summaryrefslogtreecommitdiffstats
path: root/src/network/access/qnetworkrequest.cpp
diff options
context:
space:
mode:
authorShane Kearns <shane.kearns@accenture.com>2011-10-06 14:40:34 (GMT)
committerShane Kearns <shane.kearns@accenture.com>2011-10-06 16:13:26 (GMT)
commit7e662f3727e7c3dd3c41c29ed49bc41d2b66c744 (patch)
treeb317cd719656cb7d2dca1ea0128b4a4a63ded90a /src/network/access/qnetworkrequest.cpp
parent51ed77637a6001f66805e8091ce8616d8f2530db (diff)
downloadQt-7e662f3727e7c3dd3c41c29ed49bc41d2b66c744.zip
Qt-7e662f3727e7c3dd3c41c29ed49bc41d2b66c744.tar.gz
Qt-7e662f3727e7c3dd3c41c29ed49bc41d2b66c744.tar.bz2
Fix construction races in QtNetwork
When two threads construct a QNetworkAccessManager at exactly the same time on an SMP system, there are construction races for some Q_GLOBAL_STATIC data. This is normal and expected - the losing thread deletes its instance as part of the Q_GLOBAL_STATIC macro. However, for two of the classes, destruction of the loser had side effects. For QNetworkConfigurationMangerPrivate, there was a crash because of uninitialised variable on the losing side. For QNetworkAccessBackendFactoryData, a guard mechanism intended to prevent the data being reconstructed by destructors of other global static classes was being set by the loser. To fix this, the bool is changed to a QAtomicInt. In the normal case, it will have value 0->1 on startup and 1->0 on shutdown. In the race case, it will have values 0->1->2->1 on startup and 1->0 on shutdown. Task-Number: QTBUG-20343 Reviewed-By: mread
Diffstat (limited to 'src/network/access/qnetworkrequest.cpp')
0 files changed, 0 insertions, 0 deletions