As a convenience, int overloads are provided for IntPtr size
parameters (corresponding to BufferSize or size_t). In the vast
majority of cases, a 32bit int is sufficient for buffer sizes,
so these overloads avoid the necessity of annoying (IntPtr) casts.
If more than 2^31-1 elements are required, the IntPtr overloads
remain available. (As always, this requires a 64bit runtime
environment.)
These valuators are not available on all systems. We now check
explicitly for support, instead of returning invalid results.
Additionally, unknown valuator names are now printed to the
debug log.
It appears that Mono 3.4.0 fails to marshal SDL_JoystickGUID
correctly when used as a return value. Changing the fixed buffer
to a couple of long fields allows the struct to be marshalled
correctly.
On SDL, DPad inputs were not correctly translated from joystick
hat inputs.
Updated MappedGamePadDriver so that it will correctly recognize
combined joystick hat directions e.g. DPadLeft should be true when
DownLeft, Left, or UpLeft are returned from the joystick hat.
The stored delegate list would store delegates indefinitely, causing a
memory leak. It is now the responsibility of each class to maintain
references to any callback functions it defines.
Devices are now explicitly closed when they a DeviceRemoved event is
received. Additionally, exceptions are no longer allowed to escape into
unmanaged code, where they might crash the runtime.
Callbacks are now explicitly marked as unmanaged pointers with the
correct (platform-defined) calling convention. Exceptions raised inside
these callbacks are no longer allowed to escape into unmanaged code,
where they will crash the runtime.
Additionally, the window is now only closed on the UI thread, and only
if it is actually open.
CocoaNativeWindow now checks for and reports failures to allocate
resources in its constructor. Additionally, it no longer calls UI
methods in the finalizer thread, as those methods will crash on Cocoa.
There are static methods that rely on `OpenTK.Platform.Factory` having
a static constructor. We need to fix those first, before we can remove
this constructor.
OpenTK.dll.config was previously set to be always copied to the
output directory. This caused constant rebuilds in Visual Studio.
Setting this to "Preserve Newest" improves build times
significantly.
Nvidia drivers fail in Glx.MakeCurrent() when using a 32bpp
visual on a window created with a 24bpp visual. Since we do
not know the actual visual until after the context is constructed,
the solution is to implicitly use 24bpp when 32bpp is requested.
The loss of the alpha channel does not have a user-visible effect,
since WinForms do not support translucent windows on X11.
OpenTK will now use the same GLXFBConfig to create the INativeWindow
and IGraphicsContext on Linux/X11. This resolves an issue where OpenGL
3.x contexts could not be created on some graphics drivers (e.g.
nvidia binary.)
The TTY keyboard driver requires a robust cleanup method to avoid
hogging the keyboard/console after the process exists. Without
this, it does not make sense to use enable this driver.
Non-accelerated contexts are now considered iff no accelerated contexts
are available. Additionally, a GraphicsException will be thrown if
context construction fails for any reason, instead of causing a runtime
crash.
We currently have three categories of string parameters: `string`,
`string[]` and `StringBuilder`. (OpenTK 1.2 adds one more: `ref
string`.) Each category needs to be marshaled separately into a native
character array.
This commit implements the following changes:
- string[] epilogues are now correctly emitted, instead of being
ignored.
- string[] prologues and epilogues now use the same local variable name.
- all epilogues are now generated with a single pass over the function
parameters, instead of requiring a separate pass for each category.
- string prologues and epilogues now allocate local variables based on
the relevant parameter *name* rather than the parameter *type*.
Fixes issue #144.
Fixes two issues:
1. FreeStringArrayPtr used the wrong variable in the offset to
ReadIntPtr causing an access violation.
2. Better cleanup of memory in MarshalStringArrayToPtr when any alloc
fails.
This introduces support for the following extensions:
- EXT_geometry_point_size
- EXT_primitive_bounding_box
- EXT_texture_array
- EXT_tessellation_point_size
To do that, we need a custom InputOnly window. This window is only used
for receiving the exit ClientMessage - XI2 input is still received
through the root window.
Fixes issue https://github.com/mono/MonoGame/issues/2711
You cannot create an X11/OpenGL context without the GLX extension.
OpenTK will now throw a `NotSupportedException` when this condition is
encountered.
In some cases, it may be possible to enable the EGL backend by passing
`GraphicsContextFlags.Embedded` to the `GraphicsContext` constructor.
External mice will now report horizontal scrolling in the low-level
`Mouse.GetState()` API. The touchpad apparently creates an emulated
mouse device that does not report horizontal wheel scrolling events.
Touchpad support probably requires a proper multitouch implementation.
OpenTK uses the win32 wheel coordinate system, where
(+h, +v) = (right, up). XI2 uses (+h, +v) = (right, down)
instead, so we need to flip the vertical offset.
Fixes issue #133 and https://github.com/mono/MonoGame/issues/2686
Since GL_TRUE and GL_FALSE match .Net true and false (1 and 0,
respectively), we can simply reinterpret the byte value as a bool.
In the future, we could issue a `(if value == 0 then false else true)`
statement for added safety, but this does not appear to be necessary
right now.
GLboolean is typedefed as an unsigned char. It appears that `[return:
MarshalAs(UnmanagedType.U1)]` is ignored in calli p/invokes, so we need
to take care of this ourselves by emitting byte instead of bool for the
p/invoke signature. The public API remains unchanged.
Fixes issue #129.
The original C code had if(m & 0x00800000) which is true if the
expresssion does not evaluate to zero. This was inncorretly translated
to the C# code if((m & 0x00800000) == 1) which only evaluates true if
the expression evalaute to 1, which it never does. The correct test is
to test not equal to zero (!= 0).
This test checks every public API under OpenTK.Audio, OpenTK.Compute
and OpenTK.Graphics for compatibility. Incompatible changes will cause
a compilation failure.
Versions prior to 2.0.0 are not ABI-compatible with 2.0.x and
attempting to use those will result in random instability. We now
explicitly check the SDL2 version before enabling the SDL2 backend.
This was caused by calling XDestroyWindow inside Dispose(), or
by processing events after calling XDestroyWindow. Pending events
are now discarded when Close() is called and not cancelled, and
XDestroyWindow is only called if the window exists.
We now use the display connection specified in EglWindowInfo
throughout the EGL backend, instead of opening new connections
in EglGraphicsMode. This fixes EGL errors in
Egl.CreateWindowSurface.
As part of this fix, all instances of
IPlatformFactory.CreateGraphicsMode() have been removed. This
API is no longer used in OpenTK 1.1
We have to load statically exported entry points using the regular OS
symbol loading facilities, and only use eglGetProcAddress for
extensions.
Fixes crash in https://github.com/mono/MonoGame/pull/2377
Mono can use ldlen and ldelema on both 1d arrays (vectors) and 2d/3d
arrays. However, .Net can only use these instructions on 1d arrays -
higher rank arrays must use get_Length and Address explicitly.
The scrolling valuator offsets are now calculated correctly, based on
the number of set bits in the valuator mask.
Additionally XI2Mouse now runs on a background thread, using blocking
XIfEvent. This should improve resonsiveness and reduce CPU usage, due
to reduced polling.
