Interesting, could this be related to a discussion we had on IRC today? => IRC: 12:40 <LRN> MHD tests randomly fail on the buildslave 12:40 <LRN> i suspect that this is due to the testsuite running in parallel 12:41 <LRN> (W32 slave has 4 virtual cores, whereas non-W32 slaves seem to only have 1 virtual core) 12:42 <LRN> the fix is to either use -j1 in the master config, or explicitly disable parallel testing in makefiles somehow 16:01 <grothoffhome> I don't know how to disable parallel testing, which does sound like the better idea. Do you? 16:14 <LRN> grothoff, you can enable the serial test harness by using automake option "serial-tests" 16:14 <LRN> http://www.gnu.org/software/automake/manual/html_node/Serial-Test-Harness.html#Serial-Test-Harness
16:16 <grothoffhome> LRN: then please just put that in. => SVN: r30481 | LRN | 2013-10-30 16:34:40 +0100 (Wed, 30 Oct 2013) | 2 lines Force the use of non-parallel testsuite => IRC: 17:33 <LRN> configure.ac:25: option `serial-tests' not recognized :( 17:33 <LRN> half of the slaves seem to be using old autoconf or something 17:34 <LRN> https://lists.gnu.org/archive/html/automake/2013-01/msg00060.html So this also depends on the specific automake version (older ones didn't do tests in parallel?). Anyway, please let me know if this helps... Happy hacking! Christian On 10/30/2013 08:45 PM, Andreas Enge wrote: > Hello, > > when upgrading from 0.9.30 to 0.9.31 in the GNU Guix distribution, > where all packages are built in a sparsely populated chroot, about > half the tests in testcurl fail with messages similar to > Failed to bind to port 11080: Address already in use > > I attach src/testcurl/test-suite.log. With the exact same setting, > 0.9.30 passes all its tests. Our curl version is curl-7.28.1. > > Andreas >
