GH-100805: Support numpy.array() in random.choice(). (GH-100830)

(cherry picked from commit 9a68ff1)

Co-authored-by: Raymond Hettinger <[email protected]>

Author: miss-islington

Lib/random.py

@@ -366,7 +366,10 @@ def randint(self, a, b):
 
     def choice(self, seq):
         """Choose a random element from a non-empty sequence."""
-        if not seq:
+
+        # As an accommodation for NumPy, we don't use "if not seq"
+        # because bool(numpy.array()) raises a ValueError.
+        if not len(seq):
             raise IndexError('Cannot choose from an empty sequence')
         return seq[self._randbelow(len(seq))]
 

Lib/test/test_random.py

@@ -111,6 +111,21 @@ def test_choice(self):
         self.assertEqual(choice([50]), 50)
         self.assertIn(choice([25, 75]), [25, 75])
 
+    def test_choice_with_numpy(self):
+        # Accommodation for NumPy arrays which have disabled __bool__().
+        # See: https://github.com/python/cpython/issues/100805
+        choice = self.gen.choice
+
+        class NA(list):
+            "Simulate numpy.array() behavior"
+            def __bool__(self):
+                raise RuntimeError
+
+        with self.assertRaises(IndexError):
+            choice(NA([]))
+        self.assertEqual(choice(NA([50])), 50)
+        self.assertIn(choice(NA([25, 75])), [25, 75])
+
     def test_sample(self):
         # For the entire allowable range of 0 <= k <= N, validate that
         # the sample is of the correct length and contains only unique items

Misc/NEWS.d/next/Library/2023-01-07-15-13-47.gh-issue-100805.05rBz9.rst

@@ -0,0 +1,2 @@
+Modify :func:`random.choice` implementation to once again work with NumPy
+arrays.