Inaccurate/misleading description of max in operators module (wrt. NaN comparison)

[abonie]

Not sure if this is an error in documentation or purposeful omission, or perhaps a bug in fsharp.

Here max is described as

Maximum based on generic comparison

And here compare is described as

Generic comparison.

However, even though they kind of boil down to the same thing in implementation for most types, they do not specifically for floating point numbers. Moreover, if one defines a max function with compare, for example like this:

let max2 x y = if compare x y < 0 then y else x

Then max2 behaves differently than max with respect to NaN, e.g.:

> max2 1. nan;;
val it: float = 1.0

> max2 nan 1.;;
val it: float = 1.0

whereas

> max nan 1.;;
val it: float = nan

> max 1. nan;;
val it: float = nan

If this discrepancy in behavior is expected, then perhaps it should be documented?