Add `isless` variant where `missing` is smallest

[bkamins]

The definitions to make missing smallest would be

isless2(::Missing, ::Missing) = false
isless2(::Missing, ::Any) = true
isless2(::Any, ::Missing) = false
isless2(x, y) = isless(x, y)

@nalimilan - the question is what should be the name of this function (as usual this is the hardest part).

Also a question is if Missings.jl should export it (I think it is OK to export it if we find a good name)

[nalimilan]

islessmissingsmallest or islessmissingfirst would be explicit though a bit long...

[bkamins]

This is my fear. Let us ask on Slack.

[sprmnt21]

let's start by removing isless, since it is the default value missingassmallest missingissmallest

[nalimilan]

We could also say missingisless since this function really does this: consider that missing is less than all other values.

[bkamins]

missingless? (without s in the middle)

[sprmnt21]

it sounds like a play on words (or is it? I like it). missingless could be read as no missing

[alejandromerchan]

You guys obviously know way more than me with regards of design and philosophy of these packages, but I thought of this operation as sort acting in somehow in combination with skipmissing. I don't know if that's the right approach an what consequences you have from this, but something like sort(df, :A, rev = true, skipmissing = true) was the operation I was hoping to have.

[bkamins]

The approach we are discussing is more general and would allow you to sort any collection, e.g. a vector in exactly the same way.

[nalimilan]

Any objections to missingisless?

BTW, since the main use case is to do sort(x, lt=missingisless, rev=true), we could directly provide a function which reverses the order. But I don't have a good name for that (missingisnotgreater?)

[bkamins]

missingisnotgreater feels unclear.

missingisless seems OK.