On Wed, Apr 1, 2009 at 2:12 AM, René Dudfield
<renesd@xxxxxxxxx> wrote:
hello,
An important project for gsoc could be to port pygame to SDL 1.3
If anyone wants to make a proposal for this, it would be important to
list the new features of SDL 1.3 that need to be wrapped, and tested.
Below are some of the planned features - many of which are implemented
already. The project should try and target the stable parts of the
API - since SDL 1.3 is still in development (but is getting closer to
being ready).
A first step would be to get it working at all - then to start
wrapping new functionality.
It's probably easier to start with the pygame 1.9x rather than
pgreloaded - since pgreloaded lacks tests... but to wrap new stuff for
pgreloaded (since it's cooler than new pants).
cu,
- Multiple windows! You don't need to have a single video surface anymore!
This is a new API (but SDL_SetVideoMode() still exists in the
compatibility layer).
- Multiple displays: SDL exposes details of what physical monitors are
hooked up to a machine and lets you control them individually.
- Formal API for positioning SDL windows: it sort of comes with the
multiple window/display support. No more setting environment variables
for this!
- Video device enumeration: you can get an idea of what APIs and hardware
are available to you.
- 2D acceleration: SDL can use OpenGL or Direct3D behind the scenes with
the 2D interfaces, so we can get acceleration on modern systems where
X11 or DirectDraw just aren't the fast paths anymore. The framebuffer-
oriented interfaces, like X11, are still there, though, for legacy
platforms and hardware.
- Texture support: the 2D interfaces now concern themselves with "textures"
and not surfaces. The assumption is that, even in 2D graphics, you now
want to try to push all the effort to the hardware when you can,
falling back to software where you can't. On the most basic level, this
just means you can't get at pixel-based framebuffers without locking
the "texture" and doing so may be much more expensive than in 1.2, but
in many common scenarios, a well-designed program can be significantly
more efficient in 1.3. There are some basic texture operations to offload
common per-pixel operations to hardware so you may not have to lock the
texture and do it yourself. This is meant to be a very simple API,
however: those needing more, even in 2D, should consider using OpenGL
directly.
- Multiple input devices: we'll be merging ManyMouse into SDL, which
will let us handle input from more than one connected mouse on at least
Linux, Mac OS X, and Windows. If all you care about is the usual generic
pointer input, that'll still work, too, but if you want to make a game
where all your friends plug in a USB mouse and compete, that'll now be
possible. This is also interesting for experimental work that has nothing
to do with games, or Photoshop-style things that want mice AND pens, etc.
- Pressure/tilt support: mouse-like input devices like tablets will be
able to report their pressure/tilt through SDL, for those that want to
write art programs. TuxPaint, I'm looking at you here.
- Horizontal mousewheel events: SDL 1.2 only responds to the oldschool
vertical mousewheel, but many mice you buy now can scroll horizontally,
too, including the Mighty Mouse that ships with Apple desktops. There's
a formal event for this in SDL 1.3.
- Mousewheel no longer looks like a button: SDL 1.2 treats wheelup as
button 4 and wheeldown as button 5. This was an accident of history,
since XFree86 mapped the wheel to these buttons back when Loki had an
interest in making this work Right Now, but it wasn't guaranteed to be
those buttons, and it never was on other platforms. These have been
moved to formal events in 1.3.
- Separation of text input and key events: Basically, we're fixing the
Unicode support. The keyboard can be treated as a 101-button gamepad
with the usual KEYUP/KEYDOWN events, but there will be separate
events for text input, so IME implementations can let users compose
characters as they'd expect to on their platform. This will replace the
"unicode" field in 1.2 key events, which was universally unreliable
once you stepped outside of America and Europe.
- Formal API to permit screensaver: no more environment variable hack!
This is useful if you're writing a windowed application and not a game.
Audio:
- Audio capture support: You can record from a microphone, etc, and recover
this data through SDL.
- Audio device enumeration: You can find and choose specific audio devices.
- Multiple audio device support: You can open multiple devices and playback
different audio from each at the same time. Take all these things and
you could, for example, have a game where the audio plays through a
set of speakers, but speech from your teammates comes to a USB headset
while you reply through a microphone...and probably other interesting
things we haven't thought of yet.
- Audio device disconnect notification: if someone accidentally kicks out
their USB audio widget, the app can be notified, so they can pause and
let the user get reconfigured, etc.
- Support for PCM data in int32 format.
- Support for PCM data in float32 format.
- Non-power-of-two resampling. If you had to do something other than
double or halve your sample rate, SDL 1.2 would break (and probably
corrupt memory). This will be fixed in SDL 1.3. This is useful for
programs that worked fine with data at 22050Hz when everyone was using
audio cards at 11025Hz, 22050Hz, or 44100Hz, but now there are a lot of
motherboard cards that only eat at 48000Hz and break badly in 1.2.
- 7.1 output support.
- Macros to parse audio format enumerations.
No more need to do things like this:
const int bitsize = MyAudioFormat & 0xFF;
Now you can do the more readable:
const int bitsize = SDL_AUDIO_BITSIZE(MyAudioFormat);
(...and others like this.)
- SDL_MixAudio() won't need an open audio device: In SDL 1.2, you could
only mix audio buffers based on the opened device format. This has been
changed to a generic interface in SDL 1.3 that allows mixing data without
dependency on a specific device, or a device at all.
- Better audio format converters: these should be faster and work better,
since we generate all the possible cases in a perl script now instead of
trying to tapdance at runtime. This means converting a buffer takes one
pass regardless of format, instead of a pass to change bitsize, another
to change the format, another to change the signedness, etc. This is
cleaner to maintain and extend, and doesn't thrash the CPU cache as much,
etc.
Joysticks:
- Connect notification: In 1.2, you needed to restart the
joystick subsystem to find new sticks. In 1.3, your friend can walk up
and plug in his controller, and your game will be able to announce "a
new challenger approaches!"
- Disconnect notification: ...and you'll also be able to deal with the
controller getting kicked out midgame, or your friend getting tired of
losing and leaving with his controller.
- Force feedback: sticks and platforms that support vibration, rumble, etc,
will be controllable via SDL.
SDL_main:
- No longer required on any platform. Now it does what it was intended
for: hiding differences in main/WinMain/etc. Initialization details
are being moved into SDL_Init() where they belong. This makes it easier
to have SDL as a plugin, or use a non-SDL path without it hijacking your
mainline just to function. This is really useful for scripting languages,
since, say, a Python program might want to use SDL without linking SDL
directly to python.exe and replacing its main() function.
Misc:
- Atomic operations API: a way to deal with atomic operations like
test-and-set and compare-and-swap will be added. This is becoming
increasingly important in game development, and every processor and
platform handles this differently, so it'll be nice to abstract the
details into SDL.
- SDL_KillThread() is gone. It was never safe or portable. The function
will continue to exist for binary compatibility, but it will always
be a no-op that reports failure. If you need this, your program has
problems anyhow and needs minor reworking to cleanly handle thread
termination, even if KillThread was still available.
Planned but not yet fully considered:
- Some minor cleanups and clarifications to the RWOPS api are planned
to fill some gaps, but this is still work in progress.
- A basic API to read/write from the system clipboard would be nice.
- A basic API to support drag'n'drop with the system would be nice.
- Probably other stuff! Let Ryan know what you want!
cheers,