Description of problem: QMC2 build takes ages. On F-11 my machine needed roughly 3,5 minutes to do a mockbuild. On F-12, after 7.5 minutes the process was not even halfway through. RPM Fusion buildsys took 1286 minutes to compile this simple program: http://buildsys.rpmfusion.org/build-status/job.psp?uid=5392 Memory usage is within normal limits. Version-Release number of selected component (if applicable): 4.4.2-7 How reproducible: always Steps to Reproduce: 1. Try to rebuild the source RPM in mock for F-12 Actual results: Build process takes ages Expected results: Build process takes several minutes tops. Additional info: http://belegdol.fedorapeople.org/qmc2-0.2-0.16.b12.fc11.src.rpm
Testcase added to http://gcc.gnu.org/PR41371, Alex Oliva will hopefully have a look soon. Short term workaround is to compile with -fno-var-tracking-assignments.
This bug appears to have been reported against 'rawhide' during the Fedora 12 development cycle. Changing version to '12'. More information and reason for this action is here: http://fedoraproject.org/wiki/BugZappers/HouseKeeping
In case anyone is wondering, I can confirm that adding -fno-var-tracking-assignments to CXXFLAGS brings the compilation back to reasonable values - it took 8 minutes and 28 seconds this time.
I have found that -fvar-tracking-assignments breaks mysql on s390: it's not a matter of compilation time, it's that the resulting executable fails its regression tests unless I add -fno-var-tracking-assignments. Is this likely to be the same issue described here, or should I file a new bug? (Unfortunately I do not have a good test case to submit, because I haven't localized the bug at all.)
That's certainly unrelated, so please do file a new bug. gcc shouldn't generate different code between -fvar-tracking-assignments and -fno-var-tracking-assignments (not even between -g and -g0), any such case is considered a bug, so you could just strip -fvar-tracking-assignments and -fno-var-tracking-assignments generated object files and compare them, those that differ should be reported with preprocessed source and all gcc command line options.
Oh, scratch that ... further experimentation shows that the same test sometimes passes and sometimes doesn't with identical compilation options. So the use of -fno-var-tracking-assignments probably didn't have anything to do with it. Sorry for the noise.
Please try http://kojipkgs.fedoraproject.org/packages/gcc/4.4.2/24.fc12/
Sorry for the late reply, but this needinfo has completely escaped my attention. Anyway, the results are as follows. Building qmc2-0.2-0.17.b13.fc12.src.rpm on this machine takes: - 10 min. 3 s; when using gcc-4.4.2-20 and -fno-var-tracking-assignments - 30 min. 23 s; when using gcc-4.4.2-24 w/o -fno-var-tracking-assignments In both of these cases the test was done in mock, with ccache and root_cache plugins disabled. RPMs for mock were all available locally, nothing had to be downloaded. Real-time build.log monitoring has revealed that the compiler needed most of the time to compile qrc_qmc2.cpp.
gcc-4.4.2-4.fc12 errata has been released.