new rule: remove unnecessary list wrapping

[spaceone]

''.join([chr(i) for i in range(128)])
should be:
''.join(chr(i) for i in range(128))

[ngnpope]

This was discussed way back when for C407 in adamchainz/flake8-comprehensions#156.

It was rejected because using a list comprehension was faster, and still is:

# Python 3.10.9

In [1]: %timeit ",".join([str(i) for i in range(3)])
389 ns ± 1.79 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [2]: %timeit ",".join([str(i) for i in range(300)])
17.9 µs ± 90.9 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [3]: %timeit ",".join(str(i) for i in range(3))
473 ns ± 1.68 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [4]: %timeit ",".join(str(i) for i in range(300))
20.8 µs ± 21.4 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

# Python 3.11.1

In [1]: %timeit ",".join([str(i) for i in range(3)])
384 ns ± 4.26 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [2]: %timeit ",".join([str(i) for i in range(300)])
16.2 µs ± 169 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [3]: %timeit ",".join(str(i) for i in range(3))
501 ns ± 5.38 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [4]: %timeit ",".join(str(i) for i in range(300))
19.9 µs ± 268 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

[charliermarsh]

So strange that that's faster! Intuitively I would've expected the opposite.

[spaceone]

ah, I saw this in the past as well.

The reason:
adamchainz/flake8-comprehensions#156 (comment)

because join internally converts all generators to lists

But I think the performance aspect doesn't matter here, it's minimal in the recent python versions.
code readability is more important. If performance matters use rust instead of python :-P

So a new rule which can do this and be deactivated by folks who want the faster version.

[charliermarsh]

I might err on the side of omitting if flake8-comprehensions intentionally omits it.

[charliermarsh]

(Even though I personally wouldn't mind it.)

[ngnpope]

Out of interest, the performance dip for 3.11 in the third test (small generator) is likely caused by python/cpython#100762.

Even then, it'll still be slower. If we did implement this, it should be a different rule to C407. But it would be good to have a rule that enforces a list (comprehension) too given that's most optimal.

[charliermarsh]

I'm going to close for now, we can re-open if flake8-comprehensions changes their stance.