On Wed, Mar 09, 2016 at 10:53:06PM -0500, James McCoy wrote: > On Fri, Mar 04, 2016 at 06:41:00PM +0100, Christian Hofstaedtler wrote: > > Your package FTBFS on the mips and mipsel buildds: > > > > https://buildd.debian.org/status/fetch.php?pkg=subversion&arch=mips&ver=1.9.3-2%2Bb1&stamp=1457047589 > > https://buildd.debian.org/status/fetch.php?pkg=subversion&arch=mipsel&ver=1.9.3-2%2Bb1&stamp=1457043167 > > > > Both somewhere in the testsuite. > > This worked fine previously[0] with the exact same source using gcc-5 > 5.3.1-6. > > [0]: > https://buildd.debian.org/status/fetch.php?pkg=subversion&arch=mips&ver=1.9.3-2&stamp=1452933856 > > The segfaults in the test programs have been seen with gcc-5 5.3.1-10 > and 5.3.1-11. Using sid's gcc-snapshot 20150610-1 works fine. > > Experimenting with the GCC optimizations, the tests succeed at -O0 and > -O1 but fail at -O2. Playing with the individual optimizations, using > "-O2 -fno-ipa-sra" causes the build to succeed and similarly "-O1 > -fipa-sra" causes it to fail.
Note that 5.3.1-9 backported PR#68273 which fixed a similar issue in gsoap. I wonder if that patch had the side effect of introducing the issue when compiling subversion's code. Is the follow-up patch[0] in that same bug needed? [0]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68273#c31 Cheers, -- James GPG Key: 4096R/331BA3DB 2011-12-05 James McCoy <james...@debian.org>
