Improve consistency of field shapes

[rmosolgo]

Initialize GraphQL::Schema::Field so that its instance variables are always initialized in the same order. This way Ruby's "Object Shapes" optimization will work for field instances during query execution. (However, this also requires an application's custom field state to use consistent ivar setup, too.)

Part of #4299

This ended up improving the Schema footprint memory benchmark, though I couldn't tell you why:

  $ bundle exec rake bench:profile_schema_memory_footprint

- Total allocated: 4060200 bytes (34279 objects)
+ Total allocated: 4058600 bytes (34279 objects)
- Total retained:  1832888 bytes (6185 objects)
+ Total retained:  1831288 bytes (6185 objects)


  allocated memory by location
  -----------------------------------
  ... 
-     199200  /Users/rmosolgo/code/graphql-ruby/lib/graphql/schema/resolver/has_payload_type.rb:92
+     197600  /Users/rmosolgo/code/graphql-ruby/lib/graphql/schema/resolver/has_payload_type.rb:92

[casperisfine]

Probably not worth it, but when using this kind of "unset" tokens, I prefer to inverse the check so that if I hit a weird user defined object with a custom == it won't be invoked.

You can also use equal? to only check identity:

Suggested change

	if @broadcastable != NOT_CONFIGURED
	if !NOT_CONFIGURED.equal?(@broadcastable)

But I agreed that it doesn't read as well.

[rmosolgo]

👍 Thanks for the suggestion, I'll update it!

[casperisfine]

Assuming you'd use this pattern in multiple classes, it might make sense to have a single GraphQL::NOT_CONFIGURED. You can also make it a private_constant if you don't want users to be able to pass it as argument.

[rmosolgo]

👍 I added it here and I've opened #4363 to track rolling it out across the codebase.

[casperisfine]

So this is kind of ROFLscale territory, but I wonder if all these extra references might slow down major GC pause.

One way to avoid this would be to use an immediate object, like a static symbol, e.g. :__GraphQL__NOT_DEFINED__. This allows the GC to deretctly skip that reference.

It's not as hack proof, as it's much easier for users to pass that symbol, but I can't imagine it happening by accident.

That said, I'm no GC expert, it might not make a difference at all.

[rmosolgo]

I'm going to leave it as-is 😅