Alexander Plavin

Anything wrong with this PR? It nicely gets rid of all those special cases. Behavior doesn't change at all, as evidenced by no test changes. Just stumbled across this piece...

Hm, that's... interesting. Basically, the issue is that without `lt =

The updated implementation is both correct (findall tests pass, and they do check for zeros of both signs), and keeps the performance for ranges (Base `searchsorted` funcs are called without...

This PR also fixes https://github.com/JuliaMath/IntervalSets.jl/issues/100 btw.

Maybe, the time difference is some benchmarking artifact. At least the timings definitely looks suspicious, in both cases. 1. `findall(∈(1..3), 2:8)` changing from 0.02 ns to 1.47 ns. The first...

Added the zero-step tests.

Anything else remains to be done here?..

> Maybe, it seems better to split this PR into > Add rev keyword argument in searchsorted_interval function. > Simplify findall function. > to merge this PR quickly. "Quickly" is...

> The searchsorted family functions use binary search `searchsorted` definitely doesn't use binary search for ranges by default, see around https://github.com/JuliaLang/julia/blob/24cb92d1e94367c5ef758545b69133ded4559763/base/sort.jl#L230. That's why I expect searchsorted_interval to be as performant...

As I said in a comment above, you just removed `include("staticarrays.jl")`. Add it back, to all other includes.