release: v0.8.6

python/manifoldbt/_native.pyi

@@ -5,9 +5,17 @@ import pyarrow as pa
 
 
 class DataStore:
-    """Parquet data store with SQLite metadata."""
+    """Bar data store (Parquet by default, or Arrow IPC via ``arrow_dir``) with SQLite metadata."""
 
-    def __init__(self, data_root: str, metadata_db: str = "metadata/metadata.sqlite") -> None: ...
+    def __init__(
+        self,
+        data_root: str,
+        metadata_db: str = "metadata/metadata.sqlite",
+        dataset: str = "bars_1m",
+        mega: Optional[str] = None,
+        arrow_dir: Optional[str] = None,
+    ) -> None: ...
+    def dataset(self) -> str: ...
     def data_root(self) -> str: ...
     def metadata_db(self) -> str: ...
     def active_version(self, dataset: str) -> str: ...

python/manifoldbt/dataframe.py

@@ -98,8 +98,15 @@ def arrow_to_series(
     if backend == "polars":
         import polars as pl
 
-        if hasattr(array, "to_pylist"):
-            return pl.Series(name=name, values=array.to_pylist())
+        try:
+            import pyarrow as pa
+        except ImportError:
+            pa = None
+        # Zero-copy: hand the Arrow buffers straight to polars instead of boxing
+        # every value into a Python object via to_pylist() (copies the whole
+        # column). pl.from_arrow shares the underlying buffers.
+        if pa is not None and isinstance(array, (pa.Array, pa.ChunkedArray)):
+            return pl.from_arrow(array).rename(name)
         return pl.Series(name=name, values=list(array))
 
     return array

python/manifoldbt/diagnostics.py

@@ -13,6 +13,27 @@ _EMPTY_TS = np.array([], dtype="datetime64[ns]")
 _SAFETY_PRO_FEATURE = "Safety checks (lookahead, exposure)"
 
 
+def _prepare_for_diagnostics(config, strategy, store):
+    """Mirror ``run()``'s config/store preparation for the diagnostics path.
+
+    ``run()`` resolves the config and store before serializing
+    (``_cap_output_resolution`` -> ``_resolve_store`` -> ``_prepare_config``).
+    Diagnostics must do the same: in particular a dict ``universe`` has to be
+    resolved to a ``List[SymbolId]`` first, otherwise ``config.to_json()`` emits
+    a JSON map and the Rust loader rejects it ("invalid type: map, expected a
+    sequence"). Returns the prepared ``(config, store)``.
+    """
+    from manifoldbt import (
+        _cap_output_resolution,
+        _resolve_store,
+        _prepare_config,
+    )
+    config = _cap_output_resolution(config)
+    store = _resolve_store(config, store)
+    config = _prepare_config(config, strategy, store)
+    return config, store
+
+
 @dataclass
 class LookaheadReport:
     """Result of a single look-ahead bias test."""
@@ -289,6 +310,10 @@ def detect_lookahead(
         run_on_aligned as _run_on_aligned,
     )
 
+    # Resolve config/store exactly like run() (notably dict universe -> ids),
+    # otherwise config.to_json() emits a map the Rust loader rejects.
+    config, store = _prepare_for_diagnostics(config, strategy, store)
+
     period = config.time_range_end - config.time_range_start
 
     # Load data ONCE for the full range.
@@ -813,6 +838,10 @@ def check_exposure_stability(
         run_on_aligned as _run_on_aligned,
     )
 
+    # Resolve config/store exactly like run() (notably dict universe -> ids),
+    # otherwise config.to_json() emits a map the Rust loader rejects.
+    config, store = _prepare_for_diagnostics(config, strategy, store)
+
     period = config.time_range_end - config.time_range_start
 
     # Load data ONCE.