BUG: Handle read_csv corner case (#21176)

Closes gh-21141

Author: r00ta

doc/source/whatsnew/v0.23.2.txt

@@ -64,6 +64,7 @@ Bug Fixes
 
 **I/O**
 
+- Bug in :func:`read_csv` that caused it to incorrectly raise an error when ``nrows=0``, ``low_memory=True``, and ``index_col`` was not ``None`` (:issue:`21141`)
 -
 -
 

pandas/io/parsers.py

@@ -3209,12 +3209,22 @@ def _get_empty_meta(columns, index_col, index_names, dtype=None):
             col = columns[k] if is_integer(k) else k
             dtype[col] = v
 
-    if index_col is None or index_col is False:
+    # Even though we have no data, the "index" of the empty DataFrame
+    # could for example still be an empty MultiIndex. Thus, we need to
+    # check whether we have any index columns specified, via either:
+    #
+    # 1) index_col (column indices)
+    # 2) index_names (column names)
+    #
+    # Both must be non-null to ensure a successful construction. Otherwise,
+    # we have to create a generic emtpy Index.
+    if (index_col is None or index_col is False) or index_names is None:
         index = Index([])
     else:
         data = [Series([], dtype=dtype[name]) for name in index_names]
         index = _ensure_index_from_sequences(data, names=index_names)
         index_col.sort()
+
         for i, n in enumerate(index_col):
             columns.pop(n - i)
 

pandas/tests/io/parser/common.py

@@ -238,6 +238,21 @@ def test_csv_mixed_type(self):
         out = self.read_csv(StringIO(data))
         tm.assert_frame_equal(out, expected)
 
+    def test_read_csv_low_memory_no_rows_with_index(self):
+        if self.engine == "c" and not self.low_memory:
+            pytest.skip("This is a low-memory specific test")
+
+        # see gh-21141
+        data = """A,B,C
+1,1,1,2
+2,2,3,4
+3,3,4,5
+"""
+        out = self.read_csv(StringIO(data), low_memory=True,
+                            index_col=0, nrows=0)
+        expected = DataFrame(columns=["A", "B", "C"])
+        tm.assert_frame_equal(out, expected)
+
     def test_read_csv_dataframe(self):
         df = self.read_csv(self.csv1, index_col=0, parse_dates=True)
         df2 = self.read_table(self.csv1, sep=',', index_col=0,