checking even bigger lists:
t1 = timeit.Timer('a=50000*[\'a\'];a.pop(0)')
t2 = timeit.Timer('b=50000*[\'b\'];b[1:]')
t1.timeit(10000)/10000
0.0002497872758001904
t2.timeit(10000)/10000
0.00044558706480020194shows that slicing the list carries a performance penalty of ~50% compared to just doing a pop of the first element. This penalty seems to plateau after a certain list size.