SDL1.2 internal state is not initialized before calling SDL_ functions

[ghost]

It seems 87ef78922abb8b3450b9ecfe6315dada9e9867c8 from pull request #14 broke the SDL1.2 initialization. On parts of the platforms the main function must be replaced and instead the private SDL-main (parachute) function has to be called first. This is the only use of this SDL_main.h include. The programs main function is redefined as SDL_main and the main function is included using a special static library. The normal startup will therefore be main(SDLmain.a) -> SDL_main(previously known called main in the actual program source code). Of course, main() is not always called main() but I just use it here as placeholder for the real name.

This special indirection is required to initialize the internal state of SDL before the actual program calls any SDL function.

A similar problem already caused the windows builds to break (in a very hard to debug way) in the past.

[ghost]

One way to deal with it would be to drop SDL 1.2 support

[littleguy77]

Thank you @conchurnavid for the perfectly clear explanation. Once again I cannot see beyond my own hands :(.

@richard42 @Narann I suggest reverting #14 immediately. That PR was one of several ways to address the windows build issues discussed in #12. I can submit a new PR to address #12 in a less invasive manner.

Three steps forward, three steps back :P

[Narann]

Thanks a lot for pointing this @conchurnavid. I've create the reverted PR.

@littleguy77 I would suggest a #ifndef SDL2 (with comments of course :D ) over the SDL_main.h.

There is a point where SDL1.2 support will be dropped but we would lost many platforms that still can run mupen64plus. And it's not just a question of dropping SDL1.2 but also cleanup many things. (What do you think @richard42).

[littleguy77]

I have no opinion on SDL 1.2. I merely posted #14 as one way to solve windows issues from #12, thinking I might as well cleanup what I thought was an obsolete dependency. I see no reason to surround the #include SDL_main.h with #ifndef SDL2 as it doesn't resolve the issues from #12 for windows. Might as well keep the SDL behavior the same regardless of SDL version.

[Narann]

I see no reason to surround the #include SDL_main.h with #ifndef SDL2 as it doesn't resolve the issues from #12 for windows.

Ok, I was assuming it was a potential fix. :(

[richard42]

Yeah I mentioned the original reason for the SDL_main linkage in PR #12 . As Narann mentioned, I would like to retain SDL 1.2 compatibility for now in order to support older platforms.

On another note, I've been working with Gonetz to get his new OpenGL 3.3 video plugin working in OSX, and in the course of this work I discovered that our video mode setting code in vidext.c is wrong for SDL2. If I get time this weekend I will rewrite this to use the correct (new) functions. It will be a rather low-level change, so I'm a little concerned about causing regressions.

[Narann]

Thanks @richard42.

So, we don't drop SDL1.2 for now and will wait a future PR from @littleguy77 adressing this problem.

I've been working with Gonetz to get his new OpenGL 3.3 video plugin working in OSX

Excellent news! I didn't see @gonetz code yet, I hope it's clean enough for future contribution. :)

About possible regression. If it's just few lines to change in every video plugin it wouldn't be a real problem right?

[littleguy77]

Thanks for the reminder @Narann. Will submit a new PR tonight.

[richard42]

The SDL2 change that I plan to make this weekend shouldn't require any changes to the video plugins (this is one nice thing about the video extension architecture). Mostly I'm concerned that it either: 1. won't work at all on some platforms, or 2. fullscreen/window mode switching won't work on some platforms. Of course, fullscreen switching already doesn't work on Windows due to a platform limitation.