__attribute__((destructor)) is a bad place for cleanup, so xr_3da fails assert on termination

[kosumosu]

I faced several failed assertions in mimalloc when I was trying to close the game.

I started investigating this and found out that the problem was not in the allocator itself, but in the fact that the allocator was destroyed BEFORE the failed assertion. Further investigation has shown that even though SDL_UnloadObject() is called before returning from main(), the __attribute__((destructor)) unload() is called AFTER return from main(). So global variables inside xrGame.so turn into a garbage by the time the control reaches unload().
This causes xray to crash every time it terminates on e2k. It also leads to hanging fullscreen, causing big trouble to a user.

Although a dlclose() operation is not required to remove structures from an address space, neither is an implementation prohibited from doing so.
https://pubs.opengroup.org/onlinepubs/007904975/functions/dlclose.html

So resources from shared libraries must be cleaned up explicitly instead.

[kosumosu]

This is mainly Linux-related. And in theory, not only on Elbrus.

[Xottab-DUTY]

@kosumosu, I just pushed 810692a to fix this problem. Please, test it. :)

Also, can you test it with disabled workaround for GCC in FACTORY_PTR_INSTANCIATE in src/Include/xrRender/FactoryPtr.h?

[kosumosu]

I confirm, the game doesn't crash anymore 👍

[Xottab-DUTY]

@kosumosu doesn't crash even with removed FACTORY_PTR_INSTANCIATE workaround?

[kosumosu]

I'm trying this right now...

[kosumosu]

@Xottab-DUTY Without it I got SIGSEGV when tried to "exit to windows" from menu right away.

1 CUIStatsIcon::TEX_INFO::~TEX_INFO()   0x4555b0442c58 
2 __cxa_vec_dtor                        0x455558421338 
3 __run_exit_handlers                   0x455558957d68 
4 exit                                  0x4555589587f0 
5 __libc_start_main                     0x4555588f5048 
6 _start                                0x1c1e0        

[Xottab-DUTY]

@kosumosu, hi! Could you please try to remove FACTORY_PTR_INSTANCIATE workaround in src/Include/xrRender/FactoryPtr.h again and test it?
In a914f65 I've changed how CUIStatsIcon::TEX_INFO works, so it might help with that.

[kosumosu]

That have been a long time. I'm not sure I still have access to e2k and have proper setup...

[Xottab-DUTY]

@kosumosu, this should be reproducible on any Linux machine I think.