Add sort_keys option in the generator.

[samyron]

I had some time last night so I decided to see how hard it would be to implement a sort_keys option to the generator. This adds support to the C extension, Java extension and truffleruby implementation.

irb(main):005> h = { zzz: true, a: 1, c: 2, b: "Test", 'z': 'blah 2'} 
=> {zzz: true, a: 1, c: 2, b: "Test", z: "blah 2"}
irb(main):006> JSON.generate(h, sort_keys: false)
=> "{\"zzz\":true,\"a\":1,\"c\":2,\"b\":\"Test\",\"z\":\"blah 2\"}"
irb(main):007> JSON.generate(h, sort_keys: true)
=> "{\"a\":1,\"b\":\"Test\",\"c\":2,\"z\":\"blah 2\",\"zzz\":true}"
irb(main):008> JSON.generate(h)
=> "{\"zzz\":true,\"a\":1,\"c\":2,\"b\":\"Test\",\"z\":\"blah 2\"}"

[byroot]

THanks for handling this, however in lieu of #1038, I think a sort_key option shouldn't be a a simple boolean, but should also accept a proc so we can implement custom sorting semantic.

[samyron]

THanks for handling this, however in lieu of #1038, I think a sort_key option shouldn't be a a simple boolean, but should also accept a proc so we can implement custom sorting semantic.

No problem. I can try to take a stab at that tonight night/this weekend. I think it should be as simple as accepting a proc and passing it to obj.sort_by when generating the JSON.

[samyron]

I added the ability to pass a Proc which accepts two [key, value] pairs to allow for arbitrary sorting. I'm not entirely sure if this is what you had in mind but it allows for maximum flexibility. I'm not totally sure if that's warranted but I ran with that assumption. I'm happy to make changes as requested.

Note that a boolean (true) is still supported which will sort keys lexicographically.

irb(main):002> reverse = ->(a, b) { b[0] <=> a[0] }
=> #<Proc:0x000000012d2bbc08 (irb):2 (lambda)>
irb(main):003> JSON.generate({2=>"a", 1=>"b", 3=>"c"}, sort_keys: reverse)
=> "{\"3\":\"c\",\"2\":\"a\",\"1\":\"b\"}"

[byroot]

Sorry, for not immediately communicating in details what I had in mind, I was traveling.

Based on the type of sorting json-normalization needs: https://github.com/dryruby/json-canonicalization/blob/a85e688a4a593ed6b28e5aa4610c996828c13ada/lib/json/canonicalization.rb#L64-L78

I think it would be better if the sort proc would just take the entire hash at once, and return a sorted hash.

As for the sort: true argument, that's great, that's what I had in mind, but we might be better to just convert it into a proc early on the Ruby side:

def generate(..., options)
  if options[:sort_keys] == true
    options = options.merge(sort_keys: ->(hash) { hash.sort.to_h }
  end
  ...
end