vcat(::SVector, ::Number) should return an SVector

[dlfivefifty]

I assume this behaviour is just an oversight:

julia> vcat(SVector(1), 2)
2-element Array{Int64,1}:
 1
 2

Shall I make a PR overriding vcat(and possibly hcat)?

[c42f]

I suspect that these methods don't exist because it was tricky to deal with the resulting ambiguities without support from Base. However if you can figure that out a PR would be most welcome.

[c42f]

To explain, a combinatoric explosion of ambiguities for varargs functions like vcat can arise when Base doesn't include extra machinery required to make it easy to override the output container type. This extra machinery is very well developed for broadcast (see the Base.Broadcast.BroadcastStyle trait) but for many other functions taking multiple containers it's not so easy.

[dlfivefifty]

Yes I understand. It might be easier to use LazyArrays.jl and use SVector(Vcat(SVector(1), 2)), though I'll have to figure out how to get it tell that this is an SVector{2}.

[c42f]

Well, if you just want a workaround there's plenty of options. Just wrapping the scalar in an SVector of length 1 for instance.

[dlfivefifty]

Ah, I see, it's only missing on the scalar case:

julia> vcat(SVector(1), SVector(2))
2-element SArray{Tuple{2},Int64,1,2}:
 1
 2

The benefit of the LazyArrays.jl approach is we can use ApplyStyle to determine that it should be an SVector automatically even for very complicated types. But I probably won't look at this anytime soon as there was an even simpler work around for my use case: SVector(x..., y...).

[c42f]

Should be mitigated (to the extent which is easily possible) by #768

[c42f]

Interesting XRef which would have fixed this in Base: JuliaLang/julia#20815