dbt-labs/dbt-core · error

Datafusion

Error message

Datafusion

What it means

A `todo!()` panic in `ColumnBuilder::build` (column_builder.rs:41) for the Datafusion adapter. The adapter dispatch has no Datafusion column-builder implementation, so schema-to-Column conversion with Datafusion panics with the message 'Datafusion'.

Source

Thrown at crates/dbt-adapter/src/column/column_builder.rs:41

    pub fn build(&self, field: &FieldRef, type_ops: &dyn TypeOps) -> AdapterResult<Column> {
        use AdapterType::*;
        match self.adapter_type {
            Snowflake => Ok(Self::build_snowflake(field, type_ops)),
            Bigquery => Ok(Self::build_bigquery(field, type_ops)),
            Databricks | Spark => Ok(Self::build_databricks(field, type_ops)),
            Redshift => Ok(Self::build_redshift(field, type_ops)),
            Postgres | Salesforce | DuckDB | LakeCompute => {
                Ok(Self::build_postgres_like(field, type_ops))
            }
            Fabric => Ok(Self::build_fabric(field, type_ops)),
            ClickHouse => Self::build_clickhouse(field, type_ops),
            Exasol => Ok(Self::build_exasol(field, type_ops)),
            Starburst => todo!("Starburst"),
            Athena => todo!("Athena"),
            Trino => todo!("Trino"),
            Dremio => todo!("Dremio"),
            Oracle => todo!("Oracle"),
            Datafusion => todo!("Datafusion"),
        }
    }

    pub fn build_from_parts(
        &self,
        name: String,
        dtype: String,
        char_size: Option<u32>,
        numeric_precision: Option<u64>,
        numeric_scale: Option<u64>,
        mode: Option<BigqueryColumnMode>,
    ) -> Column {
        use AdapterType::*;
        match self.adapter_type {
            Postgres => Column::new(
                Postgres,
                name,
                dtype,

View on GitHub (pinned to 0267ce9170)

Solutions

  1. Use a supported adapter (Postgres-like, Fabric, ClickHouse, Exasol) until Datafusion is wired up.
  2. Implement the Datafusion arm by delegating to `build_postgres_like` — Datafusion types are Arrow-native and map closely to the postgres_like path.
  3. Convert the panic into a typed unsupported-adapter error.

Example fix

// before
Datafusion => todo!("Datafusion"),
// after
Datafusion => Ok(Self::build_postgres_like(field, type_ops)),
Defensive patterns

Strategy: type-guard

Validate before calling

if adapter == AdapterType::Datafusion {
    return Err(anyhow!("Datafusion column building is not implemented yet"));
}

Type guard

fn build_supported(t: AdapterType) -> bool {
    matches!(t, Postgres | Salesforce | DuckDB | LakeCompute | Fabric | ClickHouse | Exasol)
}

Try / catch

std::panic::catch_unwind(AssertUnwindSafe(|| builder.build(field, type_ops)))
    .map_err(|_| anyhow!("column building unimplemented for Datafusion"))?

Prevention

When it happens

Trigger: Any call into ColumnBuilder::build (schema_to_columns, get_column_schema_from_query, update_node_columns) while the adapter type is Datafusion — e.g. introspecting the schema of a local/Arrow-backed query result.

Common situations: Using the embedded Datafusion engine for testing or local execution and then requesting column metadata from results; also hit by tests iterating all adapter variants.

Related errors

AI-assisted analysis of dbt-labs/dbt-core@0267ce9170 (2026-09-07). Data as JSON: /api/errors/36d7bd4ad997512a. Report an issue: GitHub.