Microsoft Fabric driver: pre-aggregation build crashes on NULL decimal values (Cannot read properties of null (reading 'toString'))

[jdart]

Describe the bug

Building a pre-aggregation against a Microsoft Fabric data source fails immediately whenever any decimal/numeric column in the rollup's result set contains a NULL value.

During the build, Cube downloads the source query results via the Fabric driver, and the driver's decimal type handler calls .toString() on the value without a null-check, throwing:

TypeError: Cannot read properties of null (reading 'toString')
    at Object.decimal (/cube/runtime/packages/cubejs-microsoftfabric-driver/src/MicrosoftFabricDriver.ts:66:14)
    at /cube/runtime/packages/cubejs-microsoftfabric-driver/src/MicrosoftFabricDriver.ts:270:55
    at Array.find (<anonymous>)
    at /cube/runtime/packages/cubejs-microsoftfabric-driver/src/MicrosoftFabricDriver.ts:270:16
    at Array.find (<anonymous>)
    at /cube/runtime/packages/cubejs-microsoftfabric-driver/src/MicrosoftFabricDriver.ts:266:41
    at Array.map (<anonymous>)
    at MicrosoftFabricDriver.downloadQueryResults (/cube/runtime/packages/cubejs-microsoftfabric-driver/src/MicrosoftFabricDriver.ts:263:26)
    at PreAggregationLoader.refreshReadOnlyExternalStrategy (/node_modules/@cubejs-backend/query-orchestrator/src/orchestrator/PreAggregationLoader.ts:734:19)
    at Object.query (/node_modules/@cubejs-backend/query-orchestrator/src/orchestrator/QueryCache.ts:658:26)

The build runs under the read-only external strategy (refreshReadOnlyExternalStrategy) with no export bucket configured. NULLs are extremely common in real models (e.g. a LAG()-derived prior-period column is NULL for the earliest period; any ratio measure is NULL on a zero denominator), so this effectively blocks pre-aggregating those measures.

To Reproduce

Steps to reproduce the behavior:

1. Configure a Cube deployment against a Microsoft Fabric data source.
2. Define a cube with a numeric (decimal) measure that is NULL for at least one row at the pre-aggregation's grain.
3. Add a pre-aggregation that includes that measure plus a time dimension + granularity.
4. Trigger the pre-aggregation build (e.g. from the Cube Cloud UI → Pre-Aggregations → Build).
5. The build fails in ~1s with the TypeError above.

Expected behavior

NULL values in decimal/numeric columns should be passed through as NULL — the type handler should null-check before calling .toString() — so pre-aggregations containing legitimately-null measures build successfully and preserve NULL semantics.

Screenshots

N/A — the Cube Cloud build-log error is included above.

Minimally reproducible Cube Schema

cube(`NullDecimalRepro`, {
  // Two months; February's amount is NULL, so the monthly rollup
  // materialises a NULL decimal -> driver crashes on build.
  sql: `
  SELECT CAST('2024-01-01' AS DATE) AS created_at, CAST(10.50 AS DECIMAL(10,2)) AS amount
  UNION ALL
  SELECT CAST('2024-02-01' AS DATE) AS created_at, CAST(NULL  AS DECIMAL(10,2)) AS amount
  `,

  measures: {
    maxAmount: {
      sql: `amount`,
      type: `max`,
    },
  },

  dimensions: {
    createdAt: {
      sql: `created_at`,
      type: `time`,
    },
  },

  preAggregations: {
    monthlyRollup: {
      measures: [maxAmount],
      timeDimension: createdAt,
      granularity: `month`,
    },
  },
});

Version:

Cube Cloud — 1.6.57, cubejs-microsoftfabric-driver.

Additional context

• The crash is in the read-only external pre-aggregation strategy (refreshReadOnlyExternalStrategy); the build had no export bucket configured.
• CAST(... AS FLOAT) does not work around it — the column's type changes but the value is still NULL, and it's still routed through the crashing handler. The only schema-level workaround is COALESCE(..., <sentinel>), which is lossy: it destroys legitimate NULL semantics (e.g. it breaks WHEN col IS NULL THEN ... logic and forces a fake 0/sentinel for "no data" periods).
• Likely fix: null-check in the decimal (and probably the other numeric) type handler(s) used by downloadQueryResults before calling .toString().