----- Original Message ----- > From: "Aaron Conole" <acon...@redhat.com> > To: "Lance Richardson" <lrich...@redhat.com> > Cc: dev@openvswitch.org > Sent: Wednesday, March 2, 2016 5:40:32 PM > Subject: Re: [ovs-dev] [RFC 0/4] testsuite code coverage usability fixes, lcov > > Lance Richardson <lrich...@redhat.com> writes: > > > This series addresses usability issues when executing the testsuite > > on builds configured with --enable-coverage and introduces basic > > support for generating HTML code coverage reports via lcov. > > > > Currently when running "make check" on a build with coverage enabled, > > nearly all test cases will fail due to unexpected messages on stderr > > of the form: > > profiling:...*.gcda: Merge mismatch for function * > > It's important to note that this could happen for legitimate reasons as > well; I've submitted a patch for gcov to eliminate the need for a regex > added to each test case. > > > This happens because daemons created for each test case are killed > > at the end of the test case, and if a daemon happens to be killed > > while in the middle of writing coverage information to a gcda file, > > the gcda file will be corrupted. > > But, this is still an issue, and one I'm going to put on my list of > things to-do (because gcov should sortof recover from this anyway - > after all something bad could happen during execution). >
Hi Aaron, thanks for the feedback. With this patch series, *.gcda merge errors no longer occur because daemons are no longer killed while writing coverage information. But in general I do agree it would be better for all users of gcov if the gcov runtime was more robust in this situation. It's worth noting that, with this patch series, "make check" and "make check TESTSUITEFLAGS=-j8" produce the same results with --enable-coverage as without --enable-coverage. Without these changes, "make check" with --enable-coverage results in a large percentage of test cases failing (often nearly all of them). > > These errors can be avoided by ensuring that deamons have fully > > exited when given the "exit" command before executing any cleanup > > actions. > > > > Lance Richardson (4): > > tests: Make OVS_APP_EXIT_AND_WAIT() wait for process termination > > tests: make as() function as documented > > tests: consistently use OVS_APP_EXIT_AND_WAIT() for daemon termination > > tests: enable basic lcov code coverage reporting support > > > > .gitignore | 1 + > > INSTALL.md | 6 +++ > > tests/automake.mk | 17 ++++++++ > > tests/bridge.at | 2 + > > tests/completion.at | 2 +- > > tests/daemon.at | 2 +- > > tests/interface-reconfigure.at | 2 + > > tests/ofproto-dpif.at | 77 ++++++++++++++++++------------------ > > tests/ofproto-macros.at | 6 +-- > > tests/ofproto.at | 68 ++++++++++++++++---------------- > > tests/ovn-controller-vtep.at | 12 +++--- > > tests/ovn-controller.at | 15 ++++++- > > tests/ovn-nbctl.at | 2 +- > > tests/ovn-sbctl.at | 6 ++- > > tests/ovn.at | 88 > > ++++++++++++++++++++++++++++++++++++++---- > > tests/ovs-macros.at | 5 ++- > > tests/ovs-vswitchd.at | 8 ++-- > > tests/ovsdb-monitor.at | 4 +- > > tests/ovsdb-server.at | 15 ++++--- > > tests/vlog.at | 6 ++- > > 20 files changed, 233 insertions(+), 111 deletions(-) > > I very much approve this series. Not sure if you think it's complete > yet, but I'll be running it, because it does make a huge difference. > _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
