vgi-python 0.36.0__tar.gz → 0.36.1__tar.gz

{vgi_python-0.36.0 → vgi_python-0.36.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: vgi-python
-Version: 0.36.0
+Version: 0.36.1
 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

{vgi_python-0.36.0 → vgi_python-0.36.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "vgi-python"
-version = "0.36.0"
+version = "0.36.1"
 description = "Vector Gateway Interface - Connect DuckDB to external programs via Apache Arrow"
 readme = "README.md"
 keywords = [

{vgi_python-0.36.0 → vgi_python-0.36.1}/uv.lock

@@ -2580,7 +2580,7 @@ requires-dist = [
 
 [[package]]
 name = "vgi-python"
-version = "0.36.0"
+version = "0.36.1"
 source = { editable = "." }
 dependencies = [
     { name = "click" },

{vgi_python-0.36.0 → vgi_python-0.36.1}/vgi/_test_fixtures/table/splits.py

@@ -18,8 +18,8 @@ ways it goes right:
   shape that silently truncates a scan if a reader treats an empty split as EOS.
 * ``split_skewed(n)`` — one split ~100x the others, so greedy claiming can be
   told apart from static assignment.
-* ``split_many(n, splits := 1000)`` — far more splits than reader threads, which
-  forces sequential re-init on a reused connection.
+* ``split_many(n, splits)`` — far more splits than reader threads (pass them), which
+  forces sequential re-init on a reused connection. ``splits <= 0`` means 1000.
 * ``split_echo_filters(n)`` — reports the pushdown it saw *per split*, so a
   pushdown regression is visible without inferring it from row counts.
 """
@@ -322,7 +322,9 @@ class SplitManyFunction(_SplitBase):
 
     @classmethod
     def _split_count(cls, params: BindParams[SplitSequenceArgs]) -> int:
-        return max(params.args.splits, 1000)
+        # Honour the caller's count, as every other SDK's twin does; a floor here
+        # made a test that asks for 64 splits pay for 1000 HTTP round trips.
+        return params.args.splits if params.args.splits > 0 else 1000
 
 
 @dataclass(frozen=True)

{vgi_python-0.36.0 → vgi_python-0.36.1}/vgi/aggregate_function.py

@@ -518,7 +518,14 @@ class AggregateFunction[TState: StreamStateCodec](vgi.function.Function):
         """Compute the aggregate value for one output row.
 
         Args:
-            rid: Partition-local row index being filled.
+            rid: Identifies the output row being filled; do not use it to
+                index the partition. Over the per-row ``aggregate_window``
+                RPC it is DuckDB's own ``rid``: the row's index within the
+                current output chunk (``0 <= rid < 2048``), not its row in
+                the partition. When the default :meth:`window_batch` calls
+                this, it passes the partition row index instead. Either way
+                ``subframes`` are partition-relative: use them to find the
+                rows to aggregate.
             subframes: Frame ranges ``[(begin, end), ...]`` — 1 for the default
                 frame, 3 when ``EXCLUDE`` produces multiple subframes.
             partition: The cached partition data.

{vgi_python-0.36.0 → vgi_python-0.36.1}/vgi/client/aggregate.py

@@ -371,7 +371,10 @@ class AggregateSession:
 
         Args:
             partition_id: The partition previously shipped by :meth:`window_init`.
-            rid: Row index within the partition of the output row being computed.
+            rid: Identifies the output row. The DuckDB extension sends DuckDB's
+                own per-row ``rid``, the row's index within the current output
+                chunk rather than its row in the partition; ``frames`` carry the
+                partition-relative bounds, and a worker must use those.
             frames: The row's subframes as ``(begin, end)`` half-open offsets
                 into the partition. One entry normally; two or three for
                 ``EXCLUDE TIES`` / ``EXCLUDE GROUP``.

{vgi_python-0.36.0 → vgi_python-0.36.1}/vgi/protocol.py

@@ -2887,7 +2887,10 @@ class AggregateWindowRequest(ArrowSerializableDataclass):
         execution_id: Identifier for one query execution, stable across the
             coordinator and any secondary workers; keys worker-owned state.
         partition_id: Identifier of the window partition this call addresses.
-        rid: Row index within the partition for the output row being computed.
+        rid: Which output row this call fills, as DuckDB's per-row window callback
+            reports it: the row's index within the current output chunk
+            (``0 <= rid < STANDARD_VECTOR_SIZE``), NOT its row in the partition.
+            The frames below are partition-relative; use them to find the rows.
         frame_starts: Subframe start offsets (inclusive) within the partition, one
             per subframe.
         frame_ends: Subframe end offsets (exclusive) within the partition, parallel