Unable to cross join two views that depend on the same cube since 1.3.14 (SQL API)

[DReigada]

Describe the bug

Unable to cross join two views that depend on the same cube starting with version 1.3.14.

To Reproduce
Steps to reproduce the behavior:

1. Define cubes as shown in "Minimally reproducible Cube Schema"
2. Run query against SQL API

      select
          date,
          measure(view1.value) as measure1,
          measure(view2.value) as measure2
      from view1, view2
      group by 1

3. Error Arrow error: Compute error: Can't find join path to join 'view1,view2', 'cube1,cube1', 'cube1'

Expected behavior

The following result (screenshot from version 1.3.13)

Minimally reproducible Cube Schema

cubes:
  - name: cube1

    sql: |-
      select 1 as measure1, 2 as measure2, '2025-01-01' as date UNION ALL
      select 3 as measure1, 4 as measure2, '2025-01-02' as date UNION ALL
      select 5 as measure1, 6 as measure2, '2025-01-03' as date

    dimensions:
      - name: date
        type: time
        sql: TIMESTAMP({CUBE}.date)
        primary_key: true

    measures:
      - name: measure1
        type: number
        sql: sum({CUBE}.measure1)

      - name: measure2
        type: number
        sql: sum({CUBE}.measure2)

views:
  - name: view1
    cubes:
      - join_path: cube1
        includes:
          - name: measure1
            alias: value
          - date
  - name: view2
    cubes:
      - join_path: cube1
        includes:
          - name: measure2
            alias: value
          - date

Version:
Starting with 1.3.14

Additional notes:

If we explain the query we get the same plan for both versions, except that from version 1.3.14 we also get a joinHints key.
Either way, if we use the query plan in the playground, both plans work (we don't get the join error)

This is the plan for 1.3.14: {"measures":["view1.value","view2.value"],"dimensions":["view1.date"],"segments":[],"order":[],"joinHints":[["view1","view2"]]}

[igorlukanin]

@DReigada Could you please elaborate on your use case here? Why would you like to cross join views? The idea behind views is that all joins are baked into them and need not to happen in runtime.

[DReigada]

@igorlukanin we are using a metric-first approach to define our metrics (each metric is a separate view)

We want to be able to get multiple related metrics in a single SQL query (eg: revenue + bookings per day)
We assumed the cross join was the correct way of doing it (since it was working before and gave us the results we expected).

Is there better way to do it?

[igorlukanin]

We want to be able to get multiple related metrics in a single SQL query (eg: revenue + bookings per day)
[...]
Is there better way to do it?

Yes, that would be to expose both metrics/measures on the same view.

Views aren't supposed to be joined, it worked earlier by pure luck, I'd dare to say.

[igorlukanin]

I hope this helps, please feel free to reopen if there's something else!