Overlays with alpha channel in ScummVM use pre-multiplied alpha.
Thus, is it necessary to set the source blend function to ONE, not
SOURCE_ALPHA.
While there is no firm decision on the texture format to be used to
cursors, make sure to set the key color to R=G=B=A=0.0, so that it
can be used with either pre-multiplied or non-pre-multiplied blending.
The vanilla android toolchain even is unhappy when unknown folders are
present in the resources directory, so have moved the ouya specific
resources out to another directory.
The aapt throws an "invalid resource directory name" if the
resource directory name is not known to the tools. Unfortunately,
"values-television" is an OUYA specific addition and thus breaks
with the standard vanilla Android toolchain.
This fixes two minor issues when trying to cross-compile to Android on
Win32 using the Win32 Android SDK/NDK and MinGW shell.
However, this is still no working as DX and APKBUILDER are batch files
in the Win32 SDK (which then call the real Java tools) and these fail
to run in shell. However, using "cmd /C" also fails as the paths in the
parameters have unix, rather than native separators. Unsure how to fix.
This parameter overrides any user choice in the GUI or config file,
which is not good UX design (and is making it hard to debug loading
issues associated with theme files).
If we want to set this to default to Modern, we should do this in
a different way i.e. do this by ConfMan.loadDefaultConfigFile()
This is probably related to the Android issue with the backend init
running before main() is called, rather than after...
This removes the need to convert the parameter to copyRectToScreen to
"const byte *", which is commonly used in games, which use Graphics::Surface
to store their graphics data.
All uses of the old target scale API actually wanted to disallow scaling of
the mouse cursor. This commit adapts our API to this and thus simplifies
backend implementations.
Some backends, most notable the Wii and Android, did some implementation of
the cursor target scale, which I didn't adapt yet. I added a TODO for the
porters there.
If we leave it for OSystem, the DefaultTimerManager destructor tries
constructing a StackLock which, predictably, calls a pure virtual
function, because OSystem_Android's destructor is finished.