Revert converting str.join(listcomp) to str.join(generator)

Consider reverting the listcomp->generator for the str.join() case, it’s faster to use a list comprehension than a generator in this case:

$ python -m timeit '"y".join("x" for _ in range(1000000))'
5 loops, best of 5: 40.6 msec per loop
$ python -m timeit '"y".join(["x" for _ in range(1000000)])'
10 loops, best of 5: 29.6 msec per loop

This is because str.join() constructs a list internally and short-circuits if passed an already created list.

[MarcoGorelli]

Having tried this out, it looks like even that one’s slower:

In [8]: %timeit frozenset([i for i in range(1000)])
36.6 µs ± 661 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)

In [9]: %timeit frozenset((i for i in range(1000)))
41.8 µs ± 940 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)

I’d suggested this feature, and it seems that in several cases it induces a small performance hit, so…I’m sorry about that

@asottile reckon it’s worth removing this rewrite altogether?

[sethmlarson]

@MarcoGorelli Thanks for making my eyes actually work, sorry for the noise.