it looks cleaner now and good to me but I guess I'll let @uboness take a look at well!

Would be great if this could be expanded to include the usual parsing problems, like doubling up fields. For example:

{
  "aggs" : {
    "my_agg" : {
       "terms" : {
          "field" : "a",
          "field" : "b"
       }
    }
  }
}

This agg will only use the b field when building the terms bucket. The a field is silently ignored and no exception is thrown, which is very confusing for users unfamiliar with pull parsing.

I agree this should not fail silently. I think the best way to fix this issue would be to wrap the parser in order to make sure that objects don't contain duplicate keys.

@jpountz I like the idea of having a parser wrapper that ensures we have unique keys. Yet, that is a general thing and should be a different issue?

+1 on a different issue

been thinking about it (dup keys that is)... not sure about it... wrapping the parser to check that in a generic manner comes with a cost (keeping track of all keys)... I'm wondering if it's something we should just keep simple and as cheap as possible at the price of leniency

left a comment on missing tests... other than that, LGTM