fix: handle column names with null bytes (#365)

Author: PrettyWood

python/tests/fixtures/null-bytes-in-columns-names.xls

@@ -588,3 +588,10 @@ def test_header_row_and_skip_rows(
         .to_polars(),
         pl.DataFrame(expected),
     )
+
+
+def test_null_bytes_in_column_names() -> None:
+    """https://github.com/ToucanToco/fastexcel/issues/343"""
+    reader = fastexcel.read_excel(path_for_fixture("null-bytes-in-columns-names.xls"))
+    df = reader.load_sheet(0).to_polars()
+    assert df.shape == (8_763, 11)

python/tests/test_fastexcel.py

@@ -380,7 +380,19 @@ fn column_info_from_header<D: CalamineDataProvider>(
             .map(|col_idx| {
                 data.get_as_string((*row_idx, col_idx))
                     .map(|col_name| {
-                        ColumnInfoNoDtype::new(col_name, col_idx, ColumnNameFrom::LookedUp)
+                        // Remove null bytes from column names to avoid CString panics in Arrow FFI.
+                        //
+                        // Excel strings (especially UTF-16 in .xls) may contain embedded nulls (`\0`) after
+                        // conversion to Rust `String`. Arrow’s C FFI uses `CString::new()`, which fails on
+                        // null bytes, causing panics.
+                        //
+                        // This strips nulls while keeping the readable content.
+                        let sanitized_col_name = col_name.replace('\0', "");
+                        ColumnInfoNoDtype::new(
+                            sanitized_col_name,
+                            col_idx,
+                            ColumnNameFrom::LookedUp,
+                        )
                     })
                     .unwrap_or_else(|| {
                         ColumnInfoNoDtype::new(