> On Nov. 20, 2014, 2:39 p.m., Dominic Hamon wrote: > > 3rdparty/libprocess/src/libev.hpp, line 39 > > <https://reviews.apache.org/r/28184/diff/3/?file=771698#file771698line39> > > > > this makes me uncomfortable but i can't quite pin down why. > > > > would a better construction be: > > > > ThreadLocal<bool>* in_event_loop() > > { > > static ThreadLocal<bool> in_event_loop; > > return &in_event_loop; > > } > > > > which gets you lazy initialization of the thread local but without > > macro usage, and avoids the issue of the non-determinism of static > > initialization. > > Benjamin Hindman wrote: > Cool, I like this better than the macros too! But lets make the static be > a pointer so that it doesn't get cleaned up non-determinisitically at program > shutdown. Also, I don't know what non-determinism you are referring to > regarding static initialization? > > Joris: can we do this for both __executor__ and __in_event_loop__ please? > Thanks!
Nice catch. Yes, a pointer will be safer. The global ThreadLocal in the current patch is initialized with " = new ThreadLocal...". There's no guarantee that this statement will be run before the object is accessed when a method is called from a different translation unit. The function static guarantees that when it is first accessed it is initialized. - Dominic ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/28184/#review62448 ----------------------------------------------------------- On Nov. 20, 2014, 2:34 p.m., Joris Van Remoortere wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/28184/ > ----------------------------------------------------------- > > (Updated Nov. 20, 2014, 2:34 p.m.) > > > Review request for mesos, Benjamin Hindman and Niklas Nielsen. > > > Bugs: MESOS-2125 > https://issues.apache.org/jira/browse/MESOS-2125 > > > Repository: mesos-git > > > Description > ------- > > This improves performance by short-circuiting run_in_event_loop to just call > the lambda if it is already in the event loop. > > > Diffs > ----- > > 3rdparty/libprocess/src/libev.hpp 04847e30b0d764d258b8c0971da87180043161f2 > 3rdparty/libprocess/src/process.cpp > 00cd89f7645db4401c92de25d3516047a5e21ff1 > > Diff: https://reviews.apache.org/r/28184/diff/ > > > Testing > ------- > > make check > 3rdparty/libprocess/benchmark improved throughput ~27% > > > Thanks, > > Joris Van Remoortere > >
