[
https://issues.apache.org/jira/browse/THRIFT-3038?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15228444#comment-15228444
]
ASF GitHub Bot commented on THRIFT-3038:
----------------------------------------
GitHub user jeking3 opened a pull request:
https://github.com/apache/thrift/pull/981
THRIFT-3038: fix up some volatiles in cpp
I addressed the concerns in TFileTransport (forceFlush_) and in
TThreadPoolServer, so items #2, #4, #5 from the original issue description.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/jeking3/thrift defect/THRIFT-3038
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/thrift/pull/981.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #981
----
commit cb4849abc4156d64532989344a4b2c2fa8be8148
Author: Jim King <[email protected]>
Date: 2016-04-05T17:00:24Z
THRIFT-3038: fix a couple races and removed volatile per analysis
----
> Use of volatile in cpp library
> ------------------------------
>
> Key: THRIFT-3038
> URL: https://issues.apache.org/jira/browse/THRIFT-3038
> Project: Thrift
> Issue Type: Bug
> Components: C++ - Library
> Affects Versions: 0.9.2
> Reporter: Adriaan Schmidt
>
> In the cpp library there are several member variables which are declared
> volatile, I believe with the intention of providing some sort of
> thread-safety.
> While volatile can be used in this way in Java and C#, in C++ it cannot! It
> does not provide any guarantees with regard to instruction (re-)ordering,
> i.e. there are no implied memory barriers like you would get by using
> explicit locking or atomic variables.
> This means that all uses of volatile should be examined, the volatile
> qualifier should be removed and replaced by proper synchronization.
> The affected member variables are:
> # NoStarveReadWriteMutex::writerWaiting_
> Unprotected read access in acquireRead(). Data race can be seen by running
> the unit test with Helgrind.
> # TFileTransport::forceFlush_
> Always accessed while holding mutex_. In this case, the volatile can just be
> removed.
> # TFileTransport::closing_
> Sometimes accessed while holding mutex_ (in combination with the notEmpty_
> Monitor),
> but, e.g., enqueueEvent reads closing_ without any synchronization.
> # TThreadPoolServer::stop_, TThreadedServer::stop_
> Accessed (read and written) without synchronization. These would probably be
> fine using an atomic data type. Or, use explicit locking or signaling.
> # TThreadPoolServer::timeout_, TThreadPoolServer::taskExpiration_
> Should probably use a lock.
> # Mutex.cpp has mutexProfilingCounter as static variable. This probably
> doesn’t break anything, but still the volatile serves no real purpose.
> While some of the fixes are probably simple, in general I think someone with
> better knowledge of the code should have a look at this.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
