Add ca_yaml_features macro: YAML-only features (time_dimensions, filters, data_type) in DDL

[cletoigor]

Summary

Brings the semantic-view features that are expressible in the Cortex Analyst YAML spec but not in plain CREATE SEMANTIC VIEW DDL — time_dimensions, standalone filters, and data_type declarations (YAML vs. DDL) — into the DDL path.

• Adds a ca_yaml_features Jinja macro (macros/ca_yaml_features.sql) that generates the WITH EXTENSION (CA=$$...$$) clause carrying these fields, grouping entries by table and emitting only populated keys.
• Wires the same three inputs through config keys (time_dimensions, ca_filters, ca_dimensions) in macros/relations/semantic_view/create.sql, so both an inline-macro form and a config-driven form work. A guard raises a compiler error if config-driven usage is combined with a hand-written WITH EXTENSION clause (a single CA extension is allowed). Runs before the COPY GRANTS append so COPY GRANTS stays last.
• Adds integration-test models for both forms (semantic_view_with_yaml_features.sql, semantic_view_with_yaml_features_config.sql) and a get_ddl-based data test (semantic_view_yaml_features_has_extension.sql).
• Adds an offline Jinja render test (scripts/test_ca_yaml_features.py, only requires jinja2) that exercises the real macro with no Snowflake/dbt dependency.
• Documents the feature in the README under a new "YAML-only features in DDL" section with a parameter table and both usage forms.

Motivation

Per Snowflake's YAML vs. DDL comparison, a few semantic-view features live only in the Cortex Analyst YAML spec: time_dimensions (a distinct category for date/timestamp columns), standalone filters (named table-level filter expressions), and explicit data_type declarations. In Snowflake these fields are carried in the CA extension, which a DDL-created view can attach via WITH EXTENSION (CA=$$ ...json... $$).

The existing semantic_view_with_ca_extension.sql integration test shows that users currently have to hand-write raw, escaped JSON in the with extension clause to reach these fields — error-prone and hard to maintain. This macro lets users declare them as structured Jinja dicts and relies on | tojson for correct serialization, bringing the YAML-only feature set into the package's DDL path.

Usage

Inline:

{{ config(materialized='semantic_view') }}
TABLES(orders AS {{ ref('orders') }})
DIMENSIONS(orders.status AS status)
METRICS(orders.total AS SUM(orders.amount))
{{ dbt_semantic_view.ca_yaml_features(
    time_dimensions=[{'table': 'orders', 'name': 'order_ts', 'expr': 'ORDER_TS', 'data_type': 'TIMESTAMP_NTZ'}],
    filters=[{'table': 'orders', 'name': 'recent', 'expr': 'order_ts > dateadd(day, -30, current_date)'}],
    dimensions=[{'table': 'orders', 'name': 'amount', 'expr': 'AMOUNT', 'data_type': 'NUMBER(38,2)'}]
) }}

Config-driven (equivalent result):

{{ config(
    materialized='semantic_view',
    time_dimensions=[{'table': 'orders', 'name': 'order_ts', 'expr': 'ORDER_TS', 'data_type': 'TIMESTAMP_NTZ'}],
    ca_filters=[{'table': 'orders', 'name': 'recent', 'expr': 'order_ts > dateadd(day, -30, current_date)'}],
    ca_dimensions=[{'table': 'orders', 'name': 'amount', 'expr': 'AMOUNT', 'data_type': 'NUMBER(38,2)'}]
) }}
TABLES(orders AS {{ ref('orders') }})
DIMENSIONS(orders.status AS status)
METRICS(orders.total AS SUM(orders.amount))

Testing

Offline render test (python scripts/test_ca_yaml_features.py, requires only jinja2) — 9/9 checks passed. Sample output:

WITH EXTENSION (CA=$${"tables": [{"name": "orders", "time_dimensions": [{"name": "order_ts", "expr": "ORDER_TS", "data_type": "TIMESTAMP_NTZ"}], "dimensions": [{"name": "amount", "expr": "AMOUNT", "data_type": "NUMBER(38,2)"}], "filters": [{"name": "recent", "expr": "order_ts > dateadd(day, -30, current_date)"}]}]}$$)

Checks cover: WITH EXTENSION (CA=$$...$$) wrapping, per-table grouping, time_dimensions with data_type, standalone filters, data_type on a regular dimension, stripping of the routing-only table key, empty-input → '', and the missing-table-key compiler error.

Live Snowflake — verified end-to-end on a Snowflake sandbox schema via dbt build:

• semantic_view_with_yaml_features (inline form) — created successfully
• semantic_view_with_yaml_features_config (config-driven form) — created successfully
• semantic_view_yaml_features_has_extension — passed, confirming via get_ddl that the live CREATE SEMANTIC VIEW DDL actually contains the CA extension with time_dimensions / filters / data_type (not just the offline-rendered string)

The live run selected the new models plus ancestors. A full dbt build --target snowflake (no --select) is still worth running in CI to confirm nothing else regressed; note the unrelated semantic_view_sum_matches_base_table test depends on semantic_view_basic and will error only if that ancestor is out of the selected scope.