[issue-48] Ability to serialize a hash object

[mikebaldry]

If a Hash is passed as object, it will attempt to find attributes using Hash#[].

PankoSerializer must at some point turn attributes :blah from a Symbol to a String, as it only works with String keys. I think this is good enough as it'd be hard to detect which to use, and might cause confusion.

I attempted to make it work with ActiveSupport::HashWithIndifferentAccess but was unable to get the hash lookup working in ext

    it "hash (with indifferent access)" do
      foo = ActiveSupport::HashWithIndifferentAccess.new(
        "name" => Faker::Lorem.word,
        "address" => Faker::Lorem.word
      )

      expect(foo).to serialized_as(FooSerializer,
                                   "name" => foo[:name],
                                   "address" => foo[:address])
    end

void hash_attributes_writer(VALUE obj, VALUE attributes,
                             EachAttributeFunc func, VALUE writer) {
  long i;
  for (i = 0; i < RARRAY_LEN(attributes); i++) {
    volatile VALUE raw_attribute = RARRAY_AREF(attributes, i);
    Attribute attribute = attribute_read(raw_attribute);
    volatile VALUE value = rb_funcall(obj, rb_intern("[]"), 1, attribute->name_id);

    func(writer, attr_name_for_serialization(attribute), value);
  }
}

I expect rb_hash_lookup is faster than rb_funcall with [] but that might bypass overridden [] method on Hash subclasses like ActiveSupport::HashWithIndifferentAccess. Either way, it works for a plain Hash but not for ActiveSupport::HashWithIndifferentAccess and I can't figure out why (my ext experience is limited)

Fixes #48

[mikebaldry]

Also I've seen that rb_hash_aref is used to access Hash value. That doesn't have any different result.

[yosiat]

@mikebaldry thanks for the PR!

Supporting other types of hash is possible, but it's complicated because of the reason you posted.

If you are ok with your result, I'll merge this.

[mikebaldry]

I'm happy with this. It could be improved upon in the future without breaking backwards compatibility.

Thanks

[mikebaldry]

This actually did not work at all - my test was incorrect, I had it working at some point but then with the trying to get it working with different types of hash, I broke it but the tests still passed.

I was asserting that serialized values were equal to hash[:blah] where the hash had string keys, so it was returning nil and therefore the tests passed because the code in hash.c was incorrectly returning nil.

Once I realised my mistake I was able to track down the reason it wasn't working and make it work with HashWithIndifferentAccess also.

This is now ready to merge :)

[yosiat]

@mikebaldry merged, will release a version later this week.

[mikebaldry]

Great, thanks!