fix(CSVDocumentSplitter): report column positions when read_csv_kwargs sets a header

haystack/components/preprocessors/csv_document_splitter.py

@@ -148,6 +148,12 @@ def run(self, documents: list[Document]) -> dict[str, list[Document]]:
             # Sort split_dfs first by row index, then by column index
             split_dfs.sort(key=lambda dataframe: (dataframe.index[0], dataframe.columns[0]))
 
+            # Columns are only positional when ``header=None``. A caller passing
+            # ``read_csv_kwargs={"header": 0}`` gets the first row as string labels,
+            # and ``int(label)`` then failed for every sub-table. Map labels back to
+            # their position in the original frame instead.
+            column_positions = {label: position for position, label in enumerate(df.columns)}
+
             for split_id, split_df in enumerate(split_dfs):
                 split_documents.append(
                     Document(
@@ -156,7 +162,7 @@ def run(self, documents: list[Document]) -> dict[str, list[Document]]:
                             **deepcopy(document.meta),
                             "source_id": document.id,
                             "row_idx_start": int(split_df.index[0]),
-                            "col_idx_start": int(split_df.columns[0]),
+                            "col_idx_start": column_positions[split_df.columns[0]],
                             "split_id": split_id,
                         },
                     )

haystack/components/rankers/meta_field.py

@@ -264,10 +264,12 @@ def run(
                 "The parameter <meta_field> is currently set to '{meta_field}', but none of the provided "
                 "Documents with IDs {document_ids} have this meta key.\n"
                 "Set <meta_field> to the name of a field that is present within the provided Documents.\n"
-                "Returning the <top_k> of the original Documents since there are no values to rank.",
+                "Applying the configured <missing_meta> policy instead of ranking.",
                 meta_field=self.meta_field,
                 document_ids=",".join([doc.id for doc in deduplicated_documents]),
             )
+            if self.missing_meta == "drop":
+                return {"documents": []}
             return {"documents": deduplicated_documents[:top_k]}
 
         if len(docs_missing_meta_field) > 0:

test/components/preprocessors/test_csv_document_splitter_header_kwargs.py

@@ -0,0 +1,27 @@
+"""read_csv_kwargs that sets a header must not break CSV splitting."""
+
+from haystack import Document
+from haystack.components.preprocessors.csv_document_splitter import CSVDocumentSplitter
+
+CSV = "name,score\nAda,9\n\nBob,8"
+
+
+def test_default_header_none_keeps_positional_columns():
+    splitter = CSVDocumentSplitter(row_split_threshold=1, column_split_threshold=None)
+    result = splitter.run([Document(content=CSV)])
+    assert [
+        (d.meta["row_idx_start"], d.meta["col_idx_start"]) for d in result["documents"]
+    ] == [(0, 0), (3, 0)]
+
+
+def test_read_csv_kwargs_header_reports_column_positions():
+    """With a caller-supplied header the columns are labels, not positions."""
+    splitter = CSVDocumentSplitter(
+        row_split_threshold=1,
+        column_split_threshold=None,
+        read_csv_kwargs={"header": 0},
+    )
+    result = splitter.run([Document(content=CSV)])
+    assert [
+        (d.meta["row_idx_start"], d.meta["col_idx_start"]) for d in result["documents"]
+    ] == [(0, 0), (2, 0)]

test/components/rankers/test_metafield.py

@@ -338,3 +338,22 @@ def test_none_meta_value_is_handled_as_missing_meta(self, missing_meta, expected
         ]
         output = ranker.run(documents=docs_before)
         assert [doc.id for doc in output["documents"]] == expected_ids
+
+
+def test_missing_meta_drop_when_no_document_has_the_field():
+    """The missing_meta policy applies even when no document has the field.
+
+    The ranker returned every input document in that case, so
+    missing_meta="drop" silently stopped filtering: a batch with no rated
+    document kept the unrated ones.
+    """
+    docs = [Document(id="a", content="a"), Document(id="b", content="b")]
+
+    assert (
+        MetaFieldRanker(meta_field="rating", missing_meta="drop").run(docs)["documents"]
+        == []
+    )
+
+    for policy in ("bottom", "top"):
+        ranker = MetaFieldRanker(meta_field="rating", missing_meta=policy)
+        assert [d.id for d in ranker.run(docs)["documents"]] == ["a", "b"]