Add custom get proc support

[Jupeyy]

Motvation: OpenGL isn't really well supported under windows (e.g. borderless fullscreen causes weird black flushes, or doesn't really load the taskbar correctly and stuff like that)

So in order to let the user decide if they need borderless fullscreen over "real" fullscreen, we want to support glondirect12 mesa drivers. They are, as of jan. 2022, slower than the native OpenGL drivers. So we want to have support for both.

Since GCC for windows doesn't support tricks like lazy dll loading, the easiest is to not being required to link OpenGL and add custom handlers for that.

SDL2 comes with SDL_GL_LoadLibrary which makes it easy to select a different opengl driver.

Minimal code example: e.g. by having mesa's opengl32.dll in a directory called "mesa" and all other required libs(for mesa opengl driver) where the working directory is(the .exe)

extern "C" void *GLEW_customGetProcAddress(const GLchar *name)
{
	return SDL_GL_GetProcAddress(name);
}

static void LoadOpenGLLibrary()
{
	const auto *pOpenGLLibName = SDL_getenv("SDL_OPENGL_LIBRARY");
	bool HasCustomOpenGLLib = pOpenGLLibName != nullptr;
	if(HasCustomOpenGLLib)
	{
		// message("sdl", "loading custom opengl32.dll: %s", pOpenGLLibName);
	}
	if(SDL_GL_LoadLibrary(nullptr) != 0)
	{
		// message("sdl", "loading opengl32.dll failed, please remove any start parameter. (error %s)", SDL_GetError());
	}
}

[...] somewhere in the backend
	SDL_InitSubSystem(SDL_INIT_VIDEO);
	LoadOpenGLLibrary();
        [...]
        glewContextInit(); // use context init, NOT glewInit

[...] somewhere at program start
      // optinal
      SDL_setenv("SDL_OPENGL_LIBRARY", "mesa/opengl32.dll", 1);
      SDL_setenv("GALLIUM_DRIVER", "d3d12", 1);

Important: This moves GL 1.1 functions into GLEWs loading mecanism, even on windows. I dunno if this can potentially break something, or what the exact reason was, this wasnt done before.

This probably only works with glewContextInit, because there were code paths that used wglGetCurrentDC. Maybe this can be improved for easier usage

Currently it requires GLEW_STATIC for the custom symbol

Here is the patch on DDNet, where we want to use it (https://github.com/ddnet/ddnet/pull/4600)

[nigels-com]

There is a build failure in GLEW_STATIC mode:

cc -DGLEW_NO_GLU -DGLEW_STATIC -O2 -Wall -W -Wshadow -pedantic -Iinclude -fPIC -Wcast-qual -ansi -pedantic -fno-stack-protector  -o tmp/linux/default/static/glew.o -c src/glew.c
--
967 | src/glew.c:18868:38: warning: redefinition of typedef ‘PFNGLGETSTRINGPROC’ [-Wpedantic]
968 | 18868 \| typedef const GLubyte* (GLAPIENTRY * PFNGLGETSTRINGPROC) (GLenum name);
969 | \|                                      ^~~~~~~~~~~~~~~~~~
970 | In file included from src/glew.c:34:
971 | include/GL/glew.h:1006:39: note: previous declaration of ‘PFNGLGETSTRINGPROC’ was here
972 | 1006 \| typedef const GLubyte * (GLAPIENTRY * PFNGLGETSTRINGPROC) (GLenum name);
973 | \|                                       ^~~~~~~~~~~~~~~~~~
974 | src/glew.c:18869:28: warning: redefinition of typedef ‘PFNGLGETINTEGERVPROC’ [-Wpedantic]
975 | 18869 \| typedef void (GLAPIENTRY * PFNGLGETINTEGERVPROC) (GLenum pname, GLint *params);
976 | \|                            ^~~~~~~~~~~~~~~~~~~~
977 | In file included from src/glew.c:34:
978 | include/GL/glew.h:993:28: note: previous declaration of ‘PFNGLGETINTEGERVPROC’ was here
979 | 993 \| typedef void (GLAPIENTRY * PFNGLGETINTEGERVPROC) (GLenum pname, GLint *params);
980 | \|                            ^~~~~~~~~~~~~~~~~~~~
981 | ar rv lib/libGLEW.a tmp/linux/default/static/glew.o
982 | ar: creating lib/libGLEW.a
983 | a - tmp/linux/default/static/glew.o
984 | strip -x lib/libGLEW.a
985 | cc -DGLEW_NO_GLU -O2 -Wall -W -Wshadow -pedantic -Iinclude -fPIC -Wcast-qual -ansi -pedantic -fno-stack-protector  -o tmp/linux/default/shared/glewinfo.o -c src/glewinfo.c
986 | src/glewinfo.c:131:13: error: redefinition of ‘_glewInfo_GL_VERSION_1_1’
987 | 131 \| static void _glewInfo_GL_VERSION_1_1 (void)
988 | \|             ^~~~~~~~~~~~~~~~~~~~~~~~
989 | src/glewinfo.c:122:13: note: previous definition of ‘_glewInfo_GL_VERSION_1_1’ was here
990 | 122 \| static void _glewInfo_GL_VERSION_1_1 (void)
991 | \|             ^~~~~~~~~~~~~~~~~~~~~~~~
992 | src/glewinfo.c:122:13: warning: ‘_glewInfo_GL_VERSION_1_1’ defined but not used [-Wunused-function]
993 | make: *** [Makefile:195: tmp/linux/default/shared/glewinfo.o] Error 1
994 |  
995 | [Container] 2022/01/15 10:41:28 Command did not exit successfully make exit status 2
996 | [Container] 2022/01/15 10:41:28 Phase complete: BUILD State: FAILED
997 | [Container] 2022/01/15 10:41:28 Phase context status code: COMMAND_EXECUTION_ERROR Message: Error while executing command: make. Reason: exit status 2
998 | [Container] 2022/01/15 10:41:28 Entering phase POST_BUILD

[nigels-com]

I can understand the motivation and practicality of this proposed change. There is some history here of MX mode which was unsupportable for core contexts for some reason I don't recall too clearly.

I'm not decided if GLEW can reasonably accommodate this sort of change, but I will give it some consideration and engagement. As a lone maintainer I am initially a bit daunted with how to maintain, test and support this additional mode of operation. As a cross-platform library I would also want to be confident about the viability on Linux and Apple.

The additional wglew complexity is also a bit daunting from a maintainer perspective.

[Jupeyy]

This probably only works with glewContextInit, because there were code paths that used wglGetCurrentDC. Maybe this can be improved for easier usage

I load wglGetCurrentDC dynamically now too to minimize the ifdefs and allow glewInit Also I tested it on windows XP to make sure it doesn't break any legacy things, but it seems to work

[Jupeyy]

As a cross-platform library I would also want to be confident about the viability on Linux and Apple.

I'd say it is defenitely less useful, as atleast Linux uses mesa anyway and mesa supports MESA_LOADER_DRIVER_OVERRIDE environment flag to overload the GL driver(or atleast a vendor neutral dispatcher that can be forced to load smth different probably). But theoretically I'd say it could also work there if someone wants to write a custom opengl wrapper like mesa does it for windows