[issue25159] Import time regression

New submission from Serhiy Storchaka: There are regressions in import time in 3.5 and 3.6. $ for i in `seq 5`; do ./python -I -m timeit -n1 -r1 "import enum"; done Python 3.4: 1 loops, best of 1: 3.45 msec per loop 1 loops, best of 1: 3.43 msec per loop 1 loops, best of 1: 3.55 msec per loop 1

[issue25159] Import time regression

Brett Cannon added the comment: It should also be mentioned that startup time from 3.4 to default has not been affected according to Intel's daily benchmark run which is usually the most obvious place you see import performance issues: normal_startup0.63661% 0.39540%

[issue25159] Import time regression

Serhiy Storchaka added the comment: This is not startup time, this is import time. All binaries were compiled in release mode. The regression is reproducible with other modules. $ for i in `seq 5`; do ./python -I -m timeit -n1 -r1 "import zipfile"; done Python 3.4: 1 loops, best of 1: 46 msec

[issue25159] Import time regression

STINNER Victor added the comment: Even if Python 3.5 startup is slower than Python 3.4 startup, I'm more concerned by Python 3.6! What happens in Python 3.6? Are you sure that all binaries were compiled in release mode? Why do you import enum? The bare minimum is "pass" :-) --

[issue25159] Import time regression

Brett Cannon added the comment: I don't know how much of this is directly importlib's fault, at least in the 3.5 -> 3.6 case, as very little has changed in the bootstrap code since 3.6 started: https://hg.python.org/cpython/log/default/Lib/importlib/_bootstrap.py and

[issue25159] Import time regression

Changes by STINNER Victor : -- nosy: +haypo ___ Python tracker ___ ___