Alternatively, I would think that gzip does a good job of factoring out repeatedly embedded user objects.
Yes I agree that gzip will probably fix this single-handedly. The main cost here is very likely network transfer, so gzip will do great.
Don't forget it takes time to parse large json blobs.
And to serialize it from whatever the in-memory representation is.
yes, obviously – gzip even does that explicitly, replacing repeated text with references.
But it still takes more power for your server to go through gzip every time. And it will take more RAM on the client to store those objects.
Something similar has been standardized as well with the JSON API specification (although it adds its own weight to the message as well, it does address this problem): http://jsonapi.org/format/#document-compound-documents
Its a great solution! I'd use dictionaries for faster lookups though.
But not exactly pure JSON. Client-side, you need to attach methods (or getters) that fetch the user object to the comment. I suppose you could just attach get(reference) that takes this[reference + '_id'] and looks it up inside the `result[reference]`. m:n relations will be harder though.
Otherwise you can't e.g. simply pass each comment to e.g.a react "Comment" component that renders it. You would also have to pass the user list (dictionary) to it.
Well, you could process the JSON on client side.
response.comments.forEach(comment => comment.author = response.user[comment.author]);
Preferably, even, you could do it during rendering so it can be garbage collected afterwards.Well, that counts as further deserialisation in my book. At least if you set up a convention to automate it for all kinds of objects. Otherwise you'd have to do it manually for every type of request
Or you send it like
Which is a standard way to do it that works here, too. Because you can have references in JSON, you just have to do them yourself.