setuponly: remove printing out/err from capman

[yoavcaspi]

resolve #5906
Thanks for @blueyed @asottile for the great help.
Really appreciate it.

[blueyed]

LGTM.
Consider this on top:

diff --git i/testing/python/setup_only.py w/testing/python/setup_only.py
index 8e4e3ace8..ef165c153 100644
--- i/testing/python/setup_only.py
+++ w/testing/python/setup_only.py
@@ -276,7 +276,7 @@ def test_setup_show_with_KeyboardInterrupt_in_test(testdir):
         import pytest
         @pytest.fixture
         def arg():
-            assert True
+            pass
         def test_arg(arg):
             raise KeyboardInterrupt()
     """
@@ -303,7 +303,7 @@ def test_setup_show_with_KeyboardInterrupt_in_fixture(testdir):
         def arg():
             raise KeyboardInterrupt()
         def test_arg(arg):
-            assert True
+            pass
     """
     )
     result = testdir.runpytest("--setup-show", p, no_reraise_ctrlc=True)
@@ -311,3 +311,4 @@ def test_arg(arg):
     result.stdout.fnmatch_lines(
         ["*SETUP    F arg*", "*! KeyboardInterrupt !*", "*= no tests ran in *"]
     )
+    assert "TEARDOWN F arg" not in str(result.stdout)

Also please squash and force-push it then.

[blueyed]

(I'm not really sure we need two tests here, which adds overhead to the test suite, but I guess it is OK).

[blueyed]

Note: testing/python/setup_only_new.py::test_setup_show_with_KeyboardInterrupt_in_fixture does not fail on master, i.e. we do not really need it here to test the regression/issue.

[asottile]

looks good, I think we can delete test_setup_show_with_KeyboardInterrupt_in_fixture if it doesn't fail on master 👍

[blueyed]

@nicoddemus
Have you seen the previous comments??

[yoavcaspi]

@blueyed I think the fixture test does not add much overhead and can left to provide some guard refactoring of the setup_only internals.
what do you think? I don't mind delete the test.

[blueyed]

@yoavcaspi
Yeah, it's not much by itself, but all those little things add up - so that you cannot easily run the whole suite locally already. Therefore I think it is good to restrict tests to the minimum being required.
It was removed in #6013 already, thanks for asking.
Also thanks for the contribution, of course! :)

[yoavcaspi]

great :-)