On Sat, Apr 27, 2019 at 10:53:44PM +0200, Robert Scheck wrote:
Hello Ondrej, hello Maria,
On Fri, 26 Apr 2019, Ondrej Zajicek wrote:
The suite of basic unit tests can be run by 'make test'.
that one fails for BIRD 2.0.4 on RHEL/CentOS 7 on ppc64 (big endian) like this:
[...] obj/lib/fletcher16_test: t_fletcher16_compute [FAIL] obj/lib/fletcher16_test: t_fletcher16_checksum [FAIL] obj/lib/fletcher16_test [FAIL]
https://kojipkgs.fedoraproject.org//work/tasks/1532/34501532/build.log for build logs. https://koji.fedoraproject.org/koji/taskinfo?taskID=34501529 in case you would like to compare output of different architectures.
And: It also fails on Fedora 31 on s390x in the same way like for ppc64.
https://kojipkgs.fedoraproject.org//work/tasks/1613/34501613/build.log and https://koji.fedoraproject.org/koji/taskinfo?taskID=34501608 for details.
Just let me know if I can be helpful somehow in tracking them down, even my gdb knowledge is quite limited.
Hi Seems like you compile it with your CFLAGS. There are compiler options which we expect there to be: -fno-strict-aliasing -fno-strict-overflow Although it probably is not relevant to this test, BIRD should always be build with these. Also, could you try run the test manually: ./obj/lib/fletcher16_test -vv ? It is possible that it is just a bug in the test, we should add there some explicit test vectors instead of comparison with straightforward implementation. Also, if you could add these options (which we usually use): -Wno-parentheses -Wno-pointer-sign -Wno-missing-field-initializers It would eliminate plenty of irrelevant build warnings. -- Elen sila lumenn' omentielvo Ondrej 'Santiago' Zajicek (email: santiago@crfreenet.org) OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to blame it on a computer is even more so."