vgi-python 0.48.0__tar.gz → 0.48.2__tar.gz

{vgi_python-0.48.0 → vgi_python-0.48.2}/.github/workflows/ci.yml

@@ -41,10 +41,10 @@ jobs:
           bun-version: "1.3.14"
 
       - name: Install dependencies
-        run: uv sync --all-extras
+        run: uv sync --all-extras --python ${{ matrix.python-version }}
 
       - name: Run tests
-        run: uv run pytest
+        run: uv run --python ${{ matrix.python-version }} pytest
 
   lint:
     name: Lint
@@ -147,6 +147,26 @@ jobs:
       - name: Run hosted-protocols group (stdio, unix, http)
         run: scripts/hosted_protocols_conformance.sh
 
+  bare-install:
+    # Every other job syncs --all-extras, which hides a core-dependency gap: a
+    # worker that installs vgi-python with no extras must still import the
+    # modules catalog_attach reaches (0.48.0 shipped without pycryptodome).
+    name: Bare install (no extras)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v10.2.0
+
+      - name: Build wheel
+        run: uv build --wheel --out-dir dist .
+
+      - name: Import attach-path modules from the bare wheel
+        run: |
+          uv run --isolated --no-project --python 3.14 --with dist/*.whl \
+            python -c "import vgi, vgi.worker, vgi.query_execution; print('ok')"
+
   s3-offload-localstack:
     name: S3 Offload Tests (LocalStack)
     runs-on: ubuntu-latest

{vgi_python-0.48.0 → vgi_python-0.48.2}/.github/workflows/integration.yml

@@ -27,7 +27,7 @@ concurrency:
 env:
   # The test suite and native Actions artifact must come from this exact ref.
   # The downloader verifies its provenance against the checked-out commit.
-  VGI_REF: a83ff7e8cd46aa840a10e78fe9b31603931824c2
+  VGI_REF: 7b1e1fbbe745d77fdacf52d6d135fa3ef4491d48
   VGI_NATIVE_RUNNER_BRANCH: release/vgi-v3-ci-20261028
   # CI is not a user: its ATTACHes should not land in Query.Farm's adoption data.
   # The vgi extension fires a telemetry event per successful ATTACH, and this

{vgi_python-0.48.0 → vgi_python-0.48.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: vgi-python
-Version: 0.48.0
+Version: 0.48.2
 Summary: Vector Gateway Interface - Connect DuckDB to external programs via Apache Arrow
 Project-URL: Homepage, https://query.farm
 Project-URL: Repository, https://github.com/Query-farm/vgi-python
@@ -161,6 +161,7 @@ Requires-Dist: click
 Requires-Dist: httpx2>=2.12.0
 Requires-Dist: platformdirs
 Requires-Dist: pyarrow
+Requires-Dist: pycryptodome>=3.20
 Requires-Dist: typer>=0.9
 Requires-Dist: vgi-rpc>=0.52.3
 Provides-Extra: azure

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/catalog-interface.md

@@ -122,9 +122,8 @@ Information about a table:
 | `name` | `str` | Table name |
 | `schema_path` | `SchemaPath` | Parent schema path |
 | `columns` | `SerializedSchema` | Column definitions as Arrow schema bytes |
-| `not_null_constraints` | `list[int]` | Column indices with NOT NULL |
-| `unique_constraints` | `list[list[int]]` | Column index groups for UNIQUE |
-| `check_constraints` | `list[str]` | SQL check expressions |
+| `constraints` | `list[bytes]` | Canonical `ConstraintInfo` records (see [Constraints](#constraints)) |
+| `definition_revision` | `bytes \| None` | Opaque table-definition revision |
 | `comment` | `str \| None` | Optional description |
 | `tags` | `dict[str, str]` | Key-value metadata |
 
@@ -509,7 +508,7 @@ For most use cases, the declarative catalog API provides a simpler way to define
 
 ```python
 from vgi import Worker, TableFunctionGenerator
-from vgi.catalog import Catalog, Schema, Table, View
+from vgi.catalog import Catalog, Schema, Table, View, constraint_record
 ```
 
 ### Function-Backed Tables (Recommended)
@@ -527,7 +526,7 @@ class UsersFunction(TableFunctionGenerator):
     """Generate user data."""
 
     FIXED_SCHEMA: ClassVar[pa.Schema] = pa.schema([
-        ("id", pa.int64()),
+        pa.field("id", pa.int64()),  # served NOT NULL: the PRIMARY_KEY record below names it
         ("name", pa.string()),
         ("active", pa.bool_()),
     ])
@@ -545,8 +544,16 @@ class UsersFunction(TableFunctionGenerator):
 users_table = Table(
     name="users",
     function=UsersFunction,  # Schema derived from FIXED_SCHEMA
-    not_null=["id"],         # Constraint column names validated
-    unique=[["id"]],
+    constraints=(            # Paths and nullability validated at definition time
+        constraint_record(
+            ["main"], "users", "users.id.not_null", "NOT_NULL", [["id"]],
+            enforcement="ENFORCED", validation="ALL_ROWS",
+        ),
+        constraint_record(
+            ["main"], "users", "users.id.unique", "UNIQUE", [["id"]],
+            enforcement="ENFORCED", validation="ALL_ROWS",
+        ),
+    ),
     comment="User accounts",
 )
 ```
@@ -593,7 +600,7 @@ if __name__ == "__main__":
 | Feature | Description |
 |---------|-------------|
 | **No schema duplication** | Function-backed tables derive schema automatically |
-| **Constraint validation** | `not_null`, `unique` column names validated at definition time |
+| **Constraint validation** | DuckDB's ATTACH rules for records and generated columns checked at definition time |
 | **Automatic scan handling** | Function-backed tables don't need `table_scan_function_get()` |
 | **Type safety** | Frozen dataclasses with runtime validation |
 
@@ -606,16 +613,98 @@ For tables not backed by functions, provide the schema explicitly:
 config_table = Table(
     name="config",
     columns=pa.schema([
-        ("key", pa.string()),
+        pa.field("key", pa.string(), nullable=False),
         ("value", pa.string()),
     ]),
-    not_null=["key"],
-    unique=[["key"]],
+    constraints=(
+        constraint_record(
+            ["main"], "config", "config.key.pk", "PRIMARY_KEY", [["key"]],
+            enforcement="ENFORCED", validation="ALL_ROWS",
+        ),
+    ),
 )
 ```
 
 **Note:** Tables with explicit columns require the worker to implement `table_scan_function_get()` to tell DuckDB how to scan the data.
 
+### Constraints
+
+A table states its constraints as canonical records built with
+`constraint_record()`. Each record is the worker's own attestation — the SDK
+never derives one — so `enforcement` and `validation` are always explicit:
+
+```python
+from vgi.catalog import ReferencedKey, constraint_record
+
+constraint_record(
+    ["main"], "orders", "orders.customer.fk", "FOREIGN_KEY", [["customer_id"]],
+    enforcement="ENFORCED", validation="ALL_ROWS",
+    references=ReferencedKey(
+        schema_path=["main"], table="customers", columns=[["id"]], constraint_id="customers.pk",
+    ),
+)
+constraint_record(
+    ["main"], "orders", "orders.amount.check", "CHECK", [],
+    enforcement="ENFORCED", validation="ALL_ROWS", check_sql="amount >= 0",
+)
+```
+
+DuckDB rejects a table at ATTACH unless its records follow the extension's
+rules. `Table` checks them at definition time, `ReadOnlyCatalogInterface` checks
+ownership and foreign-key targets across the catalog when it builds its
+registries, and the worker checks every table again when serving it (so a
+custom catalog's own unit tests and RPC calls see the same errors):
+
+- Only `ENFORCED` + `ALL_ROWS` records; state one only when every row the
+  table can ever return satisfies it.
+- Each record is owned by its table and schema; ids are unique, declared names
+  are unique case-insensitively, and there is at most one `PRIMARY_KEY`.
+- Column paths resolve. Only `NOT_NULL` may name a nested struct child;
+  `PRIMARY_KEY`, `UNIQUE` and `FOREIGN_KEY` name top-level columns, never the
+  `is_row_id` column.
+- A `FOREIGN_KEY` references a served table in the same schema, whose
+  `PRIMARY_KEY` or `UNIQUE` record has the referenced id and columns, with
+  `MATCH SIMPLE` and `NO_ACTION`/`RESTRICT` actions.
+- A `CHECK` is a single comparison between signed integer or floating-point
+  columns and numeric constants, such as `amount >= 0`. The grammar is checked
+  with DuckDB's own parser when `haybarn` or `duckdb` is installed.
+
+**Nullability is derived, not declared.** DuckDB requires every Arrow field
+(descending through ordinary structs, but not into extension or logical-type
+storage) to be non-nullable exactly when a `NOT_NULL` or `PRIMARY_KEY` record
+names it. The worker serves the schema that way: a non-nullable field no record
+names is served nullable, and a field a record names is served non-nullable.
+Arrow nullability in `columns` or a function's output schema is therefore never
+a constraint claim, and there is nothing to keep in sync by hand.
+`served_table_schema()` shows the schema a table is served with.
+
+**Negotiation.** Records are readable only under
+`vgi.constraint.records.read.v1` (plus `vgi.expression.duckdb.scalar.v1` for
+CHECK), and DuckDB writes a constrained table only under
+`vgi.constraint.enforce.v1`. `ReadOnlyCatalogInterface` advertises the read
+profiles when any declared table carries records, and the enforcement profile
+when such a table also has write functions: records on a writable table attest
+that the write functions enforce them. A catalog that serves tables some other
+way returns `catalog_capabilities_for(tables, indexes)` from its
+`catalog_capabilities()` override (`union` it with anything else it
+implements). If enforcement is not negotiated, the worker serves a constrained
+table read-only and logs a warning saying why.
+
+The legacy `not_null`, `unique`, `check`, `primary_key` and `foreign_key`
+fields type-check only as `()` and raise when set; `ForeignKeyDef` and
+`ForeignKeyInfo` are deprecated. Replace them with records.
+
+### Generated columns
+
+DuckDB reads only VIRTUAL generated columns of type Boolean, int8–int64 or
+string, with no default and no extension type, whose expression uses columns,
+Boolean/integer/string literals, `CAST` to those types, comparisons,
+`AND`/`OR`/`NOT`, `IS [NOT] NULL`, `COALESCE`, `+ - * %`, `||`, `abs`, `lower`
+and `upper`. A generated column must be nullable; one no record names is
+served nullable automatically, and naming it in a `NOT_NULL` or `PRIMARY_KEY`
+record is an error. Legacy `generated_expression` field metadata is translated
+to the canonical record when it is the field's only expression metadata.
+
 ### Validation
 
 Declarative catalogs include comprehensive validation:
@@ -627,10 +716,27 @@ Table(name="bad")  # ValueError: must specify either 'columns' or 'function'
 # Error: invalid constraint column
 Table(
     name="users",
-    columns=pa.schema([("id", pa.int64())]),
-    not_null=["nonexistent"],  # ValueError: column 'nonexistent' not found
+    columns=pa.schema([pa.field("id", pa.int64(), nullable=False)]),
+    constraints=(
+        constraint_record(
+            ["main"], "users", "nn", "NOT_NULL", [["nonexistent"]],
+            enforcement="ENFORCED", validation="ALL_ROWS",
+        ),  # ValueError: column path ['nonexistent'] is missing
+    ),
 )
 
+# Error: more than one PRIMARY_KEY record
+Table(
+    name="users",
+    columns=pa.schema([("id", pa.int64()), ("email", pa.string())]),
+    constraints=(
+        constraint_record(["main"], "users", "pk1", "PRIMARY_KEY", [["id"]],
+                          enforcement="ENFORCED", validation="ALL_ROWS"),
+        constraint_record(["main"], "users", "pk2", "PRIMARY_KEY", [["email"]],
+                          enforcement="ENFORCED", validation="ALL_ROWS"),
+    ),
+)  # ValueError: Table 'users': more than one PRIMARY_KEY
+
 # Error: default_schema not in schemas
 Catalog(
     name="myapp",
@@ -981,6 +1087,12 @@ The current CatalogInterface has the following limitations:
 - **Tags**: Cannot be updated after object creation
 - **Schema metadata**: Comments and tags cannot be updated on schemas
 - **Constraints**: Only NOT NULL can be added/dropped (no ALTER for UNIQUE/CHECK)
+- **Constrained CREATE TABLE**: `CREATE TABLE` with `NOT NULL`, `PRIMARY KEY`,
+  `UNIQUE`, `CHECK` or `FOREIGN KEY` reaches a catalog only when it advertises
+  `vgi.constraint.create.v1` and `vgi.constraint.enforce.v1`, i.e. when its
+  `table_create()` stores the records and its write paths enforce them. The SDK
+  ships no catalog that does (the in-memory test catalog stores no rows), so
+  such statements are refused with an error naming the missing profile.
 
 ### Table writes and indexes
 
@@ -1005,9 +1117,33 @@ on declarative tables. Table writes do not require the bundled
 [transactor](api/transactor.md); that is an optional database-access service with
 additional engine requirements.
 
-Index metadata can be declared with [`Index`][vgi.catalog.descriptors.Index]. Custom
-catalogs can implement `index_get()`, `index_create()`, and `index_drop()`; support
-depends on the catalog implementation and backing data source.
+Index metadata can be declared with [`Index`][vgi.catalog.descriptors.Index],
+building its definition with `index_definition()`:
+
+```python test="skip"
+from vgi.catalog import Index, index_definition
+
+Index(name="orders_customer", definition=index_definition(["main"], "orders", [["customer_id"]], algorithm="art"))
+```
+
+`ReadOnlyCatalogInterface` advertises `vgi.index.definition.read.v1` when any
+schema declares an index. Custom catalogs can implement `index_get()`,
+`index_create()`, and `index_drop()`; support depends on the catalog
+implementation and backing data source.
+
+### Overrides written for older signatures
+
+vgi.v3 added keyword arguments to several `CatalogInterface` methods
+(`catalog_attach(client_capabilities=)`, `schema_create(options=)`,
+`view_create(aliases=, security=, comment=, tags=, column_comments=)`,
+`table_create(constraint_definitions=, ...)`) and renamed `column_name` to
+`column_path`. The worker inspects an override's signature and passes only
+what it accepts, emitting a `DeprecationWarning` (and one log line) when it
+drops an argument. It drops one only when the value carries nothing the old
+override could have honoured (`None`, empty, a top-level column path); a
+barrier view, schema options or a nested column path fail with an error naming
+the parameter instead of being silently ignored. Update overrides to the
+current signatures, or accept `**kwargs`.
 
 ---
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/python-contracts.md

@@ -24,4 +24,4 @@ Methods take direct named arguments, with dataclasses for structured values such
 | [Notifications](reference/notify.md) | `vgi/reporting/notify.py` |
 | [Delegations](reference/delegations.md) | `vgi/reporting/delegations.py` |
 
-See [wire behavior](wire-contracts.md) for lifecycle, errors, encoding rules and worker policy. Regenerate these pages with `uv run python scripts/regen_generated.py --only vgi-python`.
+See [wire behavior](README.md#shared-conventions) for lifecycle, errors, encoding rules and worker policy. Regenerate these pages with `uv run python scripts/regen_generated.py --only vgi-python`.

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/alerts.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/alerts.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/common.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/common.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/delegations.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/delegations.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/discovery.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/discovery.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/notify.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/notify.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/objects.md

@@ -2,11 +2,11 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/objects.py`. Edit the Python module and regenerate this page.
 
-[Behavior specification](../objects.md) · [Shared records](common.md)
+[Shared records](common.md)
 
 Protocol `vgi.objects.v1`, version `1.0.0`.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/ownership.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/ownership.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/render.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/render.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/reports.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/reports.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/resource_identity.md

@@ -2,11 +2,11 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/resource_identity.py`. Edit the Python module and regenerate this page.
 
-[Behavior specification](../resource_identity.md) · [Shared records](common.md)
+[Shared records](common.md)
 
 Protocol `vgi.resource_identity.v1`, version `1.0.0`.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/resource_ownership.md

@@ -2,11 +2,11 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/resource_ownership.py`. Edit the Python module and regenerate this page.
 
-[Behavior specification](../objects.md) · [Shared records](common.md)
+[Shared records](common.md)
 
 Protocol `vgi.ownership.v1`, version `1.0.0`.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/schedules.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/schedules.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/schedules_v2.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/schedules_v2.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/sources.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/sources.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/docs/design/reporting-protocols/reference/sql_tasks.md

@@ -2,7 +2,7 @@
 
 <!-- Generated by vgi.codegen.reporting_docs; edit vgi/reporting/*.py. -->
 
-[All protocols](../python-contracts.md) · [Wire behavior](../wire-contracts.md)
+[All protocols](../python-contracts.md) · [Wire behavior](../README.md#shared-conventions)
 
 Source: `vgi/reporting/sql_tasks.py`. Edit the Python module and regenerate this page.
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/examples/cache_worker.py

@@ -75,7 +75,7 @@ class Rates(TableFunctionGenerator[RatesArgs, None]):
         description = "Exchange rates from a slow upstream, cacheable for 5 minutes"
 
     @classmethod
-    def process(cls, params: ProcessParams[RatesArgs, None], state: None, out: OutputCollector) -> None:
+    def process(cls, params: ProcessParams[RatesArgs], state: None, out: OutputCollector) -> None:
         """Answer the scan — or, if the client's copy is still current, say so."""
         vgi_out = cast(VgiOutputCollector, out)
 

{vgi_python-0.48.0 → vgi_python-0.48.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "vgi-python"
-version = "0.48.0"
+version = "0.48.2"
 description = "Vector Gateway Interface - Connect DuckDB to external programs via Apache Arrow"
 readme = "README.md"
 keywords = [
@@ -42,6 +42,10 @@ dependencies = [
     "platformdirs",
     "vgi-rpc>=0.52.3",
     "httpx2>=2.12.0",
+    # vgi.query_execution seals execution handles with vgi_rpc.crypto, which
+    # imports pycryptodome. vgi-rpc only pulls it in under its `http` extra, but
+    # catalog_attach reaches query_execution on every transport (stdio too).
+    "pycryptodome>=3.20",
 ]
 
 [project.optional-dependencies]

{vgi_python-0.48.0 → vgi_python-0.48.2}/scripts/check_duckdb2_manifest.py

@@ -79,7 +79,8 @@ def check_inventory(manifest: dict[str, Any], native_root: Path) -> list[str]:
     source_path = native_root / manifest["sources"]["audit_path"]
     if not source_path.is_file():
         return [f"required audit source is missing: {source_path}"]
-    source = source_path.read_bytes()
+    # Hash LF-normalized bytes so a CRLF checkout (Windows) matches the manifest.
+    source = source_path.read_bytes().replace(b"\r\n", b"\n")
     if hashlib.sha256(source).hexdigest() != manifest["sources"]["audit_sha256"]:
         errors.append("native audit digest changed; reconcile the manifest against the new evidence")
     rows = audit_rows(source.decode())

{vgi_python-0.48.0 → vgi_python-0.48.2}/tests/catalog/test_constraint_contract.py

@@ -322,7 +322,13 @@ def test_snapshot_rejects_duplicate_keys_labels_and_primary_keys() -> None:
         )
 
 
-@pytest.mark.parametrize("value", [b"", "not-bytes", b"x" * (contract.MAX_CONSTRAINT_TOKEN_BYTES + 1)])
+# Explicit ids: an oversized value as the test id overflows PYTEST_CURRENT_TEST,
+# and Windows limits an environment variable to 32,767 characters.
+@pytest.mark.parametrize(
+    "value",
+    [b"", "not-bytes", b"x" * (contract.MAX_CONSTRAINT_TOKEN_BYTES + 1)],
+    ids=["empty", "not-bytes", "over-limit"],
+)
 def test_revisions_and_owner_tokens_are_nonempty_bounded_bytes(value: bytes | str) -> None:
     for kwargs in [{"attach_opaque_data": value}, {"expected_revision": value}, {"transaction_opaque_data": value}]:
         with pytest.raises(ValueError):

{vgi_python-0.48.0 → vgi_python-0.48.2}/tests/catalog/test_constraint_example_cross_sdk.py

@@ -11,14 +11,19 @@ import pyarrow as pa
 import pytest
 
 from vgi.catalog import SchemaObjectType
-from vgi.catalog.constraints import ConstraintEnforcement, ConstraintValidation, validate_constraint_schema
+from vgi.catalog.constraints import (
+    ConstraintEnforcement,
+    ConstraintValidation,
+    served_table_schema,
+    validate_constraint_records,
+)
 from vgi.client import Client
 
 WORKERS: dict[str, list[str]] = json.loads(os.environ.get("VGI_METADATA_WORKERS", "{}"))
 
 
 @pytest.mark.parametrize("sdk", sorted(WORKERS))
-def test_example_constraints_are_explicit_stable_untrusted_records(sdk: str) -> None:
+def test_example_constraints_are_explicit_stable_trusted_records(sdk: str) -> None:
     with Client(server_path=WORKERS[sdk], worker_limit=1) as client:
         token = client.catalog_attach(
             name="example", data_version_spec=None, implementation_version=None
@@ -32,11 +37,11 @@ def test_example_constraints_are_explicit_stable_untrusted_records(sdk: str) ->
             assert fetched.constraints == listed.constraints
             infos = fetched.constraint_infos()
             assert infos, (sdk, name)
-            assert all(item.enforcement == ConstraintEnforcement.METADATA_ONLY for item in infos)
-            assert all(item.validation == ConstraintValidation.UNVALIDATED for item in infos)
-            assert not any(item.trusted for item in infos)
-            validate_constraint_schema(
-                pa.ipc.read_schema(pa.py_buffer(fetched.columns)),
-                [item.definition for item in infos],
-                trusted=[],
-            )
+            # DuckDB reads only ENFORCED + ALL_ROWS records; anything else fails ATTACH.
+            assert all(item.enforcement == ConstraintEnforcement.ENFORCED for item in infos)
+            assert all(item.validation == ConstraintValidation.ALL_ROWS for item in infos)
+            assert all(item.trusted for item in infos)
+            columns = pa.ipc.read_schema(pa.py_buffer(fetched.columns))
+            validate_constraint_records(name, columns, infos, schema_path=["data"])
+            # Served nullability already equals NOT_NULL/PRIMARY_KEY membership.
+            assert served_table_schema(columns, infos).equals(columns, check_metadata=True)

{vgi_python-0.48.0 → vgi_python-0.48.2}/tests/catalog/test_constraint_public_contract.py

@@ -170,14 +170,27 @@ def test_discovery_never_hides_constraints_and_mutation_requires_revision() -> N
         Worker._table_for_capabilities(table_info(definition_revision=None), PROFILES)
 
 
-@pytest.mark.parametrize("trusted", [False, True])
-def test_discovery_rejects_nullability_that_does_not_match_trust(trusted: bool) -> None:
-    info = INFO if trusted else replace(INFO, enforcement=ConstraintEnforcement.METADATA_ONLY)
-    wrong = pa.schema([pa.field("a.b", pa.int64(), nullable=trusted)])
-    with pytest.raises(ValueError, match="nullability"):
-        Worker._table_for_capabilities(
-            table_info(columns=wrong.serialize().to_pybytes(), constraints=[info.serialize_to_bytes()]), PROFILES
-        )
+def test_discovery_serves_record_nullability() -> None:
+    nullable = pa.schema([pa.field("a.b", pa.int64(), nullable=True)])
+    served = Worker._table_for_capabilities(
+        table_info(columns=nullable.serialize().to_pybytes(), constraints=[INFO.serialize_to_bytes()]), PROFILES
+    )
+    assert not pa.ipc.read_schema(pa.py_buffer(served.columns)).field("a.b").nullable
+
+
+@pytest.mark.parametrize(
+    "state",
+    [
+        {"enforcement": ConstraintEnforcement.METADATA_ONLY},
+        {"validation": ConstraintValidation.FUTURE_WRITES},
+        {"validation": ConstraintValidation.UNVALIDATED},
+    ],
+)
+def test_discovery_rejects_records_duckdb_will_not_read(state: dict[str, Any]) -> None:
+    """The extension fails ATTACH unless every record is ENFORCED + ALL_ROWS."""
+    info = replace(INFO, **state)
+    with pytest.raises(ValueError, match="accepts only ENFORCED \\+ ALL_ROWS"):
+        Worker._table_for_capabilities(table_info(constraints=[info.serialize_to_bytes()]), PROFILES)
 
 
 def test_create_checks_negotiation_and_current_support_before_provider(monkeypatch: pytest.MonkeyPatch) -> None: