[Enhancement]: $first & $after keywords in REST & GQL

[JerryNixon]

This introduces $take/$skip in DAB REST and GraphQL endpoints.

Previous discussions had this feature as a synonym of $first/$after, but that is no longer the plan. The current plan is, instead, to introduce these as new keywords with similar, but not identical, functionality.

	Data type	Behavior	REST	GraphQL	Current Status
$first	integer	TOP in T-SQL	✅	✅	Yes, DAB supports this today
$after	string	Entity key(s)	✅	✅	Yes, DAB supports this today
$take	integer	TOP in T-SQL (same as $first)	✅	✅	No (this article)
$skip	integer	OFFSET in T-SQL	✅	✅	No (this article)

Syntax

query {
  books(first: 5, after: "W3siVm1lIjoiaWQifV0=")
  {
    items {
      id
      title
    }
    hasNextPage
    endCursor
  }
}

query {
  books(take: 5, skip: 50)
  {
    items {
      id
      title
    }
    hasNextPage
  }
}

Note that $first, $take, and (later) $top are all synonyms of the same operations.

Benefits

The value of the $first/$after approach is that it allows developers to paginate in a reliable way. When the underlying collection changes, the next page is always the next page and does not reset because of ordinal position. In a busy table, this provides a consistent and non-jarring user experience.

The value of the $take/$skip approach is that it allows developers to paginate in a simple way, often in order to virtualize large datasets in visualization objects. It does not account for underlying collection changes but does return a programmatic page. It operates identically to the Take() and Skip() LINQ methods in .NET languages.

Bonus: $take/$skip are both keywords in OData and represent common expressions.

Proposed synonyms

The reason for synonyms would be to accommodate the OData specification.

Synonym	Keyword	REST	GraphQL	Current Status
$top	$first	✅	✅	No (this article)
$skiptoken	$after	✅	✅	No (this article)

Feedback

Customers have reported that using GraphQL keywords in REST endpoints instead of OData standard keywords causes some code to struggle when interacting with our endpoints. Other customers have reported that not supporting integer-based $take/$skip disqualifies some UI components looking for the standard.

[raffaeler]

Great to see you logging this @JerryNixon. Let me expand my initial request and add some more detail in relation to what you wrote here.

Revised specifications

In your table there is a semantic mismatch between $skip in oData and offset in GraphQL as for the official standards. Let me clarify with the following table whose columns specify:

1. the oData standard specs
2. the Graphql standard specs
3. the current syntax for REST in DAB
4. the current syntax for GraphQL in DAB

	oData	GraphQL	DAB/REST	DAB/GraphQL
Page size	$top=number	first:number	$first=number	first:number
Offset to skip	$skip=number	offset:number	N/A	N/A
Identifier to skip	$skiptoken=string	after:string	N/A	after:string

The perfect solution is to fully adhere to both standards by:

• filling the three N/A holes
  • the offset keyword in GraphQL is also important
• renaming the DAB/Rest $first to $top
  • This is a breaking changes and may hurt a bit at the beginning, but on the long term is definitely less confusing

Motivations

1. Support the official oData and GraphQL keywords
2. Allow replacing the WCF Data Services (oData) with DAB
3. Allow replacing the ASP.NET Core OData implementation with DAB (see (A))
4. The current .NET oData client library offers a Linq provider which transforms the Expression Trees into the oData query string. This string may include $skip and $top standard operators which should be therefore understood from the backend to correctly adhere to the client request.

Notes

(A) ASP.NET Core oData automatically exposes the CRUD for the tables but all the navigations must be manually coded. This is a major disadvantage compare to the old WCF Data Services and DAB.

HTH

[JerryNixon]

Remember @raffaeler, we need a solution that is backwards compatible with existing users of DAB. I have revised the plan but still need to talk to the engineering team about this, too. Would appreciate any feedback.

[raffaeler]

I understand the backward compatibility issues Jerry, but the compliance with standards should always come first. Even more when the original draft came from MS, even if oData was then given to oasis.

I am open to discuss but I don't really see great options to avoid the breaking change out of versioning the api with some headers.