Re: coreutils-6.5: spurious test failure (rm/one-filesystem)

[Jim Meyering]

Michael Deutschmann <address@hidden> wrote:
> I've noticed a failure on one of the "check-root" tests on my system:
>
>> make  check-TESTS
>> make[3]: Entering directory `/home/michael/build/cu-build/tests/rm'
>> mount: unknown filesystem type 'none'
>> ../../../coreutils-6.5/tests/rm/one-file-system: failure in testing framework
>> FAIL: one-file-system
>> ======================================
>> 1 of 1 tests failed
>
> It seems to be spurious -- investigating the test script shows that it is
> trying to use the "mount --bind" command, which is a feature not supported
> in Linux 2.0.40 (which I use).

Thanks for running the root-only tests and for the report.
I've changed that test to be skipped upon setup failure:

        * tests/rm/one-file-system: Upon setup failure (e.g., mount failure),
        skip the test rather than failing.  Reported by Michael Deutschmann.

diff --git a/tests/rm/one-file-system b/tests/rm/one-file-system
index bb7ffeb..66c04d5 100755
--- a/tests/rm/one-file-system
+++ b/tests/rm/one-file-system
@@ -51,8 +51,8 @@ rm: skipping `a/b', since it's on a diff
 EOF

 if test $framework_failure = 1; then
-  echo "$0: failure in testing framework" 1>&2
-  (exit 1); exit 1
+  echo "$0: setup failed; skipping this test" 1>&2
+  (exit 77); exit 77
 fi

 fail=0

> Anyways, using "mount" at all from a test script, even a root-only one,
> seems unwise, as it tends to vary across different unix-like systems.

If you find a way to exercise the same code cleanly and reliably,
but without using a bind mount, then I'd love a patch.
Then this test wouldn't have to be root-only.

FWIW, the main purpose of the root-only tests is for me to be
able to ensure/document that changes I've made really fix the
problems they purport to fix, and to make sure they stay fixed.
If other people can run those same root-only tests, that's good, too,
but not as important.