A function that returns the right constructor given a max length

[oxinabox]

I would like a way to simplify this code.

function shorten_strings(vec::AbstractVector{<:AbstractString})
    isempty(vec) && return vec

    # chose a type big enough for everything, to avoid large union
    max_len = maximum(ncodeunits, skipmissing(vec))
    return if max_len <= 7
        map(ShortString7, vec)
    elseif max_len <= 15
        map(ShortString15, vec)
    elseif max_len <= 30
        map(ShortString30, vec)
    elseif max_len <= 62
        map(ShortString62, vec)
    elseif max_len <= 126
        map(ShortString126, vec)
    else
        vec
    end
end

or even the more complex version that support missings:

using Missings: passmissing

function shorten_strings(vec::AbstractVector{<:Union{Missing, AbstractString}})
    isempty(vec) && return vec

    # chose a type big enough for everything, to avoid large union
    max_len = maximum(ncodeunits, skipmissing(vec))
    return if max_len <= 7
        map(passmissing(ShortString7), vec)
    elseif max_len <= 15
        map(passmissing(ShortString15), vec)
    elseif max_len <= 30
        map(passmissing(ShortString30), vec)
    elseif max_len <= 62
        map(passmissing(ShortString62), vec)
    elseif max_len <= 126
        map(passmissing(ShortString126), vec)
    else
        vec
    end
end

I am not sure what the best solution is.

[xiaodaigh]

This is very readable, why do you want to shorten it? I would add a function that returns a type stable function, so the type stabe version can be run.

[ScottPJones]

Instead of having ss3, ss7, etc., where the number doesn't actually match the number of characters if they are not ASCII (and note, you can't even store 1 Emoji in an ss3), why not have a macro: ss"string" that will pick the shortest power of 2 sized UInt, and then use the capability of a macro taking a second argument, so that ss"string"15 if you need to explicitly specify the size (in bytes) so that if you have a set of strings, you still can make them all in the same type.

[oxinabox]

Cute.
Using the second argument to the string macro is a great idea.

I still want a function form.
I think a function that takes a length and returns a constructor.

[xiaodaigh]

ss"string"15

definitely cute.

[xiaodaigh]

think a function that takes a length and returns a constructor.

agreed

[ScottPJones]

Yes, it should also have a simple constructor, like the ss"..." with an optional size (that would probably be how to easily implement the macro anyway)

[ScottPJones]

I can easily do a version that picks one of the UInt*s used in ShortStrings, but I think you should have the ability to pick your own (possibly larger, or possibly not a power of 2 size, such as a UInt24 or UInt48). I think to handle that, what is needed is a function that, given a vector or tuple of Unsigned types, returns the smallest one that can contain the given maxlen.
That type can then be used in the ShortString{T} constructor directly.