flag_mutex_test.go: ineffective error-string assertion masks wrong error message #2404

Found while reviewing #2386.

Bug in test harness (flag_mutex_test.go:107): the assertion checking the error string uses t.Logf instead of t.Errorf/require.Contains:

if !strings.Contains(err.Error(), test.errStr) {
    t.Logf("Invalid error string %v", err)
}

So a wrong error message never fails the test. This is currently masking a real discrepancy: the "set both flags" case passes --i 11 --ai 12 and expects "option i cannot be set along with option ai", but the actual error is "option i cannot be set along with option t". MutuallyExclusiveFlags.check uses f.Names()[0], which returns the primary name (t), not the alias used on the CLI (ai).

Fix: change the assertion to t.Errorf (or assert.Contains(t, err.Error(), test.errStr)) so wrong messages fail, and decide whether the message should reference the flag name as typed on the command line (alias ai) rather than the primary name.

Test coverage gap (related): the regression test added in #2386 (--i --q) only covers "first member of group 0 × second member of group 1". The mirror cases — second member of group 0 (--s --q), and the issue-reproduction layout where the conflicting flag is the second member of both groups — are untested. Recommend adding at least one such case to pin the cross-group fix.