Median

[benrimmington]

I might remove the two median<T: FloatingPoint> functions from the proposal, or at least mention the possibility in an Alternatives Considered section.

---

@Nevin

I’m not sure how I feel about introducing floating-point versions that use isTotallyOrdered(belowOrEqualTo:).

My original implementation didn't use a total ordering, so it only documented using a stable sorting algorithm. Swift 5.0+ has a stable sort (apple/swift#19717), but this isn't guaranteed yet, and would need another proposal.

@Nevin

Conversely, if one takes the median of some numerical data, it is also perfectly sensible to expect that any non-numerical values would simply be ignored, skipped over, and not included in the calculation. In other words, one might hope to extract the median of the non-NaN values.

TabularData.NumericSummary.median also ignores "missing" elements, but I think that refers to Optional.none rather than NaN.

@jawbroken

The only motivation given in the pitch is for clamping to a range, for which I think the ideal solution would be a method with “clamp” in the name that takes a range type, probably ClosedRange. I don't think people are going to intuitively reach for a global median function for that use case.

The idea of a global median function was suggested last year by @scanon (and endorsed by @kylemacomber). I'd like to keep the proposal short, so I'm reluctant to expand the Motivation section.

@1-877-547-7272

I also think it would be a good idea to have a separate clamped(to:) method as an extension to the Comparable protocol, like what @jawbroken suggested.

That would be the SE-0177 proposal, which was returned for revision (4 years ago), and revisited a few times since then.