SDL1 open issues

@Capehill
Something wrong with beret from github when built as SWSURFACE one. To reproduce:

1). replace HWSURFACE on SWSURFACE in 2 places in beret from github, recompile, run game
2). press alt+tab to going to the window mode
3). press alt+tab to going to fullscreenmode again

As result everything slow downs like hell :) 3-4 fps. And if i quit, it even didn't quit.

My old main.c rebuilded with new SDL give no that problem, so probably some code changes in beret about flags and co cause that.

@kas1e

Remove SDL_DOUBLEBUF when using SDL_SWSURFACE. SWSURFACEs are implicitly double-buffered.

Maybe possibility should be removed from the backend because it should not be even valid: https://www.libsdl.org/release/SDL-1.2 ... html/sdlsetvideomode.html

Can't test the issue myself right now, maybe next week.

@Capehill
Btw, for dbl-buffering, i see that minigl in os4sdl is playing dirty games with surfaces and dbl-buffering. At least i only checked on word "OPENGL" in video/amigaos4 directory and there few places which looks strange (with strange comments about using surfaces and dbl-buffering).

@kas1e

There is one check which removes SDL_DOUBLEBUF flag in OpenGL path, to avoid resource allocation. Do you mean that one? Double buffering seems to be handled separately and it's forced in OpenGL case.

I would like to do some changes, for example it would be nice to have bitdepth user-configurable (now it's 16-bit hard coded).

@Capehill
Quote:
That one yes, but also there is that part in the SDL_os4video.c:



Dunno how right comment about hack and needs to re-work surface setup code is at current state, through.

I also found that in video/SDL_glfuncs.h, we have that part:



Dunno if minigl still didn't have 1d texturing, but even if not, there we probably need somehow to make common ifdef, which we can later use for minigl, but didn't use for ogles and gl4es

@kas1e

There is cgl_GLCopyTexImage1D in MiniGL sources.

@Capehill
Strange that comment say about 1d, but function name ends on 2d still

@kas1e

That was only the function body. Check texture.c.

@Capehill
Right, i do check changelog of MiniGl, and it was added long time ago, in the 2.2 version. And that ifdef in SDL probably keeps there much longer, and we can then remove it at all.

@Capehill
Btw, did you know for what that video/SDL_glfuncs.h is need it at all ? I mean, didn't we use those gl fucntions we need , just taken includes and whatever from opengl itself, and use them in any SDL app ?

Or that file have some special meaning ?

@kas1e

I didn't check all sources but they seem to be used for OpenGL-blitting. Something we can ignore. At least I don't have interest to study the topic in SDL1. Easier to use graphics.library for 2D and OpenGL for 3D.

@kas1e

Spent quite much time to debug Beret issue. After all, the issue was that SDL creates a "shadow surface" (SW) under certain conditions and accelerated blits are canceled. In practise it could happen when asked for SDL_HWSURFACE|SDL_DOUBLEBUF but without SDL_FULLSCREEN.

Whole surface creation is so complicated business due to all combinations. Now, if the backend would use the same flipping code for both window and fullscreen, this problem might be solved but at the moment my advice is to pass SDL_DOUBLEBUF only with FULLSCREEN|HWSURFACE combo.

So, for accelerated blitting, either use:

SDL_FULLSCREEN|SDL_HWSURFACE|SDL_DOUBLEBUF or
SDL_HWSURFACE (window).

Or if your application needs to poke surfaces like many emulators do, maybe use just SWSURFACEs.

@Capehill
Porting some SDL1 stuff, and find out this part of code:



Which probabaly mean "let's hide the cursor". But , by some reassons when i use that code, i have in the cursor place some trailing black dots keeps.

So or code wrong (or wrong at all, or wrong for aos4), or that another sdl1 issue

@kas1e

Doesn't SDL_ShowCursor(int toggle) on os4 port exists?

www.libsdl.org/release/SDL-1.2 ... s/html/sdlshowcursor.html

@jabirulo
It all can be workarounded of course, point is exactly that code should't work ?

@kas1e

Example code seems to have bug. Cursor is defined to be 16*16. However, data and mask have only input for 8*16 cursor.

Double the data and mask array sizes to 2*16, or use width of 8 when creating cursor.

I didn't test it yet myself.

@Capehill
Thanks !

Found today some other issue (which can be again issues with game's code, but maybe not): i build some game via SDL1 / MiniGL which, when switch to/from fullscreen, or resize the gl window when in window mode, give a black window. Game logic works, but just black window.

If i start originally game in lets say fullscreen mode : it works.
If i start originalluy game in window mode: it works.

But once i trying to swith to/from window/fullscreen, then black window. Same for resizing when in window mode.

The functions in question are this ones:



(scroll till end for all of them).

Maybe its something with SDL_GetVideoSurface() ? Is looks like something with sys_resize() done wrong for us (or didn't done what we need to do). Maybe something with HWSURFACE there ?

On win32 same code works of course.

I use latest sld1 sources from repo.

@kas1e

SDL destroys existing OpenGL context in most cases when SetVideoMode is called. It might explain the issue if application assumes that old context exists after display change.

On the other hand, that fullscreen toggling workaround may not be needed. SDL_WM_ToggleFullScreen works, I tried testgl with ALT+ENTER combo.

Serial logs might give more information.

@Capehill
I think i found what issues is (at least in that test code).

There is original part of code:




And that didn't works. Window resizes , but black screen. Same to switch to fullscreen (its from another funciton, but i t also call that one in end).

But once i just change that line:

screen = SDL_SetVideoMode( width, height, screen->format->BitsPerPixel, screen->flags);

on that one:

screen = SDL_SetVideoMode( width, height, screen->format->BitsPerPixel, SDL_OPENGL | SDL_RESIZABLE);

Everything start works as expected. I.e. just change screen->flags on SDL_OPENGL | SDL_RESIZABLE.

What is more interesting, that in the debug output, flags in both cases are same OPENGL RESIZABLE. But in first case , when screen->flags is used, i have that kind of output when trying to resize window:



And that one, when all works, when i only change screen->flags on SDL_OPENGL | SDL_RESIZABLE:



See, it tries to resize then.

Question is : wtf ! Maybe that code produce something weird:



So, while in serial output we have same OPENGL RESIZABLE words, it still have something else in screen->flags, which make calling of os4video_SetVideoMode's Resizing window() skips ?

@Capehill

Created simple test case which show issue. There is code:



When compile and run this code over win32 , that what i have in output:

$ ./sdltest.exe
flags in dec = 18
flags in hex = 12
flags_from_screen in dec = 18
flags_from_screen in hex = 12

If i compile that for amigaos4 and run, then output are:

$ ./sdltest

flags in dec = 18
flags in hex = 12
flags_from_screen in dec = 16777234
flags_from_screen in hex = 1000012


So, in case with amigaos4, we have other value returned when do SDL_GetVideoSurface(). Seems it return something else in flags field (its not just 12 , but 1000012 , so another 1 at begining of byte).

Is it bug or feature ? At least, in case with SDL everything should works the same on all platforms, so i assume bug ?