The checks are indeed an added cost, and that’s why I said “almost certainly less efficient” the first time rather than “less efficient”. (But subsequent times I did drop it.)
But what you call “the O(1) operation of adding two items to an array” is simply not true:
① While appending to arrays is normally O(1) in the cases that there is still spare capacity in the array and the entire thing doesn’t need to be reallocated, prepending is O(n) as it has to move all of the subsequent values along.
② What’s happening is not just adding two items to an array, but rather allocating an entire new array and copying all the values into it; this will be roughly O(n), but with a high constant overhead: allocating memory is expensive.
I can’t speak of any Ruby interpreter’s performance on the matter, and I’m fairly confident that the conditionals will be much more expensive there than they would be in Rust (where I have no doubt at all that the conditional version would smoke the allocating version in all cases), but allocating memory is a pretty expensive business.
> Finally, the author doesn’t claim that the special cases are gone
I quote: “The special cases are gone”. Further, I do not grant your response that the processing is blind to them—the inside of the loop is, but the special cases were moved to the boundary conditions of the loop. They’re still there.
But what you call “the O(1) operation of adding two items to an array” is simply not true:
① While appending to arrays is normally O(1) in the cases that there is still spare capacity in the array and the entire thing doesn’t need to be reallocated, prepending is O(n) as it has to move all of the subsequent values along.
② What’s happening is not just adding two items to an array, but rather allocating an entire new array and copying all the values into it; this will be roughly O(n), but with a high constant overhead: allocating memory is expensive.
I can’t speak of any Ruby interpreter’s performance on the matter, and I’m fairly confident that the conditionals will be much more expensive there than they would be in Rust (where I have no doubt at all that the conditional version would smoke the allocating version in all cases), but allocating memory is a pretty expensive business.
> Finally, the author doesn’t claim that the special cases are gone
I quote: “The special cases are gone”. Further, I do not grant your response that the processing is blind to them—the inside of the loop is, but the special cases were moved to the boundary conditions of the loop. They’re still there.