On Oct 23, 2014, at 1:33 PM, Ben Pfaff <[email protected]> wrote: > On Thu, Oct 23, 2014 at 01:29:07PM -0700, Jarno Rajahalme wrote: >> >> On Oct 23, 2014, at 1:14 PM, Ben Pfaff <[email protected]> wrote: >> >>> On Thu, Oct 23, 2014 at 01:16:16PM -0700, Jarno Rajahalme wrote: >>>> >>>> On Oct 23, 2014, at 12:45 PM, Ben Pfaff <[email protected]> wrote: >>>> >>>>> On Thu, Oct 23, 2014 at 11:38:48AM -0700, Jarno Rajahalme wrote: >>>>>>> On Oct 23, 2014, at 8:22 AM, Ben Pfaff <[email protected]> wrote: >>>>>>> >>>>>>>> On Thu, Oct 23, 2014 at 12:04:18PM +0200, Thomas Graf wrote: >>>>>>>> Exposes the necessary structs and functions in classifier.h to allow >>>>>>>> test-classifier.c to use the header and link to the library instead >>>>>>>> of including classifier.c directly. >>>>>>>> >>>>>>>> Adds VLOG_DECLARE_THIS_MODULE() which allows to extend the namespace >>>>>>>> of a vlog module. >>>>>>>> >>>>>>>> Also adds an assert to VLOG_DEFINE_MODULE() to catch duplicate >>>>>>>> definitions with a proper backtrace. >>>>>>>> >>>>>>>> Cc: Scott Mann <[email protected]> >>>>>>>> Cc: Ben Pfaff <[email protected]> >>>>>>>> Cc: Gurucharan Shetty <[email protected]> >>>>>>>> Signed-off-by: Thomas Graf <[email protected]> >>>>>>> >>>>>>> Jarno, are you OK with this? It solves a problem building with >>>>>>> --enable-shared (which I'd add to the commit message): >>>>>>> http://archives.openvswitch.org/pipermail/dev/2014-October/047674.html >>>>>> >>>>>> I'd rather not expose the classifier internals in a shared library >>>>>> ABI, even if it is versioned. I'll see if there is another >>>>>> solution. What was the problem? >>>>> >>>>> Two different definitions of a "classifier" vlog module, in different >>>>> source files (lib/classifier.c directly and tests/test-classifier.c >>>>> #include'ing classifier.c). >>>> >>>> I tested building with ?enable-shared again on the master and it builds >>>> fine. >>> >>> Yes, it builds fine, but you end up with test failures. >> >> I thought the ?enable-shared test failures (which I see too) are due to this: >> >> $ make -k -j8 TESTSUITEFLAGS=1 check >> >> fails, the interesting piece in tests/testsuite.log: >> >> +2014-10-23T20:24:44Z|00001|daemon_unix|WARN|/home/jrajahalme/openvswitch/tests/testsuite.dir/0001/ovs-vswitchd.pid: >> open: No such file or directory >> +lt-ovs-appctl: cannot read pidfile >> "/home/jrajahalme/openvswitch/tests/testsuite.dir/0001/ovs-vswitchd.pid" (No >> such file or directory) >> >> $ ls tests/testsuite.dir/0001/ >> br0.mgmt db.sock lt-ovs-vswitchd.pid >> br0.snoop lt-ovsdb-server.36077.ctl run >> br1.mgmt lt-ovsdb-server.log stderr >> br1.snoop lt-ovsdb-server.pid testsuite.log >> cleanup lt-ovs-vswitchd.36111.ctl >> conf.db lt-ovs-vswitchd.log >> $ >> >> I.e., for some reason the names of the binaries have the ?lt-? prefix, which >> is not expected by ?lt-ovs-appctl?. > > These are a problem too but in a separate class. Scott's patch works > around the "lt-" prefix problem.
Is there some other patch doing that, as I do not see how the patch in this thread does that? Jarno _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
