Fixed #18687: Removed test_performance_scalability

Even after repeated adjustment of the constants, this test still fails
randomly. It has educated us to ignore messages from Jenkins, to a
point where we missed some actual failures. In addition, it accounts
for a non-negligible percentage of the run time of the test suite
just by itself. Since no one has proposed a convincing patch in months,
I'm going to remove the patch. We can't keep a randomly failing test
forever.

1 files changed, 0 insertions, 18 deletions

diff --git a/tests/regressiontests/utils/crypto.py b/tests/regressiontests/utils/crypto.py
index 4c6b722ca9..5cf2934ab0 100644
--- a/tests/regressiontests/utils/crypto.py
+++ b/tests/regressiontests/utils/crypto.py
@@ -144,21 +144,3 @@ class TestUtilsCryptoPBKDF2(unittest.TestCase):
             result = pbkdf2(**vector['args'])
             self.assertEqual(binascii.hexlify(result).decode('ascii'),
                              vector['result'])
-
-    def test_performance_scalability(self):
-        """
-        Theory: If you run with 100 iterations, it should take 100
-        times as long as running with 1 iteration.
-        """
-        # These values are chosen as a reasonable tradeoff between time
-        # to run the test suite and false positives caused by imprecise
-        # measurement.
-        n1, n2 = 200000, 800000
-        elapsed = lambda f: timeit.Timer(f,
-                    'from django.utils.crypto import pbkdf2').timeit(number=1)
-        t1 = elapsed('pbkdf2("password", "salt", iterations=%d)' % n1)
-        t2 = elapsed('pbkdf2("password", "salt", iterations=%d)' % n2)
-        measured_scale_exponent = math.log(t2 / t1, n2 / n1)
-        # This should be less than 1. We allow up to 1.2 so that tests don't
-        # fail nondeterministically too often.
-        self.assertLess(measured_scale_exponent, 1.2)