This allows leaf classes to update independently when the GUI has changed themes, without
their parent having to call updateTheme() manually.
Fix#5924Fix#5592
When a setting is set it locks the Settings mutex and emits a signal for that
setting. If a slot is connected to that signal and lives on the same thread as
Settings, it is executed immediately with the Settings mutex still locked. That
slot can then lock a series of other mutexes. During this time another thread
can lock a mutex and then try to read settings, which will cause a deadlock due
to the opposite order of multiple mutex locking. By always emitting signals
after unlocking the Settings mutex, we avoid holding the Settings mutex while
executing slots.
Fix#6218
Profile is required to load personal settings, and Profile was creating the
bootstraplist using personal settings. Instead create the bootstraplist in
initCore, after personal settings have been loaded. This avoids using an
uninitialized proxy type on every profile load.
- Seed random in Core Thread. Core Thread didn't seed random, resulting
in always using the same bootstrap nodes, even when you restart qTox
or change profiles.
- Use QDateTime::currentMSecsSinceEpoch() for seeding random. It
provides a bigger range of numbers than QTime::currentTime().msec()
does, and the latter somehow managed to result in approximately the
same first random number being generated, within a certain range.
- Use something a it more sensible than a mod operation to bound random
numbers within a range. It's not perfect either, but a lot better.
Using mod on random skews its distribution too much.
- Use QRandomGenerator's bounded() function to generate random values
within a range.
- Enable QRandomGenerator's usage starting with Qt 5.10.0.
QRandomGenerator is present since Qt 5.10.0, not 5.15.0.
- Bootstrap off every 5th node instead of two consecutive nodes. It's
likely that two consecutive nodes will have the same owner, which
makes some attacks more likely. The node selection algorithm should be
scraped and redone from scratch to be honest though.
This is a feature/fix to improve notification behavior when we receive
over 3 messages.
SnoreNotify prevents over 3 notifications from being displayed before a
user clears the notification. This is presumably to avoid infinite
notification spam.
Unfortunately this results in the notifications just coming in _after_
the user clears them. This means if there are 100s of messages built up
the user has to clear their notifications N messages / 3 times.
This feature/fix folds all notifications into a single notification
representing the current qTox notification state. See
notificationgenerator_test.cpp for what the new messages look like.
Previous exif transformations were not valid. The exif spec defines the
orientations as where the 0th row and the 0th column should end. The
previous mappings used in qTox did not respect these mappings and needed
to be updated.
We need to be lenient when reading from db because of ToxIds being saved in the
db from a bug introduced in e07d8d358f, which
used self ID rather than self Pk. Was subsequently fixed in
033f28e67e.
https://en.cppreference.com/w/cpp/numeric/random/random_device
"A notable implementation where std::random_device is deterministic is old
versions of MinGW (bug 338, fixed since GCC 9.2)."
Although time is less random off Windows, we don't need a strongly random
number for this, and using time avoids having to use platform specific code.
Fix#2917
* Reorder class data members and/or constructor initialisers to match,
reducing confusion about when members will be initialised.
* Remove (most) unused variables. Not removed: some global variables with
`TODO(sudden6)` on them for using them in the future. I don't know how
far into the future sudden6 wants to use them, so I left them there for
now.
* Distinguish different bootstrap nodes in the logs by index in the
bootstrap node list. Originally, we used to log the address/port of the
node we're bootstrapping to. This was removed out of privacy concerns
(even though the bootstrap nodes are public). This made the logs much
less useful when debugging why the client isn't connecting. Having
indices makes it easier to see that different nodes are being selected,
and makes it possible to determine which node was selected.
* Explicitly cast unused results of Tox API functions to `void` when all
we want is to know whether the function succeeds or not.
* Don't try to `#include <unistd.h>` on Windows. It does not exist on
MSVC.
* Remove extra `;` after function definitions.
* Remove reference indirection of QJsonValueRef, since a copy of that ref
(small pointer-like object) has to be made anyway when iterating over
QJsonArrays.
* Make some file-scope global state `static`.
* Use `nullptr` instead of `NULL`.
* Add `#if DESKTOP_NOTIFICATIONS` around the code that implements desktop
notifications, so it becomes a bit easier to compile everything with a
single compiler command - useful for manually running static analysers.
* Fix an error on MSVC where `disconnect` is looked up to be a non-static
member function and the `this` capture is missing.
* Consistently use `struct` and `class` tags for types.
* Use references in ranged-for where it reduces copies.
* Move private static data members out of the Style class and into
file-local scope. There is no need for them to be in the class. Also
marked them `const` where possible.
* Removed unused lambda capture.
* Ensure qTox can compile under NDEBUG with `-Wunused-variable` by
inlining the unused variable into the `assert` that was its only
target.
* Minor reformatting in core_test.cpp.
Our previous SQLCiper upgrade code attempted to set SQLCipher 4.0
default params. If SQLCipher 3.x was used at that time, it would result
in only half upgrading the params, since SQLCipher 3.x doesn't support
PRAGMA cipher_hmac_algorithm or PRAGMA cipher_kdf_algorithm. This means
that our databases could be saved with any of three sets of SQLCipher
params.
This will prevent qTox from auto-accepting arbitrarily large avatars. Avatars
are already limited by TCS 2.2.4 to 64KB, so we would only receive larger
avatars from badly behaving clients.
Don't add Wzero-as-null-pointer-constant by default, since on older Qt
versions that we stil support and that our CI runs agains, Qt API themselves
cause warnings which lead to build errors all over the place.
Fix#6008
Enable warnings for both. Favour casting to signed rather than casting to
unsigend for comparisons. Per isocpp's core guidelines ES.102, signed
arithmetic gives more expected results. If we need one extra bit of range,
using longer signed types achives that.
Fix#6010Fix#6012
Message caching is handled by SessionChatLog in memory even when history is
disabled. ChatLog doesn't need to worry about how the messages its rendering
are being stored. Dynamic loading up and down in chatlog is sitll functional.
Checking if history pointer is valid is not sufficient, the setting must also
be checked. This caused asserts in history when history was disabled in
settings.
Makes the CoreAV thread own all ToxCalls in order to prevent signals
from being emitted via a Direct connection from Audio to CoreAV.
(cherry picked from commit 6b468e41fa)
In other applications chatrooms allow you to idle in a call and have
people hop in and out as desired. If a user is the only one presently
online in a group but knows someone will be joining shortly they should
be able to join the call ahead of time.
(cherry picked from commit 46d57c6864)
receiver QObject is used by Qt to automatically deregister the connection when
the receiver is destroyed. Forward it on to Qt's connect.
(cherry picked from commit 24e4ec3751)
Qt doesn't support QObject multiple inheritance, so use our existing interface
macros to declare signals in the interface without QObject, and implement them
in child classes.
(cherry picked from commit 82a4f1b412)
connection is normally returned from Qt's connect, and the caller may want to
track the connection to manually disconnect it.
(cherry picked from commit 41b2b35ce3)
c-toxcore calls the groupCallCallback from it's main thread instead of
the ToxAV thread as expected, this was triggering an assertion.
Aditionally the destructors of Core and CoreAV were fixed, because they
now either crashed or deadlocked qTox when it was closed while a group
call was still running.
(cherry picked from commit 141cbf8870)
This actually fixes two problems:
1) CoreAV and Audio thread both locked the callsLock and audioLock in
different orders, resulting in a deadlock of both threads. This fixed by
using a ReadWriteLock in the CoreAV thread.
2) Multiple functions were emitting signals while holding a lock. This
is unsafe, because the connected slot may acquire any other lock. This
is fixed by releasing the locks before emitting signals.
(cherry picked from commit 4b9e4a571d)
It doesn't really make sense to assert that the callbacks are coming
from any other thread than CoreAV, when we actually want to ensure the
callback is coming from Core thread.
Remove over agressive assert from sendGroupCallAudio(...), this function
should be callable from any thread.
(cherry picked from commit 9499925fb2)
We we're calling toxav_* functions without synchronizing to any of the
Tox threads.
Additionally remove the call timeout, it creates timers from different
threads, which causes errors.
(cherry picked from commit 98cfe9838f)
This commit fixes the behavior when a message is received while the
chatlog is scrolled to the bottom. With this change, the chatlog will
stick to the bottom when it is scrolled all the way down. If it is
somewhere in the middle (e.g. for search) the chatlog will not change
its position.
In other applications chatrooms allow you to idle in a call and have
people hop in and out as desired. If a user is the only one presently
online in a group but knows someone will be joining shortly they should
be able to join the call ahead of time.
Qt doesn't support QObject multiple inheritance, so use our existing interface
macros to declare signals in the interface without QObject, and implement them
in child classes.
c-toxcore calls the groupCallCallback from it's main thread instead of
the ToxAV thread as expected, this was triggering an assertion.
Aditionally the destructors of Core and CoreAV were fixed, because they
now either crashed or deadlocked qTox when it was closed while a group
call was still running.
This actually fixes two problems:
1) CoreAV and Audio thread both locked the callsLock and audioLock in
different orders, resulting in a deadlock of both threads. This fixed by
using a ReadWriteLock in the CoreAV thread.
2) Multiple functions were emitting signals while holding a lock. This
is unsafe, because the connected slot may acquire any other lock. This
is fixed by releasing the locks before emitting signals.
It doesn't really make sense to assert that the callbacks are coming
from any other thread than CoreAV, when we actually want to ensure the
callback is coming from Core thread.
Remove over agressive assert from sendGroupCallAudio(...), this function
should be callable from any thread.
We we're calling toxav_* functions without synchronizing to any of the
Tox threads.
Additionally remove the call timeout, it creates timers from different
threads, which causes errors.
This is a minimal fix to reduce risk for the release. A more complete
re-architecture will be made.
The netcam covers much of the chat in groups, and has nothing to show since
group video calls aren't possible. Who is speaking in call is already shown by
the bold names at the top of the group, taking much less space.
Fix#5918
