[
https://issues.apache.org/jira/browse/MESOS-3326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14718174#comment-14718174
]
Neil Conway commented on MESOS-3326:
------------------------------------
Initial work:
https://reviews.apache.org/r/37876/
https://reviews.apache.org/r/37877/
https://reviews.apache.org/r/37878/
Style-wise, I opted to explicitly call load() and store(), rather than rely on
the overloads for T() and operator=. I think there's some value in being
explicit, but let me know if you guys think otherwise.
This updates most of the places where volatile and/or GCC intrinsics were used,
but not all. Remaining places I noticed:
* libprocess/src/tests/process.cpp
* libprocess/include/process/owned.hpp
* libprocess/include/process/shared.hpp
* libprocess/include/process/logging.hpp
> Make use of C++11 atomics
> -------------------------
>
> Key: MESOS-3326
> URL: https://issues.apache.org/jira/browse/MESOS-3326
> Project: Mesos
> Issue Type: Bug
> Reporter: Neil Conway
> Assignee: Neil Conway
>
> Now that we require C++11, we can make use of std::atomic. For example:
> * libprocess/process.cpp uses a bare int + __sync_synchronize() for "running"
> * __sync_synchronize() is used in logging.hpp in libprocess and fork.hpp in
> stout
> * sched/sched.cpp uses a volatile int for "running" -- this is wrong,
> "volatile" is not sufficient to ensure safe concurrent access
> * "volatile" is used in a few other places -- most are probably dubious but I
> haven't looked closely
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
