@kas1e

SDL1 concurrency code may have some issues. At least testlock and perhaps some other case was problematic (crash when quit by CTRL-C). If you have time and interest, try to find the bug :)

@broadblues

Thanks for the suggestion.

Regarding usleep, I would have thought that Newlib would have had a timer context for each process. I can accept WONTFIX of course if it works as wanted.

There must be some software that does a crazy amount of tiny usleep calls per second. Perhaps in that kind of scenario problem might become more visible?

@kas1e

Regarding iconifying, "worry-free" solution would be the best. I think there should be an iconification gadget if window has the normal borders (whenever the controls are enabled).

@thellier

Ok. Seems that SDL2 user documentation needs an update regarding WinUAE and PatchCompositeTags.

@AmigaBlitter

Please, try to improve the bug reporting. I don't have time to guess what might have happened. What is missing in your report:

- SDL2 SVN revision or git hash
- link to failing binary
- link to source code
- some details what kind of drawing might have been used, like OpenGL or compositing
- possible serial logs. You should know by now that SDL2 logs lots of stuff, being still considered as "beta"

If I have about 30 minutes per day for SDL2 / Amiga, I would rather spend that on something useful.

@thellier

Do you know what exactly happen on WinUAE when CompositeTags() is called? Does it return with success or fail? SDL2 should have some serial debug in case CompositeTags() fails. Not sure if serial debug can be easily enabled with WinUAE though.

And why compositing doesn't fallback to native SW rendering?

@kas1e

Yeah, SDL2 uses usleep(). If usleep really opens timer.device every time, I think we should a) open a ticket against newlib and b) replace it with custom timer code in SDL2.

EDIT: confirmed to be "usleep issue" by calling usleep 10 times in a row.
EDIT2: BZ opened

@kas1e

I have created a branch called fix-opengl. SDL2's OpenGL version defaulted to 2.1, changed to 1.3 (which is the MiniGL's OpenGL version I think). Then also MiniGL fallback was removed so it should be a little bit safer now.

@kas1e

There is only checking for OpenGL ES 2.0. In other cases is falls back to MiniGL and seems that this has to be removed because it conflicts with user intention (SDL_opengles2.h).

Preprocessor has replaced glCalls anyway with IOGLES->glCall so disaster waits if ogles.library wasn't opened correctly.

Regarding timer.device, please provide a minimal example that demonstrates it. SDL2 shouldn't explicitly open timer.device.

***

SDL2 project has now been imported from Sourceforge to Github, new address: https://github.com/AmigaPorts/sdl2-amigaos4

Contact me if you have any issues or would like to join the party. I have invited Bszili and Afxgroup as well.

Open tickets haven't been yet imported. Feel free to create tickets for open issues!

***

@kas1e

Filesystem may need porting but smart_ptr and format likely not. SuperTux uses OpenGL or SDL Renderer. I didn't look further but I suppose it can be compiled without OpenGL if necessary.

You can check the serial log after HW_SURFACEs have been enabled: https://github.com/capehill/os4sdl/blo ... os4/SDL_os4surface.c#L414

@kas1e

Here is a list of modules that need building:

http://www.boost.org/doc/libs/1_66_0/ ... tml#header-only-libraries

Boost is a giant set of headers mostly.

@kas1e

SuperTux seems to have SDL2 support now. If old version uses SDL_Blitsurface and can use compositing, it would be interesting to see comparison.

@kas1e

Thanks for the binaries. Disabling #define DEBUG in src/video/amigaos4/SDL_os4video.c makes grafx starting faster so the problem was indeed the slowness of serial logging. Try to disable it.

HWSURFACEs can be enabled in window mode. Takes only time and coffee.

@kas1e

Ok, thanks, now it's clear. The OGLES2 version flags stay and affect the following test. Should be easy to fix.

@AmigaBlitter

I don't remember uploading SDL with compositing support to OS4Depot. There was almost no feedback / testing so there was no reason to make it more public.

http://www.amigans.net/modules/xforum ... id=100176#forumpost100176

@kas1e

Could you provide dynamically linked exe for testing? If it's easy to build I can make one myself of course. But if there are 20 dependency libs I might lose my interest :)

Is there anything on the serial? My fork has some logging on. In theory it might cause some load.

@kas1e

Can you show some console + serial log or your test diff because I'm not completely sure what is happening there? Probably just a test code issue but anyway..

@kas1e

1.2.15 configure script is now merged...what a mind-numbing exercise. Some observations: it enables -maltivec by default and refuses to build .so. This looks similar issue than with SDL2 (-no-undefined switch in Makefile).

Maybe somebody wants to play with it. For me it's not a priority since native makefile seems to do its work nicely.

Now some more testing and maybe 1.2.15 can be merged with master.

@kas1e

"But in r214 it didn't works anyway, it just open ogles2 context and exit, even if we call testOpenGL() fucntion."

Yeah, it was written to do that: open the library and create context.