Faster itertoolz.update_in

[groutr]

The existing, recursive, form of update_in is unnecessarily slow. This borrows the non-recursive implementation from cytoolz and shows a good performance boost.

from toolz import update_in, update_in2
keys1 = list(reversed([list(range(i+100)) for i in range(100)]))
keys2 = list(reversed([list(range(i)) for i in range(1, 11)]))

# A custom assoc_in function which simply allows me to 
# change which update_in function is called.

%%timeit
d = {}
for x in keys1:
    d = assoc_in(update_in, d, x, 42)
# update_in (keys1)
26.3 ms ± 668 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
# update_in2 (keys1)
4.96 ms ± 62.2 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

# update_in (keys2)
86.7 µs ± 2.48 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
# update_in2 (keys2)
21.4 µs ± 118 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)

[groutr]

ping @eriknw

[groutr]

@eriknw do you have any feedback on this change?

[eriknw]

LGTM, thanks @groutr! Are there any other speed improvements to steal from cytoolz?

Also, aesthetically, this uses more blank lines then I generally prefer, but I'd like to get this in, so merging!

[groutr]

@eriknw I'm sure there are probably other speed improvements that could be carried over from cytoolz, but I haven't looked in awhile. Thanks for merging!