Julia: `LinRange{Int}(...)` allows creating a LinRange that doesn't really work

[simeonschaub]

eltype(LinRange{Int}(1, 4, 5)) is Int, whereas the elements are actually of type Float64, since LinRange uses / for division. Furthermore, LinRange{Int}(1, 4, 5) isa AbstractRange{Int} is also technically wrong, since the docstring of AbstractRange clearly states:

  AbstractRange{T}

  Supertype for ranges with elements of type T. UnitRange and other types are subtypes of this.

Thanks to @asinghvi17 for bringing this up on Slack!

[timholy]

This is probably not fixable with the current type system. The issue is you need

julia> struct MyLinRange{T} <: AbstractRange{typeof(zero(T)/1)}
           start::T
           stop::T
           len::Int
           lendiv::Int
       end
ERROR: MethodError: no method matching zero(::TypeVar)
Closest candidates are:
  zero(::Type{Missing}) at missing.jl:103
  zero(::Type{LibGit2.GitHash}) at /home/tim/src/julia-1/usr/share/julia/stdlib/v1.4/LibGit2/src/oid.jl:220
  zero(::Type{Pkg.Resolve.VersionWeight}) at /home/tim/src/julia-1/usr/share/julia/stdlib/v1.4/Pkg/src/Resolve/versionweights.jl:15
  ...
Stacktrace:
 [1] top-level scope at REPL[1]:1

Alternatives include adding a second type parameter,

julia> struct MyLinRange{Item,T} <: AbstractRange{T}
           start::Item
           stop::Item
           len::Int
           lendiv::Int

           function MyLinRange{Item,T}(start,stop,len) where {Item,T}
               len >= 0 || throw(ArgumentError("range($start, stop=$stop, length=$len): negative length"))
               if len == 1
                   start == stop || throw(ArgumentError("range($start, stop=$stop, length=$len): endpoints differ"))
                   return new(start, stop, 1, 1)
               end
               new(start,stop,len,max(len-1,1))
           end
       end

julia> MyLinRange{Item}(start, stop, len) where Item = MyLinRange{Item,typeof(start/1)}(start, stop, len)

but I am not sure whether this would be regarded as too breaking.

Note that this issue doesn't come up with LinRange{1, 3, 3) (i.e., if you don't specify the type parameter).

[iamed2]

The elements of LinRange{Int} are actually Int. The printing is misleading.

julia> l = LinRange{Int}(1,5,4)
4-element LinRange{Int64}:
 1.0,2.33333,3.66667,5.0

julia> l[1]
1

julia> l[end]
5

julia> l[2]
ERROR: InexactError: Int64(2.333333333333333)
Stacktrace:
 [1] Int64 at ./float.jl:710 [inlined]
 [2] lerpi at ./range.jl:687 [inlined]
 [3] unsafe_getindex at ./range.jl:681 [inlined]
 [4] getindex(::LinRange{Int64}, ::Int64) at ./range.jl:666
 [5] top-level scope at REPL[12]:1

[timholy]

Oh, right, I had forgotten about that. That's a relief. So we should just fix the printing.

[sostock]

In addition to improving the printing, one could also throw an error when creating a LinRange{<:Integer} that would contain non-integer elements. I have a PR (#32439) that does this for StepRange{<:Integer} and StepRangeLen{<:Integer} and will update it to include LinRange.

[JeffBezanson]

Alternatives include adding a second type parameter,

Yes, I think that would be the right way to do it. (The type parameter can be set by a constructor, and invalid constructor calls can be errors.)

[sostock]

The misleading printing is fixed by #35267.

[JeffBezanson]

Triage thinks this is fixed now. If you explicitly ask for Int elements, you should indeed get an InexactError if the element values can't be converted to that.