🐛 Bug: FILTER_PARAMS not populated for inDateRange filter on joined cube (regression from v0.33 to v1.3.36)

[divij123]

Describe the bug

We have two cubes:

• User (base cube)
• Conversation (joined via User.id = Conversation.user_id, 1:many)

Our business use case requires fetching all Users who have done 0 conversations within the last 30 days.
To do this, we need to apply a subquery filter on the Conversation's created date range when it’s joined to User.

This used to work correctly in Cube.js v0.33, but after upgrading to v1.3.36, the generated SQL no longer receives populated FILTER_PARAMS for the inDateRange filter, which breaks our subquery-based approach.

To Reproduce
Steps to reproduce the behavior:

1. Start the playground server (make sure you're on v1.3.36)
2. Copy and add this query

{
  "measures": [
    "Conversation.count_include_zero"
  ],
  "dimensions": [
    "User.id"
  ],
  "timeDimensions": [],
  "timezone": "Asia/Kolkata",
  "limit": 5000,
  "filters": [
    {
      "member": "User.organization_id",
      "operator": "equals",
      "values": ["1"]
    },
    {
      "or": [
        {
          "member": "Conversation.created",
          "operator": "inDateRange",
          "values": [
            "2025-06-15T00:00:00.000",
            "2025-07-14T23:59:59.999"
          ]
        },
        {
          "member": "Conversation.created",
          "operator": "notSet"
        }
      ]
    }
  ]
}

3. Go to Generated SQL, you'll notice that the FILTER_PARAMS did not get populated in the subquery

Expected behavior
Cube.js should generate SQL similar to:

SELECT
  "user".id,
  count("conversation".id) as conversation_count 
FROM tenant_user as "user"
  LEFT JOIN (
    SELECT * FROM tenant_conversation
    WHERE 1=1 AND (
      (created < ? AND created > ?)
      OR created IS NULL
    )
  ) as "conversation"
    ON "user".id = "conversation".user_id
WHERE
  "user".organization_id = 1
  AND (
    (
      "conversation".created >= ? :: timestamp_tz
      AND "conversation".created <= ? :: timestamp_tz
    )
    OR ("conversation".created IS NULL)
  )
GROUP BY 1

Screenshots
Bug screenshot in the latest version

Minimally reproducible Cube Schema
In case your bug report is data modelling related please put your minimally reproducible Cube Schema here.
You can use selects without tables in order to achieve that as follows.

// cubes/User.js
cube(`User`, {
  sql: `SELECT * FROM tenant_user`,

  joins: {
    Conversation: {
      relationship: `one_to_many`,
      sql: `${CUBE}.id = ${Conversation.user_id}`
    }
  },

  measures: {
    count: {
      type: `count`,
      drillMembers: [id]
    }
  },

  dimensions: {
    id: {
      sql: `id`,
      type: `number`,
      primaryKey: true
    },
    organization_id: {
      sql: `organization_id`,
      type: `number`
    }
  }
});

// cubes/Conversation.js
cube(`Conversation`, {
  sql: `SELECT * FROM tenant_conversation where ${FILTER_PARAMS.Conversation.created.filter("created")}`,

  measures: {
    count_include_zero: {
      type: `count`,
      sql: `id`,
    }
  },

  dimensions: {
    id: {
      sql: `id`,
      type: `number`,
      primaryKey: true
    },
    user_id: {
      sql: `user_id`,
      type: `number`
    },
    created: {
      sql: `created`,
      type: `time`
    }
  }
});

Version:
[1.3.36]

Additional context

❌ Actual behavior

Since upgrading to v1.3.36:

• The FILTER_PARAMS.Conversation.created.inDateRange is missing or empty.
• This breaks our count_include_zero measure that relies on it to apply the subquery date filter.
• The generated SQL does not include the correct subquery or date filter.

---

📦 Regression

Version	Status
v0.33	✅ Works
v1.3.36	❌ Broken

---

🛠 Environment

Item	Version
Cube.js	v1.3.36
Database	Snowflake
Node.js	22.14.0

---

This functionality is critical for supporting zero data reporting on users in a specific time range, which is a common business scenario.

[igorlukanin]

Hi @divij123 👋

I've just checked your data model and query with 1.3.39. Here's the generated SQL:

SELECT
  "user".id "user__id",
  count("conversation".id) "conversation__count_include_zero"
FROM
  tenant_user AS "user"
  LEFT JOIN (
    SELECT
      *
    FROM
      tenant_conversation
    where
      (
        (
          created >= ? :: timestamp_tz
          AND created <= ? :: timestamp_tz
        )
        OR (created IS NULL)
      )
  ) AS "conversation" ON "user".id = "conversation".user_id
WHERE
  ("user".organization_id = ?)
  AND (
    (
      "conversation".created >= ? :: timestamp_tz
      AND "conversation".created <= ? :: timestamp_tz
    )
    OR ("conversation".created IS NULL)
  )
GROUP BY
  1
ORDER BY
  2 DESC
LIMIT
  5000

As you can see, all filters are in place. So, I'd invite you to try the latest version.

[igorlukanin]

Please feel free to reopen if this is not fixed for you.