Hi Chris, Sorry for the delay in getting back to you.
I've disabled to the verify_* tests for the time being; the way that rpmbuild works is that build/install happens to a build-user-owned DESTDIR so files can be gathered for packaging. Simply installing the previous version doesn't help because the build user doesn't have access to /etc/trafficserver (due to default permissions). The koji build system sets up BuildRequires based on the spec file, but I don't think there's a way I can ask it to do something as root (like changing the directory permissions). I'll investigate further. I'll return to the linking error later, but right now the problem I've run into is an assertion failure in libswoc; see the end of the build log here: https://kojipkgs.fedoraproject.org//work/tasks/48/126810048/build.log 5/63 Test #1: test_libswoc ...........................Subprocess aborted***Exception: 0.15 sec /usr/include/c++/14/optional:482: constexpr const _Tp& std::_Optional_base_impl<_Tp, _Dp>::_M_get() const [with _Tp = swoc::_1_5_12::Errata::Severity; _Dp = std::_Optional_base<swoc::_1_5_12::Errata::Severity, true, true>]: Assertion 'this->_M_is_engaged()' failed. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ test_libswoc is a Catch v2.13.8 host application. Run with -? for options ------------------------------------------------------------------------------- Errata copy ------------------------------------------------------------------------------- /builddir/build/BUILD/trafficserver-10.0.2-build/trafficserver-10.0.2/lib/swoc/unit_tests/test_Errata.cc:84 ............................................................................... /builddir/build/BUILD/trafficserver-10.0.2-build/trafficserver-10.0.2/lib/swoc/unit_tests/test_Errata.cc:84: FAILED: {Unknown expression after the reported line} due to a fatal error condition: SIGABRT - Abort (abnormal termination) signal I'll do a local build in a mock environment and see if I can attach a gdb to this process, but any pointers on what might be going on? I haven't kept up with C++17... Thanks, --Jered ----- On Nov 17, 2024, at 11:25 AM, Chris McFarlen <ch...@mcfarlen.us> wrote: > Hi Jered, > You are correct there is a chicken-egg issue with the tests that verify plugin > linkage (verify_.* tests). They use traffic_server's verify plugin command and > the traffic_server initialization code wants the install directories to exist. > I plan on fixing this before the next release. If installing before testing > isn't an option for your build, a workaround would be to comment or remove the > bodies of the verify_remap_plugin and verify_global_plugin in > cmake/add_atsplugin.cmake. Just leave the function definitions in place. > I might be able to help with the linking issues as well if you can provide > some > steps to reproduce. > Thanks! > Chris > Sent with Proton Mail secure email. > On Saturday, November 16th, 2024 at 5:30 PM, Jered Floyd <je...@convivian.com> > wrote: >> [This is specifically regarding trafficserver 10.0.2.] >> I've managed to get a successful build by disabling link time optimization >> features (default Fedora build flags "-flto=auto -ffat-lto-objects") -- that >> works in 9.2.6 so there's something else at play, but I'll look into that >> later. >> Now I'm running into ctest problems, which happen even if I build in a >> vanilla >> environment following the README.md build instructions. In particular, ctest >> doesn't use config files nor run or logdirs from the build environment. It >> appears that running "cd build; ctest" will not pass unless there is an >> existing install of traffic server installed. Have I missed something here? >> Thanks, >> --Jered >> ----- On Nov 14, 2024, at 11:56 AM, Jered Floyd je...@convivian.com wrote: >> > Hello! >> > I am trying to get ATS 10.x ready for Fedora rawhide/f42 and EPEL 10, and >> > am >> > running into some problems with CMake execution. I've banged my head >> > against it >> > for a few days so now I'm reaching out in hopes of finding an ATS CMake >> > expert. >> > What seems to be happening is that the build nearly completes, and then >> > linking >> > some of the last tests and application components fails due to missing >> > symbols. >> > This seems to be at least partly due to the library ordering, as a quick >> > hack >> > of repeating the library list (manually) gets a bit further through the >> > build >> > process. I'm not very familiar with CMake, so I'm not sure where to go from >> > here. >> > You can find an entire build log here, where I've disabled build >> > parallelism for >> > clarity. Everything blows up with linker errors about 50% through the >> > file; I >> > recommend searching on "[100%] Linking CXX executable test_PluginFactory" >> > as >> > the subsequent link 2 lines later fails: >> > [ https://kojipkgs.fedoraproject.org//work/tasks/5517/125855517/build.log | >> > https://kojipkgs.fedoraproject.org//work/tasks/5517/125855517/build.log ] >> > Any debugging help would be greatly appreciated! >> > Regards, > > > --Jered
