| Age | Commit message (Collapse) | Author |
|---|
|
A batching mechanism was added to EventLoop::handleNextEvent, which
caused it to be renamed to handleNextEvents. The problem with the
batching was that it breaks EventLoop::removeEventsFromOwner: events
already grabbed off the events_ queue for invocation could be removed,
leading to issues in cases where two events were grabbed off the queue
that referred to the same entity, the second event was a timer event,
and the first event caused the timer to be stopped. The timer event
would in this case be executed, leading to unexpected behaviour or
crashes, as shown by the added unit test.
Test-Information:
Unit tests pass on Debian 9 and macOS 10.14.
Benchmarked the eventloop on Debian and macOS, and did not notice a
performance degradation.
Transferred files using S5B and IBB, and checked there were no UI hangs.
Transfer speed before and after the change are roughly the same.
Change-Id: Ife7312f533e8f0976c2e8077d16e0b63fbac6eb1
|
|
This fixes Doxygen warnings about missing parameter
documentation or documentation of non-existing parameters.
Test-Information:
Ran ./scons doc=1 without warnings with Doxygen 1.8.11.
Change-Id: I360d042e5c80f418e7574a26fa1283a82fc504d4
|
|
This cuts down our dependency on Boost further. Another
benefit is that mutex classes of C++11 standard library are
recognized by TSAN.
Test-Information:
Unit and integration tests pass on OS X 10.11.4.
Change-Id: Id4dcdb42e3d5155e107ce1d7618acbf26f913b6f
|
|
This change was done by applying the following 'gsed'
replacement calls to all source files:
's/\#include <boost\/shared_ptr\.hpp>/\#include <memory>/g'
's/\#include <boost\/enable_shared_from_this\.hpp>/\#include <memory>/g'
's/\#include <boost\/smart_ptr\/make_shared\.hpp>/\#include <memory>/g'
's/\#include <boost\/make_shared\.hpp>/\#include <memory>/g'
's/\#include <boost\/weak_ptr\.hpp>/\#include <memory>/g'
's/boost::make_shared/std::make_shared/g'
's/boost::dynamic_pointer_cast/std::dynamic_pointer_cast/g'
's/boost::shared_ptr/std::shared_ptr/g'
's/boost::weak_ptr/std::weak_ptr/g'
's/boost::enable_shared_from_this/std::enable_shared_from_this/g'
The remaining issues have been fixed manually.
Test-Information:
Code builds on OS X 10.11.4 and unit tests pass.
Change-Id: Ia7ae34eab869fb9ad6387a1348426b71ae4acd5f
|
|
Removed trailing spaces and whitespace on empty lines
in the process.
Changed CheckTabs.py tool to disallow hard tabs in source
files.
Test-Information:
Manually checked 30 random files that the conversion worked
as expected.
Change-Id: I874f99d617bd3d2bb55f02d58f22f58f9b094480
|
|
The old event loop had this protection, however it is
missing in the new design. This adds this protection again
as some external event loop implementations, e.g. the Qt
event loop, directly process their event queue from an event
handler.
Test-Information:
Unit and system tests pass on OS X 10.11.3.
Change-Id: I10ce7160c3f201e2d5f53ab8289ddde1eb3262e8
|
|
Test-Information:
Unit and integration tests pass on OS X 10.10.5 and Debian 8.2.
UI remains responsive when transferring a file between two
Swift instances.
Change-Id: I7841347a5d6c55121e02e274a7087a2fc200f879
|
|
The new event loop design has a single event queue that is
synchronized to protect against data races.
The removal of events in the queue by EventOwner is
deterministic.
Test-Information:
All unit and integration tests with TSAN and
cxxflags=-DBOOST_SP_USE_PTHREADS on Debian 8.2 pass without
reports.
Multiple Swiften/QA/ClientTest/ClientTest runs under
different CPU stress caused no TSAN reports on Debian 8.2.
Swift itself only causes TSAN reports related to Qt itself,
out of our control, and most likely false positives, i.e.
TSAN not detecting the synchronization method inside Qt
correctly.
Unit tests pass without errors and successfully connected
to Slimber on OS X 10.10.5.
Change-Id: Ia1ed32ac2e758c5b9f86e0dac21362818740881e
|
|
Change-Id: I94ab4bbb68c603fe872abeb8090575de042f5cb4
|
|
Includes fixes to:
Build with new Visual Studio and Boost
Avoid error caused when Avahi is present but Qt is not
Make declaration of XMPPRosterImpl::addContact consistent with implementation
Includes enhancements to:
Allow user-configurable mt.exe
Allow splitting openssl paths
Allow disabling gconf lookup
Make idn support optional
Allow disabling various library detections
Remove use of non-Python2.4 features in sconscripts
Test-Information:
Builds
Change-Id: Iee91ee80291a8bdf87cc169c915e4dad1cc1055b
|
|
Change-Id: I6d2364dc85464f238d95978793f35953a2947799
|
|
All applications succesfully link against Swiften.dll.
|
|
|
|
|
|
When EventLoop::handleEvent() was called recursively (i.e. by calling
processEvents() from a slot), weird things happened, especially in the
XMPP parser (assertion triggers, parse error from server, ...). Now, callbacks
are put in a queue handled by the topmost handleEvent.
Resolves: #592, #568
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Swift