> diff --git a/t/t4012-diff-binary.sh b/t/t4012-diff-binary.sh > index 0a8af76aab..6579c81216 100755 > --- a/t/t4012-diff-binary.sh > +++ b/t/t4012-diff-binary.sh > @@ -102,10 +102,8 @@ test_expect_success 'apply binary patch' ' > > test_expect_success 'diff --no-index with binary creation' ' > echo Q | q_to_nul >binary && > - (: hide error code from diff, which just indicates differences > - git diff --binary --no-index /dev/null binary >current || > - true > - ) && > + # hide error code from diff, which just indicates differences > + test_might_fail git diff --binary --no-index /dev/null binary >current && I am not sure why we need to be non-deterministic here, i.e. I would rather test for success or non-success error code and not just *any* error code. This code was introduced in 71b989e7dd1 (fix bogus "diff --git" header from "diff --no-index", 2008-10-05) whereas the test_must_fail was introduced in 74359821020 (tests: introduce test_must_fail, 2008-02-28). However this code was authored without Junios involvement (he was AFK 2008-09-23..2008-10-08), so maybe test_must_fail was not so popular back then? While I think this patch is a strict improvement for the test suite, I do wonder if we can tighten the exit code check here (maybe in a follow up series?). Thanks, Stefan
