Issue with Array eltype promotions about json3.jl

[quinnj]

Thanks for opening an issue! I included a few tweaks to help in #10. I fixed the promotion from Union{Int, Float64} to Any, and also changed so if you've parsed a Float64, then it will only try to parse Float64 for the rest of the array. The one case where you'll still get Union{Int, Float64} is a case like JSON3.read("[1.0, 2.1]"), which may not be ideal, but I'm not sure quite how we could handle that without just parsing Float64 for every number.

[quinnj]

Merged #10

[lbilli]

Thanks for the quick fix!

However, I still find some odd behavior:

JSON3.read("[1.2, null, 3.3]")
3-element JSON3.Array{Union{Nothing, Float64}, ... # ok

JSON3.read("[1.2, null, 3.0]")
3-element JSON3.Array{Any, ...                     # I was hoping for the same as above 

My use case is a numeric array that may contain missing values, encoded as null. It'd be useful to know whether there are any null without scanning the array, since the parser already has this information.

It already works as expected for Int64 or String.

[lbilli]

Thanks again for taking care of this so promptly.

I'm sorry to keep dragging this on, but I'd like to mention that the last remaining case that is still less then ideal is when the first non-null element in a Float array happens be an Int, in which case the eltype depends on the order of the elements:

JSON3.read("[1.2, 3, null]")    # Union{Nothing, Float64}

JSON3.read("[3, 1.2, null]")    # Union{Nothing, Float64, Int64}

JSON3.read("[3, null, 1.2]")    # Any

JSON3.read("[null, 3, 1.2]")    # Any

JSON3.read("[null, 1.2, 3]")    # Union{Nothing, Float64}

[quinnj]

Sorry for being slow here, here's some more improvements: #16