Cleanup testsuite with ASAN

[GunChleoc]

Since we now have easy ASAN compiling in the pipeline

https://code.launchpad.net/~widelands-dev/widelands/run-asan

I'm planning to go through the test suite until it passes.

This bug will be fixed with that as well:

https://bugs.launchpad.net/widelands/+bug/1734088

[GunChleoc (gunchleoc)]

All testsuite scenarios are producing the same memory leak:

=================================================================
==3752==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 288 byte(s) in 1 object(s) allocated from:
    #0 0x7f4c7d87f602 in malloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x98602)
    #1 0x7f4c7bfaf593 in glXGetFBConfigs (/usr/lib/x86_64-linux-gnu/mesa/libGL.so.1+0x1a593)

Direct leak of 144 byte(s) in 1 object(s) allocated from:
    #0 0x7f4c7d87f79a in __interceptor_calloc (/usr/lib/x86_64-linux-gnu/libasan.so.2+0x9879a)
    #1 0x7f4c7bd3fb38 (/usr/lib/x86_64-linux-gnu/libSDL2-2.0.so.0+0xbdb38)

SUMMARY: AddressSanitizer: 432 byte(s) leaked in 2 allocation(s).

[SirVer (sirver)]

This one seems to be either in libMesa or in SDL. Maybe it is also something we are doing wrong with cleaning up, hard to say.

[GunChleoc (gunchleoc)]

Yes, I think so too. I tried to find a way to exclude them from the LeakSanitizer for now, but all my attempts were ignored. I'll concentrate on the errors in our own code first and then try to find out where glXGetFBConfigs is called from.

[GunChleoc (gunchleoc)]

tl;dr: #1 is not our bug.

I have done some digging and came up with this:

https://stackoverflow.com/questions/10065849/memory-leak-using-glxcreatecontext

> Checking the docs, on glXChooseFBConfig (https://www.opengl.org/sdk/docs/man2/xhtml/glXChooseFBConfig.xml) the return value is expected to be freed with XFree called on it.

==============================

WIDELANDS:

We trigger this function once from Graphic::initialize and store the SDL_GLContext in gl_context_.

On shutdown, in Graphic::~Graphic(), we call:

    SDL_GL_DeleteContext(gl_context_);

So, we are doing everything right.

==============================

SDL2:

I found this in the SDL2 source code for it:

    void X11_GL_DeleteContext(_THIS, SDL_GLContext context)
    {
        Display *display = ((SDL_VideoData *) _this->driverdata)->display;
        GLXContext glx_context = (GLXContext) context;

        if (!_this->gl_data) {
            return;
        }
        _this->gl_data->glXDestroyContext(display, glx_context);
        X11_XSync(display, False);
    }

glXDestroyContext is documented here:

https://www.khronos.org/registry/OpenGL-Refpages/gl2.1/xhtml/glXDestroyContext.xml

No idea if that calls XFree or not - I guess we have to dig into libmesa next to see if we can file a bug with them.

[GunChleoc (gunchleoc)]

Could somebody with Valgrind powers check if WaresWorkersMap in the attached branch is indeed leaking?

I'm stumped right now, it's a unique_ptr of a map that contains only basic data types.

[GunChleoc (gunchleoc)]

P.S. the test I'm running for #5 is:

./widelands --language=en --scenario=test/maps/lua_testsuite.wmf --datadir_for_testing=.

[GunChleoc (gunchleoc)]

Migrated to https://github.com/widelands/widelands/issues/3232