feat: PageInfo is nullable

[maraisr]

PageInfo is a nullable entity, making all members NonNull please see https://relay.dev/graphql/connections.htm#sec-undefined.PageInfo.Fields

[sikanhe]

@n1ru4l You are familiar with Relay, can you help take a look?

[sikanhe]

I remmeber this - I ported the behavior from graphql-js graphql/graphql-relay-js#243

The reason cursors are nullable is because it's unclear what they should be when array is empty. But if array is empty, we can still get information about whether there is next page. So I am not sure that returning null for PageInfo is correct when array.length === 0

[maraisr]

The spec clearly says the cursors must be non-null, see link I attached in pr description.

The field is aptly named "PageInfo" alluding to the fact it's information about a page, if there is no page — aka edges is empty — then the field (PageInfo) is null.

There is chatter about the repo you linked "graphql-relay-js" so still waiting on an official answer.

For now, maybe our best bet is to check what our c# friends are doing with HotChocolate.

[n1ru4l]

https://relay.dev/graphql/connections.htm#sec-Connection-Types.Fields.PageInfo

A “Connection Type” must contain a field called pageInfo. This field must return a non-null PageInfo object, as defined in the “PageInfo” section below.

[maraisr]

You're a tad hypocritical here mate. How can you drop the spec on me here and ignore it on the other moment?

[n1ru4l]

In the case of an empty list, I tend to use "" as the start/end cursor values, as the pageInfo field is non-nullable per spec. However, there is ambiguity on what to do.

See https://stackoverflow.com/q/70448483, facebook/relay#3708

Hot Chocolate seems to use nullable end and start cursors: https://github.com/ChilliCream/hotchocolate/blob/c479c10343f22328f708dc4d8e80f98a76bbf518/src/HotChocolate/Core/src/Types.CursorPagination/PageInfoType.cs#L29-L39

java too: https://github.com/graphql-java/graphql-java/blob/1ee9630c7d25ccb55a5b1d9b1ccf7bf7895a8c7f/src/main/java/graphql/relay/Relay.java#L54

---

Furthermore, ID != cursor, so it might make sense to to have some kind of cursorFromRecord function. IMHO this can be part of a follow-up PR.

[maraisr]

You're just giving me opinions.

The spec clearly says what must happen?

[sikanhe]

Closing this as all other relay implementations do the same

[maraisr]

Just because they all do the same, doesn't make it right.