|
From: Rhys K. <rhy...@gm...> - 2015-06-09 05:27:42
|
Valgrind developers, You will have seen the OS X-related commits over the last six months, and I wanted to provide an update on current status and where to next. Valgrind had putative support for OS X as a platform since the initial work of Greg Parker (since at least December 2004), Nicholas Nethercote, Julian Seward and others hit trunk in 2009. First release with highlighted support was 3.5.0, if I recall correctly. Since that time there have been five OS X releases (Mac OS X 10.6 Snow Leopard - OS X 10.10 Yosemite) and the platform saw the transition from gcc to a clang-based compiler chain. Valgrind support didn't quite keep up, and despite the efforts of many the Valgrind user experience often fell below expectations. Over the last six months I've been working to improve Valgrind's support for the OS X platform. Pleasingly, progress has been made and already the regression test failures are down ~20%. The improvement is actually better within the core memcheck toolset, as the % failure rate remains much higher in helgrind and drd. Based on my commit logs (performance slightly different between OS X releases, not the full time horizon): 1 March: == 585 tests, 257 stderr failures, 23 stdout failures, 0 stderrB failures, 0 stdoutB failures, 31 post failures == 8 June: == 598 tests, 200 stderr failures, 12 stdout failures, 0 stderrB failures, 0 stdoutB failures, 30 post failures == Now that I have development environments for OS X 10.7 through OS X 10.10, I'll look to setup frequent buildbots to post nightly regressions to the mailing list. Others are of course free to do so if you have the infrastructure, please consult the docs or ask on the mailing list/IRC! Based on the rate of bugfixes during 2015 YTD, we are on track to clear all core memcheck and 'none' regression test failures this year. This week's announcement of OS X 10.11 (El Capitan) is a good juncture to set an audacious goal of no new regressions in Valgrind on El Capitan when it is available for general release later this year. Stretch targets are day 1 support for Firefox and the Swift runtime. Progress is being tracked in the meta bug here: https://bugs.kde.org/show_bug.cgi?id=348909 Importantly, this OS X-related activity has also led to new users reporting bugs, providing patches and suggesting new tests. This is a positive for the sustained viability of the port. We would value your comments, thoughts and contributions to achieve our goal for Valgrind support on OS X 10.11. Regards, Rhys Kidd |