Force strict xfail on pytest, and update/remove tests accordingly

setup.cfg

@@ -61,6 +61,7 @@ allow_untyped_defs = True
 
 [tool:pytest]
 addopts = --ignore src/pip/_vendor --ignore tests/tests_cache -r aR --color=yes
+xfail_strict = True
 markers =
     network: tests that need network
     incompatible_with_sysconfig

tests/functional/test_install_extras.py

@@ -172,15 +172,22 @@ def test_install_requirements_no_r_flag(script):
 
 
 @pytest.mark.parametrize(
-    "extra_to_install, simple_version",
+    "extra_to_install, simple_version, fails_on_legacy",
     [
-        ["", "3.0"],
-        pytest.param("[extra1]", "2.0", marks=pytest.mark.xfail),
-        pytest.param("[extra2]", "1.0", marks=pytest.mark.xfail),
-        pytest.param("[extra1,extra2]", "1.0", marks=pytest.mark.xfail),
+        ("", "3.0", False),
+        ("[extra1]", "2.0", True),
+        ("[extra2]", "1.0", True),
+        ("[extra1,extra2]", "1.0", True),
     ],
 )
-def test_install_extra_merging(script, data, extra_to_install, simple_version):
+@pytest.mark.usefixtures("data")
+def test_install_extra_merging(
+    script,
+    resolver_variant,
+    extra_to_install,
+    simple_version,
+    fails_on_legacy,
+):
     # Check that extra specifications in the extras section are honoured.
     pkga_path = script.scratch_path / "pkga"
     pkga_path.mkdir()
@@ -200,6 +207,9 @@ def test_install_extra_merging(script, data, extra_to_install, simple_version):
 
     result = script.pip_install_local(
         f"{pkga_path}{extra_to_install}",
+        expect_error=(fails_on_legacy and resolver_variant == "legacy"),
     )
 
-    assert f"Successfully installed pkga-0.1 simple-{simple_version}" in result.stdout
+    if not fails_on_legacy or resolver_variant == "2020-resolver":
+        expected = f"Successfully installed pkga-0.1 simple-{simple_version}"
+        assert expected in result.stdout

tests/functional/test_install_reqs.py

@@ -213,7 +213,7 @@ def test_multiple_constraints_files(script, data):
     assert "installed Upper-1.0" in result.stdout
 
 
-@pytest.mark.xfail(reason="Unclear what this guarantee is for.")
+# FIXME: Unclear what this guarantee is for.
 def test_respect_order_in_requirements_file(script, data):
     script.scratch_path.joinpath("frameworks-req.txt").write_text(
         textwrap.dedent(

tests/functional/test_wheel.py

@@ -240,7 +240,6 @@ def test_pip_wheel_fail(script, data):
     assert result.returncode != 0
 
 
-@pytest.mark.xfail(reason="The --build option was removed")
 def test_no_clean_option_blocks_cleaning_after_wheel(
     script,
     data,

tests/unit/test_models_wheel.py

@@ -92,14 +92,6 @@ def test_not_supported_osx_version(self) -> None:
         w = Wheel("simple-0.1-cp27-none-macosx_10_9_intel.whl")
         assert not w.supported(tags=tags)
 
-    @pytest.mark.xfail(
-        reason=(
-            "packaging.tags changed behaviour in this area, and @pradyunsg "
-            "decided as the release manager that this behaviour change is less "
-            "critical than Big Sur support for pip 20.3. See "
-            "https://github.com/pypa/packaging/pull/361 for further discussion."
-        )
-    )
     def test_supported_multiarch_darwin(self) -> None:
         """
         Multi-arch wheels (intel) are supported on components (i386, x86_64)