On Fri, May 24, 2019 at 2:14 AM Todd Zullinger <tmz@xxxxxxxxx> wrote: > > I wrote: > > While running the 2.22.0-rc1 tests on Fedora, I hit a few > > new test failures since 2.21.0 -- but only on the s390x > > architecture. > > > > I haven't had time to dig into these the past few days, so I > > thought I would send what I do have in case the problem is > > obvious to someone else. I think all of the failing tests > > are due to `git diff` commands. > [...] > > I don't have direct access to these s390x builders. I may > > be able to arrange shell access (or even reproduce this with > > qemu's s390x emulation). > > I poked around a little with a qemu s390x instance and see > the same failures. > > One simple failure in t4015 is with: > > git diff -w >out && > test_must_be_empty out > > Using git-2.21.0 this succeeds, while git-2.22.0-rc1 fails > and produces: > > diff --git a/x b/x > index d99af23..22d9f73 100644 > --- a/x > +++ b/x > @@ -1,6 +1,6 @@ > -whitespace at beginning > -whitespace change > -whitespace in the middle > -whitespace at end > + whitespace at beginning > +whitespace change > +white space in the middle > +whitespace at end > unchanged line > -CR at end > +CR at end > > I am guessing it's no coincidence that this only fails on > s390x and it is the only big endian architecture in the > fedora build system. I see a problem with -w and wrong type casting. sizeof(int) and sizeof(long) on s390x are not the same, are they? -- Duy
