Add Android support

[cyberarm]

Draft PR for tracking changes needed to compile for Android

See: https://github.com/cyberarm/gosu-android for Android project
Current goal is to get a very basic gosu app running

[cyberarm]

@jlnr Do you know why the app crashes when FileAndroid::read returns?
The expected log output of "Read asset." from FileAndroid is printed but the log in IO of "Loaded file: filename" doesn't.

Call tree:

Gosu::Image("gosu.png");

gosu/src/BitmapIO.cpp

Lines 55 to 62 in 0a58b39

	Gosu::Bitmap Gosu::load_image_file(const std::string& filename)
	{
	Buffer buffer;
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Loading image: %s\n", filename.c_str());
	load_file(buffer, filename);
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Loaded image: %s\n", filename.c_str());
	return load_image_file(buffer.front_reader());
	}

gosu/src/IO.cpp

Lines 52 to 59 in 0a58b39

	void Gosu::load_file(Buffer& buffer, const string& filename)
	{
	File file(filename);
	buffer.resize(file.size());
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Reading file: %s\n", filename.c_str());
	file.read(0, buffer.size(), buffer.data());
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Loaded file: %s\n", filename.c_str());
	}

gosu/src/FileAndroid.cpp

Lines 72 to 80 in 0a58b39

	void Gosu::File::read(size_t offset, size_t length, void* dest_buffer) const
	{
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Reading asset with length: %d/%d...\n", length, size());
	AAsset_seek(pimpl->asset, offset, SEEK_SET);
	AAsset_read(pimpl->asset, &dest_buffer, length);
	__android_log_print(android_LogPriority::ANDROID_LOG_INFO, "Gosu", "Read asset.\n");
	}

2021-02-08 14:45:17.065 8048-8103/org.libsdl.app I/Gosu: Loading image: gosu.png
...
(List of files in assets/ dir)
2021-02-08 14:45:17.066 8048-8103/org.libsdl.app I/Gosu: Asset filename: gosu.png
...
2021-02-08 14:45:17.066 8048-8103/org.libsdl.app I/Gosu: Opening asset: gosu.png
2021-02-08 14:45:17.066 8048-8103/org.libsdl.app I/Gosu: Reading file: gosu.png
2021-02-08 14:45:17.066 8048-8103/org.libsdl.app I/Gosu: Reading asset with length: 1617/1617...
2021-02-08 14:45:17.067 8048-8103/org.libsdl.app I/Gosu: Read asset.
2021-02-08 14:45:17.067 8048-8103/org.libsdl.app A/libc: Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0x30518635e7854 in tid 8103 (SDLThread), pid 8048 (SDLActivity)

[jlnr]

@cyberarm You are passing &dest_buffer into AAsset_read, which will read into the memory that starts with the dest_buffer parameter on the stack. The extra & is the problem.

[cyberarm]

Pointer, reference, value... 🤦🏽‍♂️ Thanks, @jlnr, it compiles now. 😀

[cyberarm]

@jlnr How do we want to implement Gosu's touch callbacks?
SDL provides us with 3 touch events SDL_FINGERMOTION, SDL_FINGERDOWN, and SDL_FINGERUP SDL TouchFingerEvent

SDL doesn't provide a touch_cancelled event.

Do we want to keep the struct based callback?

struct Touch {
  int id; // device_id + finger_id?
  int device_index, finger_index;
  double x, y;
};

void touch_began(Touch touch);
void touch_moved(Touch touch);
void touch_ended(Touch touch);

or use a more verbose callback?

void touch_start(int device_index, int finger_index, double x, double y);
void touch_move(int device_index, int finger_index, double x, double y);
void touch_end(int device_index, int finger_index, double x, double y);

[jlnr]

Hmm, good question(s). Some thoughts after a busy couple of weeks:

• I like the struct because it's easier to add stuff like touch pressure later on.
• I also like the concept of touch_cancelled. I think for the most part it can be replaced with the lose_focus callback – basically, if something happens in the foreground while the finger is down, we should not count the lost (cancelled) touch as a button tap, and so on.
  Let's leave it for now, even if only iOS calls it (which is already the case for the lose_focus callback...).
• Yes, I think we should try to make it a single id. Maybe hashing both ids together is safer than relying on addition or other simple operations, but we can always tweak that later.

[cyberarm]

@jlnr what is the best way for Gosu::Input to access Gosu::Window's instance of Gosu::Graphics for width() and height() to convert SDL's normalized (0..1) screen touch coordinates to window coordinates?

[jlnr]

@jlnr It is actually supposed to work the other way around: Window should set the mouse factors of its Input object to scale correctly. Technically touches are not a mouse, but can we reuse the same mechanism? (Sorry, still no time to dig into the code a lot...)