We are currently migrating Bugzilla to GitHub issues.
Any changes made to the bug tracker now will be lost, so please do not post new bugs or make changes to them.
When we're done, all bug URLs will redirect to their equivalent location on the new bug tracker.

You can run valgrind with --leak-check=full and --show-leak-kinds=all to get a real flood of data about where leaks happened.

Doing this on your test program shows that most of these happen in:

- dlopen()  (apparently glibc leaks some memory?)
- XOpenIM (Xlib)
- Some stuff from Nvidia's OpenGL driver
- D-Bus functions

Note that most of this is "still reachable" and not "definitely lost"...which is to say that there's still (what could be) a valid pointer to the allocated block somewhere in memory. In some cases (probably dlopen is doing this), the code is intentionally keeping a cache of some sort around, with the intention of it living until the process terminates. This isn't really a "leak" in that it's intentional behavior.

The D-Bus ones _might_ be something we're doing wrong, but I'd have to research more. At any rate, most of these are either intentionally not freed, or at least out of our control in external libraries.

The only one that might be ours is this...

==18003== 88 bytes in 1 blocks are still reachable in loss record 52 of 95
==18003==    at 0x4C2AB80: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==18003==    by 0x4C2CF1F: realloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==18003==    by 0x4EEFE68: SDL_realloc_REAL (SDL_malloc.c:41)
==18003==    by 0x4EF1008: SDL_TLSSet_REAL (SDL_thread.c:66)
==18003==    by 0x4EF1426: SDL_GetErrBuf (SDL_thread.c:241)
==18003==    by 0x4E4C321: SDL_ClearError_REAL (SDL_error.c:216)
==18003==    by 0x4E4AD72: SDL_InitSubSystem_REAL (SDL.c:116)
==18003==    by 0x4E4AF3A: SDL_Init_REAL (SDL.c:244)
==18003==    by 0x4E90234: SDL_Init_DEFAULT (SDL_dynapi_procs.h:89)
==18003==    by 0x4E95D3D: SDL_Init (SDL_dynapi_procs.h:89)
==18003==    by 0x400715: main (test.c:528)

...but that might be that the TLS slot array is meant to live past SDL_Quit().

--ryan.

(In reply to Ryan C. Gordon from comment #1)
> The D-Bus ones _might_ be something we're doing wrong, but I'd have to
> research more. At any rate, most of these are either intentionally not
> freed, or at least out of our control in external libraries.

I'm taking a look at the SDL D-Bus code now, since I wrote a lot of it, but I can't see any obvious places that memory is leaking; the messages are all being unref'd correctly as far as I can tell.

I'll keep investigating but the problem may well be internal to the D-Bus library.

(In reply to Alex Baines from comment #2)
> (In reply to Ryan C. Gordon from comment #1)
> > The D-Bus ones _might_ be something we're doing wrong, but I'd have to
> > research more. At any rate, most of these are either intentionally not
> > freed, or at least out of our control in external libraries.
> 
> I'm taking a look at the SDL D-Bus code now, since I wrote a lot of it, but
> I can't see any obvious places that memory is leaking; the messages are all
> being unref'd correctly as far as I can tell.
> 
> I'll keep investigating but the problem may well be internal to the D-Bus
> library.

As a follow up, the following trivial program has the same leaks reported by valgrind, so the problems are almost certainly internal to DBus.

#include <dlfcn.h>
#include <dbus/dbus.h>

DBusConnection *session_conn;
DBusConnection *(*bus_get_private)(DBusBusType, DBusError *);
void (*connection_close)(DBusConnection *);
void (*connection_unref)(DBusConnection *);

int main(void){
	void* handle = dlopen("libdbus-1.so.3", RTLD_NOW);

	bus_get_private = dlsym(handle, "dbus_bus_get_private");
	connection_close = dlsym(handle, "dbus_connection_close");
	connection_unref = dlsym(handle, "dbus_connection_unref");

	session_conn = bus_get_private(DBUS_BUS_SESSION, NULL);

	connection_close(session_conn);
	connection_unref(session_conn);

	dlclose(handle);

	return 0;
}

Created attachment 1738 [details]
Fix DBus related memory leak.

Turns out I was wrong and there's a dbus_shutdown function that I had no idea about until now. 

Calling it fixes the leaks, I have attached a patch.

Great news !
I'll try this right away =D

The patch is in, thanks!
https://hg.libsdl.org/SDL/rev/84ae33058c67