Use testdir fixture in test_double_test to ensure controlled environment for execution

Because the test relies that two subsequent tests try to use pytest.warns to capture
the exact same warning, it is better to use testdir to ensure test execution
occurs in the order we expect (which might be different with pytest-xdist or
pytest-random for example)

testing/test_recwarn.py

@@ -168,8 +168,17 @@ class TestWarns(object):
         assert str(record[0].message) == "user"
         assert str(record[1].message) == "runtime"
 
-    @pytest.mark.parametrize('run', [1, 2])
-    def test_doubletest(self, run):
+    def test_double_test(self, testdir):
         """If a test is run again, the warning should still be raised"""
+        testdir.makepyfile('''
+            import pytest
+            import warnings
+
+            @pytest.mark.parametrize('run', [1, 2])
+            def test(run):
                 with pytest.warns(RuntimeWarning):
                     warnings.warn("runtime", RuntimeWarning)
+        ''')
+        result = testdir.runpytest()
+        result.stdout.fnmatch_lines(['*2 passed in*'])
+