[CT-3176] [spike] infer fixture data types for unit testing

[graciegoheen]

Housekeeping

• I am a maintainer of dbt-core

Short description

From #8649

Instead of requiring folks to run dbt run --select +my_model --exclude my_model before executing unit tests on my_model or requiring the use of --deferral (which would cause issues in a CI context when you're deferring to a production environment and may have changes to your inputs as part of the PR)...

we should be able to run dbt unit-test --select my_model in a new environment without having previously run dbt run --select +my_model --exclude my_model or dbt docs generate

• we will use the data_types defined in the direct parent's yml config
• if no data_type is provided...
  • we will have the warehouse "presume" the data type based on the mock data provided (users could put cast statements in their fixture)
  • if no mock data is provided for a column, we will guess that the data type is a string
    raise a warning during unit test execution if an input fixture specifies columns that don’t have user-provided data_type values.

We will need to confirm when we should add '' around the mocked data.

We can get all column names from adapter.get_column_schema_from_query so users don't have to provide a yml name for all columns of direct parents.

Acceptance criteria

• I can run dbt unit-test --select my_model in a new environment without having previously run dbt run --select +my_model --exclude my_model or dbt docs generate
• edit test to remove the build prior to executing unit tests https://github.com/dbt-labs/dbt-core/blob/unit_testing_feature_branch/tests/functional/unit_testing/test_unit_testing.py#L123-L124

Impact to Other Teams

No

Will backports be required?

No

Context

No response

[graciegoheen]

Let's make sure we get feedback on this during the beta!

[graciegoheen]

The goal of this spike is to determine an appropriate path forward - additional ideas tracked here #8649

[graciegoheen]

Opened - #8979 and #8980

[graciegoheen]

Our assumption:

We can get all column names from adapter.get_column_schema_from_query so users don't have to provide a yml name for all columns of direct parents.

was incorrect. So closing in favor of:

• use --empty to build "empty" versions of models
• dbt build includes unit tests to materialize upstream nodes before running unit tests