maybe the code should go into 'default:' ?

> but this code makes the keyboard freak out and doesn't help smoothing CPU usage

I missed that SDL_EventState does the mask-generation already, so
the SDL_JOYEVENTMASK should be replaced by
SDL_JOYAXISMOTION | SDL_JOYBALLMOTION | SDL_JOYHATMOTION | SDL_JOYBUTTONDOWN | SDL_JOYBUTTONUP

Nope, doesn't work.
Neither does

1SDL_EventState(
2		 SDL_EVENTMASK(SDL_JOYAXISMOTION) |
3		 SDL_EVENTMASK(SDL_JOYBALLMOTION) |
4		 SDL_EVENTMASK(SDL_JOYHATMOTION) |
5		 SDL_EVENTMASK(SDL_JOYBUTTONDOWN) |
6		 SDL_EVENTMASK(SDL_JOYBUTTONUP),
7		 SDL_ENABLE);

nor

1SDL_EventState(SDL_JOYAXISMOTION,SDL_ENABLE);
2	SDL_EventState(SDL_JOYBALLMOTION,SDL_ENABLE);
3	SDL_EventState(SDL_JOYHATMOTION,SDL_ENABLE);
4	SDL_EventState(SDL_JOYBUTTONDOWN,SDL_ENABLE);
5	SDL_EventState(SDL_JOYBUTTONUP,SDL_ENABLE);

Dunno then, don't have an analog joystick to test.
Does the keyboard work at least (in the 2nd version you posted,
the first one does the same bad thing as my first code snippet).

no, keys are rarely recognized.

Does this code work?

1--- ./src/gui/sdlmain.cpp	Fri Jul  7 16:07:56 2006
2+++ ./src/gui/sdlmain.cpp	Fri Jul  7 16:06:24 2006
3@@ -1646,8 +1646,12 @@
4 static Bit8u laltstate = SDL_KEYUP;
5 static Bit8u raltstate = SDL_KEYUP;
6 
7+int poll_delay=0;
8+
9 void GFX_Events() {
10 	SDL_Event event;
11+	int time=SDL_GetTicks();
12+
13 	while (SDL_PollEvent(&event)) {
14 	    switch (event.type) {
15 #ifdef WIN32
16@@ -1929,6 +1935,15 @@
17 				}
18 			}
19 			break;
20+		case SDL_JOYAXISMOTION:
21+		case SDL_JOYBALLMOTION:
22+		case SDL_JOYHATMOTION:
23+		case SDL_JOYBUTTONDOWN:
24+		case SDL_JOYBUTTONUP:
25+			if (time-poll_delay>20) {
26+				SDL_EventState(SDL_EVENTMASK(SDL_JOYAXISMOTION) | SDL_EVENTMASK(SDL_JOYBALLMOTION) | SDL_EVENTMASK(SDL_JOYHATMOTION) | SDL_EVENTMASK(SDL_JOYBUTTONDOWN) | SDL_EVENTMASK(SDL_JOYBUTTONUP),SDL_ENABLE);
27+			} else SDL_EventState(SDL_EVENTMASK(SDL_JOYAXISMOTION) | SDL_EVENTMASK(SDL_JOYBALLMOTION) | SDL_EVENTMASK(SDL_JOYHATMOTION) | SDL_EVENTMASK(SDL_JOYBUTTONDOWN) | SDL_EVENTMASK(SDL_JOYBUTTONUP),SDL_IGNORE);
28+			break;
29 		case SDL_MOUSEMOTION:
30 			HandleMouseMotion(&event.motion);
31 			break;

The problem is inside SDL_PollEvent(&event) which we want to not (always) poll the joystick. Therefore the code must be before the SDL_PollEvent call.

> Does this code work?

Well you're effectively kicking out the joystick, as it is only re-enabled in
the (not anymore called) SDL_JOYxyz events.

> no, keys are rarely recognized

That's really odd. Could be that they're flushing the event queue when
calling SDL_EventState, so maybe we shouldn't call the ENABLE that
often but only once.

That is like the following:

1static int poll_delay=0;
2static bool poll_ignoring=false;

...

and at the end (!) of GFX_Events() :

1	int time=SDL_GetTicks();
2	if (time-poll_delay>20) {
3		poll_delay=time;
4		SDL_EventState(SDL_JOYAXISMOTION | SDL_JOYBALLMOTION | SDL_JOYHATMOTION | SDL_JOYBUTTONDOWN | SDL_JOYBUTTONUP, SDL_ENABLE);
5		poll_ignoring=false;
6	} else {
7		if (!poll_ignoring) {
8			poll_ignoring=true;
9			SDL_EventState(SDL_JOYAXISMOTION | SDL_JOYBALLMOTION | SDL_JOYHATMOTION | SDL_JOYBUTTONDOWN | SDL_JOYBUTTONUP, SDL_IGNORE);
10		}
11	}

Seems to work (keyboard-wise) though it might still loose keys. The modified
sdl.dll might still be the preferred option.

It doesn't help the joystick.
There is something strange about this SDL stuff...
SDL_JOYAXISMOTION is defined as 8, SDL_JOYBALLMOTION as 9 and so on.
ORing all these constants does not really make sense to me. They overlap. But also SDL_EVENTMASK(SDL_JOYAXISMOTION) does not work (SDL_EVENTMASK(x) is defined as 1<<x) because SDL_EventState only takes uint8 as first parameter and 1<<8 is too big.

Hm indeed, we need to call SDL_EventState() for every event.
Everything else won't work. Then they build the correct bitmask
using the SDL_EVENTMASK makro.

That helps the CPU usage, but input (joy and key) behaves really bad now. Joystick keeps moving, buttons don't always respond and the same with the keyboard keys.
It's worse on a slower computer.

Yes, as i said the clearing of the event buffer in SDL_EventState() is pretty
bad (that's why i always tried the oring of events...) thus my idea is not
really usefull. Guess the SDL_EventState() function should be called only
once at program start, and not repeatedly.

Hmm interesting comment.

1/*
2 * Enable/disable joystick event polling.
3 * If joystick events are disabled, you must call SDL_JoystickUpdate()
4 * yourself and check the state of the joystick when you want joystick
5 * information.
6 * The state can be one of SDL_QUERY, SDL_ENABLE or SDL_IGNORE.
7 */
8extern DECLSPEC int SDLCALL SDL_JoystickEventState(int state);

Do you think that'd work? That is disabling the joystick events and
call the SDL_JoystickUpdate() ourselves??

And then call SDL_JoystickGetAxis and SDL_JoystickGetButton. We could do this in a PIC_Event.

Or maybe we can build a custom joystick event and call the mapper with it.
The rest was straightforward, hope to find some time to finish it and test it 😀

well we can just do that on a joystick port write (or the bios call)

That way we are really the same as a pc.

Dunno, but wouldn't this skip the mapper totally (never used the remapping
of joystick things so i don't know in what direction this must be supported).

well the remapping of joysticks is currently not supported.
It was the diea, but we still have to figure out some sort of directional feeling as well when mapping it to the keyboard.
Maybe with some define or so.