Fwd port of PR 18658: Load pandas in import cache before binding

Author: evertlammerts

src/duckdb_py/pandas/bind.cpp

@@ -1,6 +1,7 @@
 #include "duckdb_python/pandas/pandas_bind.hpp"
 #include "duckdb_python/pandas/pandas_analyzer.hpp"
 #include "duckdb_python/pandas/column/pandas_numpy_column.hpp"
+#include "duckdb_python/pyconnection/pyconnection.hpp"
 
 namespace duckdb {
 
@@ -123,6 +124,14 @@ void Pandas::Bind(const ClientContext &context, py::handle df_p, vector<PandasCo
 		throw InvalidInputException("Need a DataFrame with at least one column");
 	}
 
+	auto &import_cache = *DuckDBPyConnection::ImportCache();
+	auto pandas = import_cache.pandas();
+	if (!pandas) {
+		throw InvalidInputException("'pandas' is required for this operation, but it wasn't installed");
+	}
+	(void)import_cache.pandas.NA();
+	(void)import_cache.pandas.NaT();
+
 	return_types.reserve(column_count);
 	names.reserve(column_count);
 	// loop over every column

tests/fast/pandas/test_import_cache.py

@@ -0,0 +1,28 @@
+from conftest import NumpyPandas, ArrowPandas
+import duckdb
+import pytest
+
+
+@pytest.mark.parametrize('pandas', [NumpyPandas(), ArrowPandas()])
+def test_import_cache_explicit_dtype(pandas):
+    df = pandas.DataFrame(
+        {
+            'id': [1, 2, 3],
+            'value': pandas.Series(['123.123', pandas.NaT, pandas.NA], dtype=pandas.StringDtype(storage='python')),
+        }
+    )
+    con = duckdb.connect()
+    result_df = con.query("select id, value from df").df()
+
+    assert result_df['value'][1] is None
+    assert result_df['value'][2] is None
+
+
+@pytest.mark.parametrize('pandas', [NumpyPandas(), ArrowPandas()])
+def test_import_cache_implicit_dtype(pandas):
+    df = pandas.DataFrame({'id': [1, 2, 3], 'value': pandas.Series(['123.123', pandas.NaT, pandas.NA])})
+    con = duckdb.connect()
+    result_df = con.query("select id, value from df").df()
+
+    assert result_df['value'][1] is None
+    assert result_df['value'][2] is None