[orblivion]

One thing this article does not cover is that, even if a comparison function doesn't do anything slow, there is still the fact that it still has to do a Perl function call O(n log n) times. At least in the equivalent with Python, I think it is advised to use the key function rather than comparison funtion for this reason, for speed. Though I guess a key function needs to take more space if it is to memoize.

[raiph]

A one arg closure to the P6 sort builtin (eg { .lc }) is a key function, not a comparison function. Imo the P6 sort builtin is an elegant rethink of P5's Schwartzian Transform, which was invented in 1994 to address precisely the point you make.

[orblivion]

I agree, this wasn't a point about Python vs Perl.

All I'm saying is that the author bills it as not wanting to run { .lc } twice per comparison. What I'm adding is that the comparison function itself, even if trivial, arguably has a bit of an overhead just from calling it. Thus, having O(n) calls to a key function { .lc } may be better than O(n log n) calls to a comparison function {.lc <=> .lc}.

At least that's how people convinced me to use key= instead of cmp= in Python.