Scale the coordinates passed to MouseMove() by the GTK scaling factor to
ensure that the mouse clicks are really delivered to the correct windows,
without this change wxUIActionSimulator was unusable on systems using high DPI
(i.e. GDK_SCALE of 2).
Per-monitor DPI scaling is still not supported however.
Adding a call to XSync() to wxUIActionSimulatorX11Impl seems to fix the
problems with the focus snapping back to the control having it previously
after simulating a mouse click and changing the focus afterwards using
wxWindow::SetFocus(), as the uiaction sample did.
Admittedly, it's not really clear why does doing it help, but at the very
leasy this XSync() call shouldn't do any harm.
Choose between wxUIActionSimulatorXTestImpl and
wxUIActionSimulatorPlainX11Impl dynamically, depending on whether XTest
extension is available during run-time or not.
Also decouple the two implementation to keep them clearly separated.
This is not a real move-ctor but std::auto_ptr<>-like "stealing" ctor. It
still allows to pass Display ownership to another function which is all that
is needed for our purposes.
Instead of opening connection to X display in every method, do it once in
ctor, as we can store the display across multiple calls after the recent
refactoring.
This does require allocating a separate "impl" object for each public object,
but should still be much less wasteful than opening and closing the display
connections all the time.
Extract platform-specific code in a wxUIActionSimulatorImpl-derived class
instead of keeping it in wxUIActionSimulator itself.
This will allow determining which implementation to use dynamically (i.e. at
run-time and not compile-time) to use later and already allows to get rid of
an __WXOSX__ #ifdef in common code.
The implementation using XSendEvent() with classic input events can't work
with GTK+ 3 anyhow because it uses XInput2 which is incompatible with them, so
warn the user about this and don't compile useless code into the library.
Just returning false is enough, there is no need to log a debug message as
well.
Also simplify the logic of this function a little by always returning true at
the end and returning false before on failure.
Closes#13759.
This merges a branch with a (slightly tweaked to apply) copy of the
cve-2016-0718-fix-2-2-1 branch from the official Expat repository at
https://sourceforge.net/p/expat/code_git/
On Gtk and OS X, but unfortunately not MSW, an assert fails if a
wxFileSystemWatcher is created too early. Therefore give a hint about
where and when one should be created.
Closes https://github.com/wxWidgets/wxWidgets/pull/287
The headers for this compiler, at least in the version packaged under Debian,
define BCN_DROPDOWN but not the NMBCDROPDOWN struct we also need, so we have
no choice but to redeclare the struct ourselves to allow this code to compile.
For some reason, "long" was used for window coordinates even though they're
really "int", and implicitly converting from the former to the latter resulted
in warnings in user code including this header with e.g. Apple clang.
Just use int to avoid conversions.
Selecting a leaf node in a wxDataViewCtrl and then pressing the right arrow
key resulted in using an out-of-bounds index for accessing the columns list.
Fix this by setting the current column to the first one, and not to the second
one which might not exist.
Closes#17537.
std::isnan() is always available when using C++11, so just use it.
Also pout the test for C++11 implementations of both wxIsNaN() and wxFinite()
first, eventually the rest of the checks will become obsolete and will be
removed.
Closes https://github.com/wxWidgets/wxWidgets/pull/283
Ensure the proper class hierarchy in wxGTK, this notably allows to override
SetValue() in classes derived from wxGauge, which didn't work in wxGTK (but
did in wxMSW and wxOSX) before.
Also remove the now unnecessary IsVertical().
When using the keyboard to select an item by typing the first characters of
its name, we need to make the item visible in addition to selecting it.
Closes#16336.
Since only geometry with closed sink (immutable) can be transferred to another geometry object with ID2D1PathGeometry::Stream() so we have to check if actual transfer succeeded and return NULL if not.
To ensure compatibility with Cairo renderer the new sub-path should be started at the joined endpoint of the current (just closed) sub-path.
Closes#17532
Because only ID2D1PathGeometry with closed sink (in the immutable state) can be transferred to another geometry object with ID2D1PathGeometry::Stream() so appending one wxGraphicsPath to another has to be done with source path geometry in the non-writable state and closing this geometry has to be done prior to any other operation.
But we want source wxGraphicsPath to stay in the writable state after appending it to another path so we have to prepare a writable copy of the source ID2D1PathGeometry and assign it to the source wxGraphicPath after the appending.
See #17532
Cairo renderer places MOVE_TO element into the path immediately after after the CLOSE_PATH element so to ensure that GDI+ will behave in the same way the new sub-path should be started at the joined endpoint of the current just closed sub-path.
See #17532
Since resulting wxGraphicPath should have the same state as appended path so we have to grab in wxGraphicsPath::AddPath also auxiliary data from appended wxGraphicsPath.
See #17532
When current point is not set before the call to wxGraphicsPath::AddCurveToPoint() then it should be set at first control point prior to adding a curve (function should behave as if preceded by MoveToPoint).
Closes#17526
When current point is not set before the call to wxGraphicsPath::AddCurveToPoint() then it should be set at first control point prior to adding a curve (function should behave as if preceded by MoveToPoint).
See #17526
