Created attachment 1384856 [details] failing dot file Description of problem: The latest graphviz is segving (on i686 only) converting the attached file. The 2.40.1-16.fc28.i686 works while the 2.40.1-18.fc28.i686 build fails Version-Release number of selected component (if applicable): graphviz-2.40.1-18.fc28.i686 How reproducible: Every time. Steps to Reproduce: 1. dot multipolygon__collector_8hpp__incl.dot -Tpng -o multipolygon__collector_8hpp__incl.png Actual results: dot segvs with no output. Expected results: dot produces output.
same behaviour in ticket 1541436 build log: https://kojipkgs.fedoraproject.org//work/tasks/2707/24642707/build.log
Yes I believe this affects any package using doxygen to build documentation - that is certainly how I found it. The outcome is that doxygen fails to build the graphs but doesn't actually abort the documentation build so you just wind up with missing graphs, and the dot input files left in the documentation tree instead of being cleaned. After that if the doco is put in an arched rpm it will just be wrong and if (as is common) it is put in a noarch rpm then it won't match the other arches and the koji build will fail as a result.
We have graphviz-2.40.1 for more than year. Between 2.40.1-16 and 2.40.1-18 there was only rebuild for new dep (which seems unrelated), so it's interesting that it started segfaulting.
Yes I got roughly as far as figuring out that nothing really changed and going WTF a lot...
graphviz-2.40.1-18.fc28.i686 is compiled with gcc-7.2.1-7.fc28 -march=i686 -fstack-clash-protection , so maybe this graphiviz hit bug 1536555 (at least openssh, rpm got broken by this compination). Maybe rebuilding with latest gcc helps.
(In reply to Tom Hughes from comment #0) > Created attachment 1384856 [details] > failing dot file > This test case seems okay with rawhide i386 mock environ and graphviz-2.40.1-19.fc28.i686
Yes that seems to have fixed it.