Fix resource serialization involving last_result_set

[stokarenko]

Sometimes JsonApiClient::Resource instance evaluates self.relationships.last_result_set = last_result_set or self.relationships.last_result_set = nil. This behavior was introduced by performance optimization commit.

But we never read the last_result_set property on JsonApiClient::Relationships::Relations objects, only the writes everywhere. That means that these writes are kind of rudiments.

Even worse, there is no attr_accessor :last_result_set defined for JsonApiClient::Relationships::Relations class, this way an attempt to write last_result_set property triggers Helpers::DynamicAttributes routine - last_result_set value become to be a part of attributes hash, and will be serialized then.

Lets remove rudimental writes to relationships.last_result_set. Without this fix the newly added test fails like this:

1. Failure:
   SerializingTest#test_as_json_involving_last_result_set [/Users/st/dev/ror/everfi/json_api_client/test/unit/serializing_test.rb:75]:
   --- expected
   +++ actual
   @@ -1 +1 @@
   -{"type"=>"articles", "id"=>"1", "attributes"=>{"title"=>"Rails is Omakase"}}
   +{"id"=>"1", "type"=>"articles", "relationships"=>{"last_result_set"=>{"data"=>[{"type"=>"articles", "id"=>"1"}]}}, "attributes"=>{"title"=>"Rails is Omakase"}}

[stokarenko]

@gaorlov Hi again )

as an option we can add attr_accessor :last_result_set to JsonApiClient::Relationships::Relations class definition, but I'm pretty sure that these writes are rudimental )

[gaorlov]

Looks like relationships.last_result_set was being handled by the dynamic attribute helper; I don't think we need to add an accessor.
This looks fine to me, but I'd like to understand what that code did int he first place. @senid231 can you explain your thinking here?

[stokarenko]

@gaorlov @senid231

Looks like relationships.last_result_set was being handled by the dynamic attribute helper

Indeed, and that is a problem - this way such last_result_set will take a part in serialization data, raising json size for nothing..

This PR just removes all rudimental writes to relationships.last_result_set, that cures the core reason of problem. As an alternate solution we can add attr_accessor :last_result_set to relations class, and avoid the garbage in relation attributes this way - in case if I missed something important around relationships.last_result_set :)

[gaorlov]

@senid231 have you had a chance to look over this? I'm fine with adding the attr_accessor, but i'd like to better understand why the changes were introduces before we expand the API surface.
Thanks!

[senid231]

@gaorlov @stokarenko indeed last_result_set is useless for relationships

I've introduce it by mistake while add relationship caching

thanks for fixing it, @stokarenko

[gaorlov]

1.14.0 is live. Thanks for the contribution!

[stokarenko]

Thank you a lot! )