Add --xfail-flaky to report exhausted flaky failures as xfail

README.rst

@@ -304,6 +304,20 @@ This is useful in large test suites to bound resource usage when many tests
 are flaky at the same time. The cap applies after rerun selection, including
 tests configured with ``--force-reruns`` and ``@pytest.mark.flaky``.
 
+Mark exhausted flaky failures as xfail
+--------------------------------------
+
+To keep flaky failures from failing the test run, pass ``--xfail-flaky``.
+A test that still fails after all its re-runs is then reported as
+``xfailed`` instead of ``failed``:
+
+.. code-block:: bash
+
+   $ pytest --reruns 2 --xfail-flaky
+
+Only tests that were actually re-run are affected. Failures excluded from
+re-running by ``--only-rerun``/``--rerun-except`` still report as failed.
+
 Show tracebacks for retried failures
 ------------------------------------
 

changes/92.feature.rst

@@ -0,0 +1,2 @@
+Add ``--xfail-flaky`` flag that reports a test as ``xfailed`` instead of
+``failed`` when it still fails after all its re-runs are exhausted.

src/pytest_rerunfailures.py

@@ -169,6 +169,14 @@ def pytest_addoption(parser):
         dest="fail_on_flaky",
         help="Fail the test run with exit code 7 if a flaky test passes on a rerun.",
     )
+    group._addoption(
+        "--xfail-flaky",
+        action="store_true",
+        dest="xfail_flaky",
+        help="Mark tests that still fail after exhausting their reruns as "
+        "xfailed instead of failed. Only applies to tests that were actually "
+        "rerun, and not to failures excluded by --only-rerun/--rerun-except.",
+    )
     group._addoption(
         "--rerun-show-tracebacks",
         action="store_true",
@@ -1285,6 +1293,19 @@ def pytest_runtest_protocol(item, nextitem):
                 # nonmatching failure as a final result first.
                 continue
             elif _should_not_rerun(item, report, reruns, condition):
+                if (
+                    item.session.config.option.xfail_flaky
+                    and report.failed
+                    and item.execution_count > reruns
+                    and item.execution_count > 1
+                    and not any(item._terminal_errors.values())
+                ):
+                    # reruns are exhausted: report the final failure as xfail
+                    report.outcome = "skipped"
+                    report.wasxfail = (
+                        f"test failed after {item.execution_count - 1} "
+                        "rerun(s), marked as xfail"
+                    )
                 # no rerun needed or one already triggered, log normally
                 item.ihook.pytest_runtest_logreport(report=report)
             else:

tests/test_pytest_rerunfailures.py

@@ -548,6 +548,68 @@ def test_pass():
     assert_outcomes(result, passed=1, rerun=2)
 
 
+def test_xfail_flaky_marks_exhausted_reruns_as_xfail(testdir):
+    testdir.makepyfile("def test_fail(): assert False")
+    result = testdir.runpytest("--reruns", "2", "--xfail-flaky")
+    assert result.ret == 0
+    assert_outcomes(result, passed=0, failed=0, xfailed=1, rerun=2)
+
+
+def test_xfail_flaky_does_not_affect_eventual_pass(testdir):
+    testdir.makepyfile(
+        f"""
+        def test_pass():
+            {temporary_failure()}"""
+    )
+    result = testdir.runpytest("--reruns", "2", "--xfail-flaky")
+    assert_outcomes(result, passed=1, rerun=1)
+
+
+def test_xfail_flaky_does_not_affect_test_without_reruns(testdir):
+    testdir.makepyfile("def test_fail(): assert False")
+    result = testdir.runpytest("--xfail-flaky")
+    assert_outcomes(result, passed=0, failed=1, xfailed=0, rerun=0)
+
+
+def test_xfail_flaky_ignores_unmatched_only_rerun_error(testdir):
+    testdir.makepyfile("def test_fail(): raise ValueError('nope')")
+    result = testdir.runpytest(
+        "--reruns", "2", "--only-rerun", "AssertionError", "--xfail-flaky"
+    )
+    assert_outcomes(result, passed=0, failed=1, xfailed=0, rerun=0)
+
+
+def test_xfail_flaky_with_flaky_marker(testdir):
+    testdir.makepyfile(
+        """
+        import pytest
+
+        @pytest.mark.flaky(reruns=1)
+        def test_fail(): assert False
+        """
+    )
+    result = testdir.runpytest("--xfail-flaky")
+    assert result.ret == 0
+    assert_outcomes(result, passed=0, failed=0, xfailed=1, rerun=1)
+
+
+def test_xfail_flaky_marks_exhausted_setup_failures_as_xfail(testdir):
+    testdir.makepyfile(
+        """
+        import pytest
+
+        @pytest.fixture
+        def broken_fixture():
+            raise ValueError("setup always fails")
+
+        def test_fail(broken_fixture): pass
+        """
+    )
+    result = testdir.runpytest("--reruns", "2", "--xfail-flaky")
+    assert result.ret == 0
+    assert_outcomes(result, passed=0, failed=0, xfailed=1, rerun=2)
+
+
 def test_no_extra_test_summary_for_reruns_by_default(testdir):
     testdir.makepyfile(
         f"""