[DRAFT] Inefficiency of ODataResource and ODataWriter and exploring alternative designs

[habbes]

Current serialization architecture in OData libraries

Our libraries currently use what I would call a "push" based serialization model, where the higher-level components push the data that they want written down to the lower-level components. The lower-level part of the serialization happens in the Microsoft.OData.Core library, higher level processes take place in Microsoft.AspNetCore.OData, Microsoft.OData.Client and/or the user's application.

The following diagram illustrates how data is serialized into JSON using OData libraries:

Let's say we have a web API based on Microsoft.AspNetCore.OData. The user application has a /People({id}) endpoint handled by an OData controller that returns an instance of type Person. The value returned by the controller will eventually find its way to an ODataResourceSerializer defined in Microsoft.AspNetCore.OData.

The ODataResourceSerializer will create an ODataResource object for this instance and populate it with properties and values from the Person instance taking into account the $select and $expand query options. The serializer might also add other metadata to the ODataResource, like annotations, navigation links, ETag, etc. Then it will pass the ODataResource to the ODataMessageWriter defined in Microsoft.OData.Core

...