Nice API! My biggest critic is that people will want to paginate on models that dont have a natural order via the PK. And supporting that will require a breaking change.

I am a little confused that why there isn't any raw SQL unit tests to illustrate the behaviour
And that the mock test part is not proving anything

Nice API! My biggest critic is that people will want to paginate on models that dont have a natural order via the PK. And supporting that will require a breaking change.

Hey @Sytten, I have updated the API to... which user have full control over what column to filter

Entity::find()
    .order_by_asc(Column::Id)
    .cursor()
    .after(Column::Id, 10)
    .last(2)
    .all(&db)
    .await?,

Hey @tyt2y3, unit tests were being updated and now it asserts on plain SQL string

What about order by non-unique keys? 🤔

What about order by non-unique keys? 🤔

You always need a column that is totally orderable for cursor pagination, even if you order by a partially orderable one.

Thank you so far. Great work!

I think there are 2 things we need to address:

Instead of "inheriting" the order by clause, we should override it by the cursor column the user passes in via the cursor() method, and so we don't assume it is ordered already, and don't have to flip-flop with that "reverse ordering" thing.

The Cursor itself should hold which column to paginate on. As a result, the API can be simplified into:

Entity::find()
    .cursor(Column::Id)
    .after(10)
    .last(2)
    .all(&db)
    .await?

And by the way, is there a real use case where we want to have a tuple of columns as cursor? We can always add that later with IntoCursor and impl IntoCursor for (A, B) trick we have been doing all along.

@tyt2y3 like I commented above, you need to support two column if you want to support partially orderable columns. If not then you can use one column.

oh I didn't merge the sea-query side yet

I reverted this merge. We need to upgrade sea-query to 0.25 first.
As stated before, git = XXX can never be merged into master to avoid dependency confusion.