Table Customization

Pass columns= to proof.dataset_strategy() or the @sqlproof decorator to override selected columns while keeping the rest of the dataset schema-aware. The keys are "table.column"; values can be Hypothesis strategies, fixed values, or callables receiving a ColumnContext and returning a value.

from hypothesis import strategies as st

datasets = proof.dataset_strategy(
    sizes={"products": 3},
    columns={
        "products.price": st.decimals(min_value="0.01", max_value="9999.99", places=2),
        "products.sku": st.from_regex(r"[A-Z]{2}-[0-9]{4}", fullmatch=True),
    },
)

Draw a dataset inside a Hypothesis test and use proof.client_for_dataset(dataset) to load it. Overrides must satisfy the schema’s constraints. An override replaces the normal generator, including CHECK refinement and nullable-column handling. Use st.just(None) to force SQL NULL in a nullable column.

Default-bearing columns are omitted unless requested explicitly. Single-column primary keys use a separate assignment path and ignore ordinary columns= overrides. A scenario can pin relationships to named rows’ generated keys; it does not provide arbitrary primary-key values.

proof.customize() is currently a placeholder and does not apply overrides. Use columns= for working customizations.

See custom generators for default sampling limits and edge-case recipes, and FK distributions for controlling parent selection.