keys and hash don't work on ranges with many elements

[rfourquet]

julia> keys(1:Int128(2)^64)
ERROR: InexactError: trunc(Int64, 18446744073709551616)
Stacktrace:
[...]

And hash fails because keys fails, but it's unfriendly to not be able to store a huge range in a Set.

[goretkin]

julia> LinearIndices((Base.OneTo(typemax(Int128)),))
ERROR: InexactError: trunc(Int64, 170141183460469231731687303715884105727)

LinearIndices assumes the indices fit into Int, which makes sense for Array, but not for every AbstractArrays.

But also the O(log(n)) hashing time probably starts making a situtation where you're hashing huge integer intervals really require a different type that is not an AbstractArray and is therefore not beholden to the isequal/hash invariants. I was surprised / found it unfriendly that hashing a range is not O(1), but it makes sense.

It might be more friendly to for hash(1:typemax(Int128)) to error (although, better error messages are always friendly), than for the hashing a 32-byte struct to take milliseconds or worse. Perhaps a suggestion to use Intervals.jl?

julia> using BenchmarkTools

julia> using Intervals

julia> b = typemax(Int64)
9223372036854775807

julia> @benchmark hash(1:typemax($b))
BenchmarkTools.Trial:
  memory estimate:  0 bytes
  allocs estimate:  0
  --------------
  minimum time:     2.676 ms (0.00% GC)
  median time:      2.816 ms (0.00% GC)
  mean time:        2.967 ms (0.00% GC)
  maximum time:     5.337 ms (0.00% GC)
  --------------
  samples:          1683
  evals/sample:     1

julia> @benchmark hash(1..typemax($b))
BenchmarkTools.Trial:
  memory estimate:  0 bytes
  allocs estimate:  0
  --------------
  minimum time:     59.122 ns (0.00% GC)
  median time:      59.973 ns (0.00% GC)
  mean time:        67.024 ns (0.00% GC)
  maximum time:     221.062 ns (0.00% GC)
  --------------
  samples:          10000
  evals/sample:     982

julia> hash(1..typemax(Int128))
0x1a5c49cfe49ef6f4