[Python-Dev] PEP 471 (scandir): Poll to choose the implementation (full C or C+Python)

[Victor Stinner]

2015-02-13 12:27 GMT+01:00 Serhiy Storchaka <storchaka at gmail.com>:
> On 13.02.15 12:07, Victor Stinner wrote:
>>
>> * C implementation: scandir is at least 3.5x faster than listdir, up
>> to 44.6x faster on Windows
>
>
> Results on Windows was obtained in the becnhmark that doesn't drop disk
> caches and runs listdir before scandir.

The benchmark code is here:
http://bugs.python.org/file38120/bench_scandir2.py

Eeach test is repeated 5 times, I compared the duration of the fastest
call. A test calls the function 5 times in a loop.

Anyway, it's not the first call to listdir() which fills the disk
cache, but the call to count_entries() (which is implemented with
os.scandir).

So is there any issue in the benchmark script or not?

On Linux, you can use "bench_nocache" (and "bench_nostat_nocache")
which flushs the cache: it writes "3" into /proc/sys/vm/drop_caches.
scandir is always faster when the disk cache is flushed.

Victor