Nested CTE Query

Question Is it possible to emit nested CTE blocks in a query? (postgres dialect)

Example Here’s a minimal example of a nested CTE query.

with res_cte as (
    select
        account_0.name acct_name,
        (
            with offer_cte as (
                select
                    offer_0.id
                from
                    offer offer_0
                where
                    offer_0.account_id = account_0.id
            )
            select
                array_agg(offer_cte.id)
            from
                offer_cte
        ) as offer_arr
    from
        account account_0
)
select 
    acct_name, offer_arr
from res_cte

In my attempts, the emitted SQL unnests the CTEs which results in an incorrect scope for the offer_0.account_id=account_0.id join

The specific usecase is returning (maybe deeply) joined data as a single json object.

Useful links https://stackoverflow.com/questions/58513337/sqlalchemy-nested-cte-query

[zzzeek]

it might not be impossible to do nested CTEs but I would need some simple and straightforward example that can show what they accomplish, as this would drive how the feature would be implemented from an API point of view and how we would describe the use case. at the moment I don’t have plans to work on this but certainly if we can make the case a little more simply and there’s a good reason for this, it’s possibly doable.

[zzzeek]

Here’s the thing about CTEs. In SQLAlchemy, you get the same syntactical approach to using a CTE as you do to a subquery, it’s literally the difference between calling cte() or subquery() on the objects. The reason cte() is helpful is because the “subquery” is independent of the rest of the query, can affect planning results and things like that.

but in this case we can take the nested CTE and just turn it into a subquery:

with res_cte as (
    select
        account_0.email acct_email,
        (
            select
                array_agg(post_cte.id)
            from
                (
                select
                    post_0.id
                from
                    post post_0
                where
                    post_0.account_id = account_0.id
            ) post_cte
        ) as post_arr
    from
        account account_0
)
select 
    acct_email, post_arr
from res_cte

if you are “nesting” anyway, there’s not much difference.

another form features the two CTEs as siblings but then you have to do the “correlated subquery” part in a different place which I guess is what you’re trying to avoid:

with post_cte as (
    select
        post_0.id, post_0.account_id
    from
        post post_0
),
res_cte as (
    select
        account_0.email acct_email,
        (
            select
                array_agg(post_cte.id)
            from
                post_cte where post_cte.account_id = account_0.id
        ) as post_arr
    from
        account account_0
)
select 
    acct_email, post_arr
from res_cte

I haven’t myself gotten into the “CTE habit” yet because when you write SQLAlchemy queries you usually can hardly notice if the object your SELECTing from is a cte() or an subquery(). I understand that it does change query planner results though.