We now set deviceScale to always the mapping from logical window system pixels
to device independent pixels (which is what Ebiten API users expect), and
introduce a new concept of videoModeScale that maps from video mode to logical
window system pixels.
videoModeScale is now only used for computing full-screen resolutions, while
deviceScale is used for any other conversion.
Fixes window sizes on X11, should be a NOP otherwise.
Closes#1774.
This should fix fullscreen mode on Linux/X11 systems in general,
while not affecting other systems.
Note that this deletes a bunch of OS X specific and Windows specific code,
as GLFW already provides this functionality.
This change is not expected to cause regressions, however,
the current behavior is still wrong and leads to wrong/unintended window sizes.
To be fixed in further PRs.
Updates #1774
This is basically a revert of 93a156a718.
This implements parsing the SDL gamepad mappings by Ebiten instead
of GLFW, so that Ebiten can handle parsing errors completely.
Closes#1802
This is only supported on desktops yet (on mobile standard layout isn't
implemented yet, and on the web this is the browser's responsibility).
Closes#1723
This change introduces the standard gamepad layout. This changes adds
these APIs:
* func HasGamepadStandardLayoutMapping
* func IsGamepadStandardButtonPressed
* func GamepadStandardAxisValue
* type StandardGamepadButton
* type StandardGamepadAxis
The standard gamepad layout is based on the web standard. See
https://www.w3.org/TR/gamepad/#remapping.
On desktops, the SDL's gamecontrllerdb.txt is used. If the gamepad is
listed in the text file, the mapping works. GLFW's mapping featrue is
not used.
On browsers, the property of a gamepad 'mapping' is used. When the
mapping value is 'standard', the gamepad is recognized to have the
standard mapping.
On mobiles, the implementation is still WIP.
Updates #1557
In `internal/uidriver/glfw/run_notsinglethread.go`, if the `UserInterface.init()` function returns an error, the `loop` is
executed regardless and the error is discarded. This behavior will hide the error returned by `init()` and might trigger
some crashes (see #1688).
A partial fix was implemented in 6c4edf8 , however that commit alone is not enough: the code now is correctly returning
the error via the `ch` channel, but it still executes the `loop()` function. This merge request skips `loop()` call if `init()`
had an error.
Updates #1688
IsWindowBeingClosed reports whether the window is being closed by
the user.
SetWindowClosingHandled sets whether the window closing is handled
or not. If the state is true, the window is not closed immediately
by the user and the game can handle the closing state. In this case,
the Update function should return an error in order to end the game.
This change also adds examples/windowclosing.
Closes#1574
Callbacks created by NewCallbackCDecl were never released and then
they are leaked, especially when the window size was changed by
SetWindowSize on Windows.
This change defines new callback ID types with uintptr, and reuse
the callbacks.
Closes#1672
glfw.PollEvents might invoke multiple FramebufferSize callbacks in
theory, this is very rare though. In this case, the sending an object
to the channel never ends.
This change fixes this deadlock by using 'select'.
Closes#1618
When the window is fullscreen, calling MaximizeWindow never returned.
Apparently the attribute glfw.Maximized can never be true on the
fullscreen mode.
This change fixes the issue by checking the fullscreen state.
On Linux (and probably other UNIX OSes), glfwRestore, glfwMaximize,
and glfwIconify returns even though the window state has not changed.
Wait for its finish by a for loop and glfw.PollEvents.
Closes#1608
At the initial phase, GLFW window's GetSize seems unreliable. Do not
call it on Linux. On the other hand, this is still necessary on macOS.
Updates #1606Closes#1609
