BUG: Throw a ParserError when header rows have unequal column counts …

doc/source/whatsnew/v1.4.0.rst

@@ -377,7 +377,7 @@ I/O
 - Bug in :func:`read_fwf`, where difference in lengths of ``colspecs`` and ``names`` was not raising ``ValueError`` (:issue:`40830`)
 - Bug in :func:`Series.to_json` and :func:`DataFrame.to_json` where some attributes were skipped when serialising plain Python objects to JSON (:issue:`42768`, :issue:`33043`)
 - Column headers are dropped when constructing a :class:`DataFrame` from a sqlalchemy's ``Row`` object (:issue:`40682`)
--
+- Bug in :func:`read_csv` where reading multi-header input with unequal lengths incorrectly raising uncontrolled ``IndexError`` (:issue:`43102`)
 
 Period
 ^^^^^^

pandas/io/parsers/base_parser.py

@@ -343,6 +343,10 @@ def _extract_multi_indexer_columns(
         # extract the columns
         field_count = len(header[0])
 
+        # check if header lengths are equal
+        if not all(len(header_iter) == field_count for header_iter in header[1:]):
+            raise ParserError("Header rows must have an equal number of columns.")
+
         def extract(r):
             return tuple(r[i] for i in range(field_count) if i not in sic)
 

pandas/tests/io/parser/test_header.py

@@ -604,3 +604,18 @@ def test_read_csv_multiindex_columns(all_parsers):
     tm.assert_frame_equal(df1, expected.iloc[:1])
     df2 = parser.read_csv(StringIO(s2), header=[0, 1])
     tm.assert_frame_equal(df2, expected)
+
+
+def test_read_csv_multi_header_length_check(all_parsers):
+    # GH#43102
+    parser = all_parsers
+
+    case = """row11,row12,row13
+row21,row22, row23
+row31,row32
+"""
+
+    with pytest.raises(
+        ParserError, match="Header rows must have an equal number of columns."
+    ):
+        parser.read_csv(StringIO(case), header=[0, 2])