findextrema: compute findmin and findmax in single pass

[andrewjradcliffe]

This is a simple extension of extant findmin and findmax methods. Depending on context (cost of f; whether reduction is over dims; size of array) the speedup increase is somewhere between 1.0-1.6 (no regressions). Interestingly, I noticed but could not locate a findextrema; there is some mention of it, but nothing in Base. If it was deemed unworthy, please excuse this errant PR.

[mikmoore]

extrema is to (minimum,maximum) as minmax is to (min,max), so perhaps findminmax is the a more consistent name here? Although there isn't an obvious notion of a distinction between findminmax and findextrema, so either would be appropriate. Wait to see what other people feel about this before you bother to change it.

[StefanKarpinski]

Thanks for the PR—this does seem like desirable functionality! Anyone familiar with the reduce machinery willing to code review this?

[N5N3]

I agree with @mikmoore on the name's consistency.
As the added funtion return findmin(), findmax() rather than extrema(), indexes.
Although findminmax! has been used for a internal method, we can rename it without breakage.

[andrewjradcliffe]

I agree with @mikmoore on the name's consistency. As the added funtion return findmin(), findmax() rather than extrema(), indexes. Although findminmax! has been used for a internal method, we can rename it without breakage.

I went with findextrema as per the link in the OP, but perhaps opinions have changed since 2019. Whichever name we choose, we will most likely be stuck with.
My thoughts on name choice are: at first glance, findminmax seems most natural, as it confers the notion of find_min_and_max. Alas, minmax looks quite similar to minimax, which is an entirely different concept. Alternatively, findextrema, while arguably uglier, is unambiguous in relation to extrema.

[KristofferC]

Can you add something to the NEWS.md file about this?