[fix] [#149] do not merge/extract fragment fields into node fields

[kindermax]

If there is no such field in node, leave field in fragment

There was a part of algo in nodes merging which will extract fields from fragment if same fields exist in node - this is to prevent same field to be resolved multiple times. The simplest example is { user { id ... on User { id } } } would become { user { id } } so id field will be resolved only once.

But previous algo (before this fix) would merge eagerly, so even if no field exits in node, algo will try to eliminate fragment at all, for example { user { ... on User { id } } } would become { user { id } } - fragment is eliminated.

This behavior breaks union queries and we must not touch union fragment, since this breaks unions. This fix tells algo to merge less eagerly, thas is - if fragment field not exist in node, we leave this field in fragment since there is nothing to optimize, no fields duplication will occur.

Note that although we do not merge fragment field to node fields, we still merge fields for the very same fragment field (see _merge_link).

Here are the rules:

{ 
  user { 
    ... on User { id }  # <- leave this fragment (this is fixed by PR)
  } 
}

{ 
  user { 
    id
    ... on User { id } # <- merge id into parent node
  } 
}
# to
{ 
  user { 
    id
  } 
}

{ 
  user { 
    id
    ... on User { id email } # <- merge id field into parent node
  } 
}
# to
{ 
  user { 
    id
    ... on User { email } 
  } 
}

{ 
  user { 
    ... on User {  # <- leave this fragment (this is fixed by PR)
      id
      company {
        id
        .. on Company {  # <- but merge this fragment
          id
        }
      } 
    } 
  } 
}
# to
{ 
  user { 
    ... on User { 
      id
      company {
        id
      } 
    } 
  } 
}

[n4mespace]

lgtm