fix from empty pandas (#2757)

py-polars/polars/internals/frame.py

@@ -58,7 +58,7 @@
     _DOCUMENTING = True
 
 from polars._html import NotebookFormatter
-from polars.datatypes import Boolean, DataType, UInt32, py_type_to_dtype
+from polars.datatypes import Boolean, DataType, UInt32, Utf8, py_type_to_dtype
 from polars.utils import (
     _prepare_row_count_args,
     _process_null_values,
@@ -360,12 +360,13 @@ def _from_pandas(
         """
         # path for table without rows that keeps datatype
         if data.shape[0] == 0:
-            # We do a loop and materialize the series.
-            # There can be series of len != null due to pandas indexes.
             series = []
             for name in data.columns:
-                col = pli.Series(name, data[name])
-                if len(col) == 0:
+                pd_series = data[name]
+                if pd_series.dtype == np.dtype("O"):
+                    series.append(pli.Series(name, [], dtype=Utf8))
+                else:
+                    col = pli.Series(name, pd_series)
                     series.append(pli.Series(name, col))
             return DataFrame(series)
 

py-polars/tests/test_interop.py

@@ -231,6 +231,14 @@ def test_from_empty_pandas() -> None:
     assert polars_df.dtypes == [pl.Float64, pl.Float64]
 
 
+def test_from_empty_pandas_strings() -> None:
+    df = pd.DataFrame(columns=["a", "b"])
+    df["a"] = df["a"].astype(str)
+    df["b"] = df["b"].astype(float)
+    df_pl = pl.from_pandas(df)
+    assert df_pl.dtypes == [pl.Utf8, pl.Float64]
+
+
 def test_from_empty_arrow() -> None:
     df = pl.from_arrow(pa.table(pd.DataFrame({"a": [], "b": []})))
     assert df.columns == ["a", "b"]  # type: ignore
@@ -240,5 +248,9 @@ def test_from_empty_arrow() -> None:
     df1 = pd.DataFrame(columns=["b"], dtype=float)
     tbl = pa.Table.from_pandas(df1)
     out = pl.from_arrow(tbl)
+    assert out.columns == ["b", "__index_level_0__"]  # type: ignore
+    assert out.dtypes == [pl.Float64, pl.Utf8]  # type: ignore
+    tbl = pa.Table.from_pandas(df1, preserve_index=False)
+    out = pl.from_arrow(tbl)
     assert out.columns == ["b"]  # type: ignore
     assert out.dtypes == [pl.Float64]  # type: ignore