Replace top-level optional imports with lazy import helpers

Heavy optional dependencies (numpy, pandas, polars, pyarrow) were
imported at module load time, causing failures in WASM environments
where these packages may not be available. This adds a lazy import
utility module and converts all eager try/except import patterns to
use cached lazy accessors. Type maps in dtypes.py are also converted
from module-level dicts to lru_cached factory functions. The pandas
DataFrame isinstance check in connection.py is replaced with a
duck-type hasattr check to avoid importing pandas at module scope.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: kesmit13

singlestoredb/connection.py

@@ -25,12 +25,6 @@
 from urllib.parse import urlparse
 
 import sqlparams
-try:
-    from pandas import DataFrame
-except ImportError:
-    class DataFrame(object):  # type: ignore
-        def itertuples(self, *args: Any, **kwargs: Any) -> None:
-            pass
 
 from . import auth
 from . import exceptions
@@ -1175,7 +1169,7 @@ def _iquery(
             out = list(cur.fetchall())
             if not out:
                 return []
-            if isinstance(out, DataFrame):
+            if hasattr(out, 'to_dict') and callable(getattr(out, 'to_dict')):
                 out = out.to_dict(orient='records')
             elif isinstance(out[0], (tuple, list)):
                 if cur.description:

singlestoredb/converters.py

@@ -26,11 +26,7 @@
 except (AttributeError, ImportError):
     has_pygeos = False
 
-try:
-    import numpy
-    has_numpy = True
-except ImportError:
-    has_numpy = False
+from .utils._lazy_import import get_numpy
 
 try:
     import bson
@@ -563,8 +559,9 @@ def float32_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.float32)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.float32)
 
     return map(float, json_loads(x))
 
@@ -591,8 +588,9 @@ def float32_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.float32)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.float32)
 
     return struct.unpack(f'<{len(x)//4}f', x)
 
@@ -619,8 +617,9 @@ def float16_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.float16)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.float16)
 
     return map(float, json_loads(x))
 
@@ -647,8 +646,9 @@ def float16_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.float16)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.float16)
 
     return struct.unpack(f'<{len(x)//2}e', x)
 
@@ -675,8 +675,9 @@ def float64_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.float64)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.float64)
 
     return map(float, json_loads(x))
 
@@ -703,8 +704,9 @@ def float64_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.float64)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.float64)
 
     return struct.unpack(f'<{len(x)//8}d', x)
 
@@ -731,8 +733,9 @@ def int8_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.int8)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.int8)
 
     return map(int, json_loads(x))
 
@@ -759,8 +762,9 @@ def int8_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.int8)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.int8)
 
     return struct.unpack(f'<{len(x)}b', x)
 
@@ -787,8 +791,9 @@ def int16_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.int16)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.int16)
 
     return map(int, json_loads(x))
 
@@ -815,8 +820,9 @@ def int16_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.int16)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.int16)
 
     return struct.unpack(f'<{len(x)//2}h', x)
 
@@ -843,8 +849,9 @@ def int32_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.int32)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.int32)
 
     return map(int, json_loads(x))
 
@@ -871,8 +878,9 @@ def int32_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.int32)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.int32)
 
     return struct.unpack(f'<{len(x)//4}l', x)
 
@@ -899,8 +907,9 @@ def int64_vector_json_or_none(x: Optional[str]) -> Optional[Any]:
     if x is None:
         return None
 
-    if has_numpy:
-        return numpy.array(json_loads(x), dtype=numpy.int64)
+    np = get_numpy()
+    if np is not None:
+        return np.array(json_loads(x), dtype=np.int64)
 
     return map(int, json_loads(x))
 
@@ -928,8 +937,9 @@ def int64_vector_or_none(x: Optional[bytes]) -> Optional[Any]:
         return None
 
     # Bytes
-    if has_numpy:
-        return numpy.frombuffer(x, dtype=numpy.int64)
+    np = get_numpy()
+    if np is not None:
+        return np.frombuffer(x, dtype=np.int64)
 
     return struct.unpack(f'<{len(x)//8}l', x)
 

singlestoredb/functions/dtypes.py

@@ -11,10 +11,9 @@
 from ..converters import converters
 from ..mysql.converters import escape_item  # type: ignore
 from ..utils.dtypes import DEFAULT_VALUES  # noqa
-from ..utils.dtypes import NUMPY_TYPE_MAP  # noqa
-from ..utils.dtypes import PANDAS_TYPE_MAP  # noqa
-from ..utils.dtypes import POLARS_TYPE_MAP  # noqa
-from ..utils.dtypes import PYARROW_TYPE_MAP  # noqa
+from ..utils.dtypes import get_numpy_type_map  # noqa
+from ..utils.dtypes import get_polars_type_map  # noqa
+from ..utils.dtypes import get_pyarrow_type_map  # noqa
 
 
 DataType = Union[str, Callable[..., Any]]

singlestoredb/functions/ext/json.py

@@ -7,10 +7,9 @@
 from typing import TYPE_CHECKING
 
 from ..dtypes import DEFAULT_VALUES
-from ..dtypes import NUMPY_TYPE_MAP
-from ..dtypes import PANDAS_TYPE_MAP
-from ..dtypes import POLARS_TYPE_MAP
-from ..dtypes import PYARROW_TYPE_MAP
+from ..dtypes import get_numpy_type_map
+from ..dtypes import get_polars_type_map
+from ..dtypes import get_pyarrow_type_map
 from ..dtypes import PYTHON_CONVERTERS
 
 if TYPE_CHECKING:
@@ -140,7 +139,7 @@ def load_pandas(
             (
                 pd.Series(
                     data, index=index, name=spec[0],
-                    dtype=PANDAS_TYPE_MAP[spec[1]],
+                    dtype=get_numpy_type_map()[spec[1]],
                 ),
                 pd.Series(mask, index=index, dtype=np.longlong),
             )
@@ -172,7 +171,7 @@ def load_polars(
     return pl.Series(None, row_ids, dtype=pl.Int64), \
         [
             (
-                pl.Series(spec[0], data, dtype=POLARS_TYPE_MAP[spec[1]]),
+                pl.Series(spec[0], data, dtype=get_polars_type_map()[spec[1]]),
                 pl.Series(None, mask, dtype=pl.Boolean),
             )
             for (data, mask), spec in zip(cols, colspec)
@@ -203,7 +202,7 @@ def load_numpy(
     return np.asarray(row_ids, dtype=np.longlong), \
         [
             (
-                np.asarray(data, dtype=NUMPY_TYPE_MAP[spec[1]]),  # type: ignore
+                np.asarray(data, dtype=get_numpy_type_map()[spec[1]]),  # type: ignore
                 np.asarray(mask, dtype=np.bool_),  # type: ignore
             )
             for (data, mask), spec in zip(cols, colspec)
@@ -235,7 +234,7 @@ def load_arrow(
         [
             (
                 pa.array(
-                    data, type=PYARROW_TYPE_MAP[dtype],
+                    data, type=get_pyarrow_type_map()[dtype],
                     mask=pa.array(mask, type=pa.bool_()),
                 ),
                 pa.array(mask, type=pa.bool_()),

singlestoredb/functions/ext/rowdat_1.py

@@ -12,10 +12,9 @@
 from ...config import get_option
 from ...mysql.constants import FIELD_TYPE as ft
 from ..dtypes import DEFAULT_VALUES
-from ..dtypes import NUMPY_TYPE_MAP
-from ..dtypes import PANDAS_TYPE_MAP
-from ..dtypes import POLARS_TYPE_MAP
-from ..dtypes import PYARROW_TYPE_MAP
+from ..dtypes import get_numpy_type_map
+from ..dtypes import get_polars_type_map
+from ..dtypes import get_pyarrow_type_map
 
 if TYPE_CHECKING:
     try:
@@ -212,7 +211,7 @@ def _load_pandas(
     index = pd.Series(row_ids)
     return pd.Series(row_ids, dtype=np.int64), [
         (
-            pd.Series(data, index=index, name=name, dtype=PANDAS_TYPE_MAP[dtype]),
+            pd.Series(data, index=index, name=name, dtype=get_numpy_type_map()[dtype]),
             pd.Series(mask, index=index, dtype=np.bool_),
         )
         for (data, mask), (name, dtype) in zip(cols, colspec)
@@ -247,7 +246,7 @@ def _load_polars(
     return pl.Series(None, row_ids, dtype=pl.Int64), \
         [
             (
-                pl.Series(name=name, values=data, dtype=POLARS_TYPE_MAP[dtype]),
+                pl.Series(name=name, values=data, dtype=get_polars_type_map()[dtype]),
                 pl.Series(values=mask, dtype=pl.Boolean),
             )
             for (data, mask), (name, dtype) in zip(cols, colspec)
@@ -282,7 +281,7 @@ def _load_numpy(
     return np.asarray(row_ids, dtype=np.int64), \
         [
             (
-                np.asarray(data, dtype=NUMPY_TYPE_MAP[dtype]),  # type: ignore
+                np.asarray(data, dtype=get_numpy_type_map()[dtype]),  # type: ignore
                 np.asarray(mask, dtype=np.bool_),  # type: ignore
             )
             for (data, mask), (name, dtype) in zip(cols, colspec)
@@ -318,7 +317,7 @@ def _load_arrow(
         [
             (
                 pa.array(
-                    data, type=PYARROW_TYPE_MAP[dtype],
+                    data, type=get_pyarrow_type_map()[dtype],
                     mask=pa.array(mask, type=pa.bool_()),
                 ),
                 pa.array(mask, type=pa.bool_()),
@@ -565,7 +564,7 @@ def _load_pandas_accel(
     numpy_ids, numpy_cols = _singlestoredb_accel.load_rowdat_1_numpy(colspec, data)
     cols = [
         (
-            pd.Series(data, name=name, dtype=PANDAS_TYPE_MAP[dtype]),
+            pd.Series(data, name=name, dtype=get_numpy_type_map()[dtype]),
             pd.Series(mask, dtype=np.bool_),
         )
         for (name, dtype), (data, mask) in zip(colspec, numpy_cols)
@@ -610,7 +609,7 @@ def _load_polars_accel(
             pl.Series(
                 name=name, values=data.tolist()
                 if dtype in string_types or dtype in binary_types else data,
-                dtype=POLARS_TYPE_MAP[dtype],
+                dtype=get_polars_type_map()[dtype],
             ),
             pl.Series(values=mask, dtype=pl.Boolean),
         )
@@ -653,7 +652,7 @@ def _load_arrow_accel(
     numpy_ids, numpy_cols = _singlestoredb_accel.load_rowdat_1_numpy(colspec, data)
     cols = [
         (
-            pa.array(data, type=PYARROW_TYPE_MAP[dtype], mask=mask),
+            pa.array(data, type=get_pyarrow_type_map()[dtype], mask=mask),
             pa.array(mask, type=pa.bool_()),
         )
         for (data, mask), (name, dtype) in zip(numpy_cols, colspec)