QApplication takes argc by reference, so copying it in to AppManager's
constructor causes QApplication's to have an invalid reference once AppManager's
constructor returns.
Instead just reference main's argc.
Fix#6641
main() couldn't hold its own state since cleanup() needed access to it.
Having the state in a class allows QApplication::aboutToQuit to call
into a member function with access to the state.
Data used in logging still needs to be global due to
qInstallMessageHandler not accepting a void* to get back to this.
Set QGuiApplication attributes before constructing Qapplication as required,
by using comma operator to call an initialization function before construction
all members in the member initializer list.
Keep logic largely unchanged, but with a defined destruction order.
Destruction is still abnormal due to QApplication::aboutToQuit forcing
us to do partial cleanup before QApplication returns, since some OSes
will kill qTox before the QApplication returns.
Prior to 2f4e8dc3e8 we would take
the written ToxID and insert that straight into history without
any case check
Must be done prior to schema 11 since even though the UNIQUE constraint
on the peers table is fooled by the different case, the UNIQUE
constraint on the new chats and authors table which are stored as BLOBS
fail during upgrade when the two different case but equal ToxPks
collide.
Unfortunately it can't be done as its own upgrade since 11 was already
merged, and this is a prerequisite for 11 to pass for some users.
Execute prior to starting the split peer upgrade instead of as a larger
transaction for simplicity of the split upgrade, and since executing
this deduplication is idempotent.
Text comparisons in SQL are not case sensitive. Since we want to use this check
to specifically check for case mismatch, cast to blob which then does an exact
comparison.
Register in Widget's constructor so that handler can always get back to
Widget.
Add unregistration functionality so that the handler doesn't run when
we've returned to the login window or when exiting.
Covers char* without size, and QByteArray.
Catches the case of QByteArray to QString implicit conversion like was
fixed in 47ee51c61d, but still allows
construction or assignment from string literals in source. Gives most
of the type safety of QT_NO_CAST_FROM_ASCII without having to wrap
every literal string in QStringLiteral.
No functional issues found during change.
QByteArray usually holds arbitrary binary data, which can have null
characters or invalid characters when converted to ASCII or UTF8.
Disallow implicit casting for safety, forcing an explicit decision if
the conversion is wanted.
It's invalid to save a different profile than was loaded, so no need to
require the profile to be passed in again on save.
Removes dependence on Nexus singleton for Nexus::getProfile.
The function is documented as async, but is only async when called from
a different thread. Save is called from within Settings, it looks like
only from public functions that would be called on other threads, but
make it async regardless of calling thread for consistency.
There is already a sync function that is used, so safety is not lost.
Setting the parent of MessageBoxManager to Widget has a dual effect of
centreing the QMessageBox's on Widget' window which we want, but also
giving ownership of MessageBoxManager to Widget, which causes Widget to
destroy it in Widget's destructor. Since the original MessageBoxManager
must outlive Widget to be used in Settings and when loading Profiles
before Widget is constructed, we don't want Widget to destroy it.
Instead of juggling MessageBoxManager's parent around dynamically when
Nexus changes the active Window, just construct a second one in Widget
to be used by all its children with the parent set on Widget, centreing
the window and allowing it to have ownership.
Settings and Profile still use the original parent-less
MessageBoxManager since they show popups before Widget is constructed.
The same user may have sent a message both in the group and 1:1, so
generically remove aliases that have no associated messages, rather than
aliases that were used in the group.
Similarly, generically remove authors that have no associated aliases.
The group itself is a chat so will be removed through existing query.
peers had combined meaning, both being referenced by history for which
chat a message was in, and being reference by aliases for who authored a
message. This means that peers had conceptually different sub-groups:
all friends are both a chat and an author, but self is an author but not
a chat. With the addition of group chats this is amplified by groups
themselves being chats but not authors, and group members being authors
but not chats. Instead of having four sub-groups all within peers,
splitting peers into chats and authors gives a clean mapping,
simplifying interactions with the data.
In the new chats and authors tables, store what used to be a public_key
string as a BLOB, since it’s inherently a 32-byte binary value in both
cases. Call the public_key a UUID for chats, since group IDs are not
defined as public keys by toxcore.
Even though the data change is quite minor, the upgrade is large because
of SQLite's lack of support for modifying foreign key constrains for
existing tables. This means when peers are moved to new tables, all
tables referencing peers need to be cloned with a new foreign key
constraint, as well as all tables referencing those, recursively.
Replace underlength resume_file_id's with arbitrary data. Loaded file IDs are
not used for anything at this time, but they should not be null nor invalid.
Any 32-byte value is valid, so use all 0's for consistency.
Fix#6553
The implicit cast in addNewFileMessage's argument causes a conversion from the
binary data of file_id to be interpreted as UTF8 characters. On null bytes or
invalid UTF8, the resulting QString can be empty or truncated. Empty IDs causes
null file_restart_id's to be inserted into the database.
Fixes the cause of #6553, but the database damage still needs to be healed.
"finished" is not a field in file_transfers. Bug has been present since
initially introduced in d9b39b3102.
Had no real effect since file transfers are inserted with initial state
CANCELLED, which is usually the case for a 0-length finished file, but
it still logged an SQL error.
We already deep copy all other data types. Especially with Qt types like
QByteArray, having a deep copy to start is important since it shallow
copies on any subsequent copy, so the callback provider only gets a
shallow copy in the QByteArray of the temporary data from SQLite.
* Pass MessageBoxManager instantiation around instead of relying on a singleton
* Mock MessageBoxManager for unit tests when needed, since they don't have a
QApplication which is required for creating QWidgets
* Remove GUI class, which didn't have a clear purpose
Widget is the only caller, let it set its own title. Wrap inherited setTitle to
prepend "qTox". Check for thread is not needed since it is only called
from Widget's slots.
Style already has themeChanged, which was unused. GUI's themeReload just
dispatches to Style, which is already piped to GUI classes. Give the signal
to Style itself so that classes can connect to that, rather than relying on
GUI's singleton.
