Proposed Changes to the Range API for Phobos v3

[Steven Schveighoffer]

On Tuesday, 24 September 2024 at 14:36:05 UTC, Steven 
Schveighoffer wrote:
> Results:
> ```
> opApplyRange took 186 ms, 123 μs, and 9 hnsecs for 100 runs
> phobosRange took 1 sec, 345 ms, 67 μs, and 5 hnsecs for 100 runs
> newRangeStyle took 83 ms, 706 μs, and 2 hnsecs for 100 runs
> simpleRanges took 136 ms and 47 μs for 100 runs
> ```
>
> Interestingly here, the simple ranges seem properly inlined. So 
> somehow the optimizer could see through those despite having 3 
> function API, but not the normal ranges? Is there something I 
> did special in there that the normal functions prevent?

Some more info, my simple filter range pre-computes the first 
element, and then avoids the whole "check if it's initialized" 
dance on `front` and `empty`.

If I add that "feature" to my range, it gets a lot slower, like 
5x slower. Still not as bad as Phobos, but it explains a bit of 
it.

This is the cost of "optimizing" ranges just in case someone does 
something unexpected.

Oh, and someone asked on discord, I am using `-O3` on the build.

-Steve