dbt sync: composite primary key test uses double-quoted identifiers, which Databricks/Spark read as string literals, so the test always fails #1713

Summary

For a schema with a composite primary key, datacontract dbt sync generates a singular test that quotes column names with double quotes. Databricks and Spark SQL read a double-quoted value as a string literal, not as an identifier. The query then groups every row under one constant tuple, so on any table with more than one row it returns exactly one row and the test fails. Data that is actually unique still fails.

Version

• datacontract-cli 1.2.0. The same code is on main, so 1.2.4 is affected too.
• dbt-core 1.11, dbt-databricks 1.12.1

Steps to reproduce

Contract excerpt (ODCS v3.1.0):

schema:
- name: my_fact
  properties:
  - name: machine_number
    logicalType: integer
    physicalType: int
    required: true
    primaryKey: true
    primaryKeyPosition: 1
  - name: calendar_date
    logicalType: date
    physicalType: date
    required: true
    primaryKey: true
    primaryKeyPosition: 2
  - name: shift
    logicalType: string
    physicalType: string
    required: true
    primaryKey: true
    primaryKeyPosition: 3

Run datacontract dbt sync and then dbt build on Databricks.

Generated test

SELECT "machine_number", "calendar_date", "shift"
FROM {{ ref('my_fact') }}
GROUP BY "machine_number", "calendar_date", "shift"
HAVING COUNT(*) > 1

Actual result

Failure in test my_fact__1_0_0__my_fact__unique_combination
  Got 1 result, configured to fail if != 0

The equivalent dbt_utils.unique_combination_of_columns test on the same table passes.

Cause

_quote_identifier in datacontract/integration/dbt_sync.py always uses ANSI double quotes, whatever the target adapter:

def _quote_identifier(name: str) -> str:
    return '"' + name.replace('"', '""') + '"'

_composite_pk_singular_test uses it for both the SELECT and GROUP BY lists. On Databricks/Spark, unless spark.sql.ansi.doubleQuotedIdentifiers is enabled, "machine_number" is the string 'machine_number'.

Workaround

Do not declare a composite primaryKey in the contract, and keep a dbt_utils.unique_combination_of_columns test in the model YAML.