Repeated key creates a problem when reading objects

[nlw0]

Having a repeated key in an object definition causes the last value to be dropped. For example:

julia> using JSON3

julia> JSON3.read("""{"keyA": 1, "keyB": 2, "keyC": 3}""")
JSON3.Object{Base.CodeUnits{UInt8, String}, Vector{UInt64}} with 3 entries:
  :keyA => 1
  :keyB => 2
  :keyC => 3

julia> JSON3.read("""{"keyA": 1, "keyA": 2, "keyC": 3}""")
JSON3.Object{Base.CodeUnits{UInt8, String}, Vector{UInt64}} with 2 entries:
  :keyA => 1
  :keyA => 2

[quinnj]

Digging into this; a few thoughts up front though:

• as far as I understand, duplicate keys in objects are highly discouraged and not well supported in many implementations
• even in JSON3.jl, we don't have a way to retrieve a JSON3.Object value by index, only by key, which means there's fundamentally not a way to distinguish between getting the value 1 or the value 2
• So the core issue here is that the JSON3.Object knows there are 2 unique keys (keyA and keyC), and that's where the length(x) == 2 property comes from. But then when printing, once it has printed 2 entries, it thinks it's "done" because of the length
• Note however, that you are still able to correctly retrieve the keyC property by doing

julia> x = JSON3.read("""{"keyA": 1, "keyA": 2, "keyC": 3}""")
JSON3.Object{Base.CodeUnits{UInt8, String}, Vector{UInt64}} with 2 entries:
  :keyA => 1
  :keyA => 2

julia> x.keyC
3

Anyway, I'll have to think about what to do here. Maybe we need to expose a way to index a property by ordinal number to allow getting either keyA value. It seems we also need to fix the length computation.

[quinnj]

Here's a PR that at least fixes the length computation: https://github.com/quinnj/JSON3.jl/pull/217

[nlw0]

Great, thanks for looking into it. I must check in my actual application if the last key really is present, because I think that's how I found the problem, not just looking at the print.

I would expect a repeated key to just be discarded, with any one the possible values being chosen... Or maybe even an error. I had a repeated entry in my file by accident, but other JSON parsers just ignored it, so I never noticed.

[nlw0]

The problem I had before was that converting the object to a Dict with copy drops the last value, but it's in the original object indeed. Hopefully the length fix already solves that as well.