Error on shared column names when writing (#1335)

* Error on shared column names when writing

* add test

* Release note

anndata/_io/specs/methods.py

@@ -663,10 +663,23 @@ def write_dataframe(f, key, df, _writer, dataset_kwargs=MappingProxyType({})):
         if reserved in df.columns:
             raise ValueError(f"{reserved!r} is a reserved name for dataframe columns.")
     group = f.require_group(key)
+    if not df.columns.is_unique:
+        duplicates = list(df.columns[df.columns.duplicated()])
+        raise ValueError(
+            f"Found repeated column names: {duplicates}. Column names must be unique."
+        )
     col_names = [check_key(c) for c in df.columns]
     group.attrs["column-order"] = col_names
 
     if df.index.name is not None:
+        if df.index.name in col_names and not pd.Series(
+            df.index, index=df.index
+        ).equals(df[df.index.name]):
+            raise ValueError(
+                f"DataFrame.index.name ({df.index.name!r}) is also used by a column "
+                "whose values are different. This is not supported. Please make sure "
+                "the values are the same, or use a different name."
+            )
         index_name = df.index.name
     else:
         index_name = "_index"

anndata/tests/test_io_elementwise.py

@@ -300,3 +300,32 @@ def test_read_zarr_from_group(tmp_path, consolidated):
     with read_func(pth) as z:
         expected = ad.read_zarr(z["table/table"])
     assert_equal(adata, expected)
+
+
+def test_dataframe_column_uniqueness(store):
+    repeated_cols = pd.DataFrame(np.ones((3, 2)), columns=["a", "a"])
+
+    with pytest_8_raises(
+        ValueError,
+        match=r"Found repeated column names: \['a'\]\. Column names must be unique\.",
+    ):
+        write_elem(store, "repeated_cols", repeated_cols)
+
+    index_shares_col_name = pd.DataFrame(
+        {"col_name": [1, 2, 3]}, index=pd.Index([1, 3, 2], name="col_name")
+    )
+
+    with pytest_8_raises(
+        ValueError,
+        match=r"DataFrame\.index\.name \('col_name'\) is also used by a column whose values are different\.",
+    ):
+        write_elem(store, "index_shares_col_name", index_shares_col_name)
+
+    index_shared_okay = pd.DataFrame(
+        {"col_name": [1, 2, 3]}, index=pd.Index([1, 2, 3], name="col_name")
+    )
+
+    write_elem(store, "index_shared_okay", index_shared_okay)
+    result = read_elem(store["index_shared_okay"])
+
+    assert_equal(result, index_shared_okay)

docs/release-notes/0.10.6.md

@@ -4,6 +4,7 @@
 ```
 
 * Defer import of zarr in test helpers, as scanpy CI job relies on them {pr}`1343` {user}`ilan-gold`
+* Writing a dataframe with non-unique column names now throws an error, instead of silently overwriting {pr}`1335` {user}`ivirshup`
 
 ```{rubric} Documentation
 ```