Support for reusable datasets and relationships across semantic models

[olivierdet]

The current spec requires every semantic_model to fully re-declare its datasets and relationships inline. This makes it impossible to share a common dataset (e.g., a date_dim or customer dimension) across multiple semantic models without duplicating its definition.

Use case:

In most data warehouses, dimension tables (dates, customers, products, etc.) are shared across many subject areas. It should be possible to define a dataset once — with its fields, primary keys, and relationships — and reference it from multiple semantic models.
Without this, any real-world multi-domain semantic layer leads to significant duplication, drift between definitions, and maintenance burden — defeating the purpose of a single source of truth.

[bcp-software]

And .. perhaps a little off topic, 'single source of truth' is dated. At best we can reach a 'single source of definition'. Everything documented is a subjective perspective on the data (since it always has context) and therefor at best inter-subjective, and definitely not 'truth'.

Not trying to nitpick, but the language shapes the mind, and the correct mindset seems relevant in an open interchange standard.