34297 sub slow in first call

pandas/core/indexes/multi.py

@@ -3218,6 +3218,8 @@ def equals(self, other) -> bool:
                 # other cannot contain tuples, so cannot match self
                 return False
 
+            if len(self) != len(other):
+                return False
             return array_equivalent(self._values, other._values)
 
         if self.nlevels != other.nlevels:

pandas/tests/frame/test_arithmetic.py

@@ -1484,3 +1484,26 @@ def test_pow_nan_with_zero():
 
     result = left["A"] ** right["A"]
     tm.assert_series_equal(result, expected["A"])
+
+
+def test_performance_subsequent_sub_calls():
+    date_range = pd.date_range("20200101 00:00", "20200102 0:00", freq="S")
+    level_0_names = [str(i) for i in range(30)]
+
+    index = pd.MultiIndex.from_product([level_0_names, date_range])
+    column_names = ["col_1", "col_2"]
+
+    df = pd.DataFrame(np.random.rand(len(index), 2), index=index, columns=column_names)
+
+    sub_df = pd.DataFrame(
+        np.random.randint(1, 10, (len(level_0_names), 2)),
+        index=level_0_names,
+        columns=column_names,
+    )
+
+    start = datetime.now()
+    df.sub(sub_df, level=0)
+    middle = datetime.now()
+    df.sub(sub_df, level=0)
+    end = datetime.now()
+    assert (middle - start).microseconds < (end - middle).microseconds * 2