重用SystemRandom是一种很好的Python实践吗？

Question

For example, a block of Python code to generate 1 million random strings off the alphabet 'abcdef'.

Compare

from random import SystemRandom

for _ in range(1000000):
    ''.join(SystemRandom().choice('abcdef') for __ in range(5))

with

from random import SystemRandom

r = SystemRandom()
for _ in range(1000000):
    ''.join(r.choice('abcdef') for __ in range(5))

I timed it on IPython. The 1st block took 50s while the 2nd block took 30s. I didn't measure the memory usage.

In Java, there is SecureRandom where the pattern is clear that it should be created once and reused. Coming from Java to Python, it is not clear to me whether I should create a single SystemRandom instance and reuse it. The 1st block is easier to read and the performance does not seem too bad.

In general, does Python programming pay more attention to readability than performance?

Answer 1

SystemRandom builds on the platform C's urandom() function, which is as secure as your operating system developers knew how to make it. Virtually everyone uses a single instance of it (as in your second example).

There's no real conflict between performance and readability here. Idiomatic Python would be like:

from random import SystemRandom
r = SystemRandom()
choice = r.choice

for _ in range(1000000):
    ''.join(choice('abcdef') for _ in range(5))

Answer 2

When in doubt, I usually look at the implementation ...

As far as I can tell, SystemRandom works pretty much exactly the same as Random ( which is reused by all of the module-level functions ) except that the random method is swapped out to call os.urandom . So, as far as I can tell, you should be able to re-use a SystemRandom instance.

In general, does Python programming pay more attention to readability than performance?

In general, I think that the mentality is that readability counts and code is read more often than written . You should only really spend time (or code complexity) optimizing things which are taking a significant amount of your over-all program running time. If creating and calling SecureRandom takes a tiny percentage of your overall program running time, then it's probably not worth optimizing and you should definitely go with the more easy to read version. If it takes 50% of the run-time and waiting for the program to complete is annoying, you might want to look into optimizing.

Of course, with that said, I don't think there's much of a readability difference between your two code snippets, so ... I'd probably create one and reuse it...