----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/39561/#review103641 -----------------------------------------------------------
trunk/qpid/cpp/src/qpid/broker/amqp/Session.cpp (line 645) <https://reviews.apache.org/r/39561/#comment161704> There's a teensy weensy race here: it is possible that accepted() will be called with sync=True even when clone() has been called (as all completers may finish while the clone() is executing). So the delivery will also be sitting in the pending set and gets re-settled on detach. Maybe the deliveries should be put on / removed from the pending set unconditionally? - Kenneth Giusti On Oct. 22, 2015, 5:37 p.m., Gordon Sim wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/39561/ > ----------------------------------------------------------- > > (Updated Oct. 22, 2015, 5:37 p.m.) > > > Review request for qpid, Alan Conway and Kenneth Giusti. > > > Bugs: QPID-6790 > https://issues.apache.org/jira/browse/QPID-6790 > > > Repository: qpid > > > Description > ------- > > This is an alterntaive fix that simply shortcircuits the asynchornous > completion for deliveries whose link has already been freed. It does this by > tracking deliveries for which asynchronous acceptance is pending, and on > deleting links removes the records of any deliveries affected. On handling an > asynchronous accept, the epnding set is consulted first and the accept only > proceeds if it is found. > > > Diffs > ----- > > trunk/qpid/cpp/src/qpid/broker/amqp/Incoming.cpp 1710066 > trunk/qpid/cpp/src/qpid/broker/amqp/Session.h 1710066 > trunk/qpid/cpp/src/qpid/broker/amqp/Session.cpp 1710066 > > Diff: https://reviews.apache.org/r/39561/diff/ > > > Testing > ------- > > Passes make test. > > > Thanks, > > Gordon Sim > >
