"***Fatal error***" with legacy branch

[vanfanel]

Hi there,

I am building drfrag's legacy branch on my X86_64 desktop GNU/Linux system against latest SDL2 running on Xorg+OpenGL/GLES, and I get this on exit, and the GZdoom process gets stuck there, it never gets to quit unless I manually kill it:

Code: Select all

*** Fatal Error ***

The game runs fine, this error is happening on quit only.

I have noticed that it only happens with SDL2 when it's built without desktop OpenGL support, wich forces SDL2 to use the "opengles2" or "opengles" renderers (note I am talking about the SDL2 renderer, not the GZDoom one).
Since GZDoom should not care about what renderer is SDL2 using, my guess is that it's trying to free something on quit that shouldn't be freed in GLES but only on desktop OpenGL.

Strangely enough, forcing a GLES renderer on an SDL2 built with desktop OpenGL support using this:

Code: Select all

SDL_HINT_RENDER_DRIVER="opengles2" ./gzdoom

does not cause the **Fatal Error**.

Building SDL2 without OpenGL support is as easy as configuring with:

Code: Select all

./configure --enable-video-opengles --disable-video-opengl

Why is this bug important to me? Because it happens on both SDL2 running on generic X86_64 Xorg, but also on SDL2 running on X-less enviroments (X86, ARM), where GLES is the only option available.

[_mental_]

Run GZDoom in gdb and when the process hangs, do Ctrl+C and bt in the debugger. This may give some clues what's going on.

[vanfanel]

@_mental_: Backtrace is different each time... I think whatever is wrong, its causing a memory corruption that has different consequences each time:

Execution 1:

Code: Select all

(gdb) bt
#0  0x00007ffff6159804 in _int_free (av=0x7ffff64a9c20 <main_arena>, p=<optimized out>, have_lock=0) at malloc.c:4328
#1  0x00007ffff615e44e in __GI___libc_free (mem=<optimized out>) at malloc.c:3145
#2  0x00005555559d209c in PClass::StaticShutdown() ()
#3  0x0000555555749256 in call_terms() ()
#4  0x00007ffff610aec0 in __run_exit_handlers (status=0, listp=0x7ffff64a96f8 <__exit_funcs>, 
    run_list_atexit=run_list_atexit@entry=true, run_dtors=run_dtors@entry=true) at exit.c:83
#5  0x00007ffff610af1a in __GI_exit (status=<optimized out>) at exit.c:105
#6  0x000055555574c910 in ST_Endoom() ()
#7  0x0000555555bef680 in ?? ()
#8  0x0000555555bef516 in ?? ()
#9  0x0000555555cb0f2f in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#10 0x0000555555cb2317 in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#11 0x0000555555cbe941 in VMCall(VMFunction*, VMValue*, int, VMReturn*, int) ()
#12 0x0000555555be23da in DMenu::CallMenuEvent(int, bool) ()
#13 0x0000555555be4141 in M_Responder(event_t*) ()
#14 0x00005555559b62e8 in D_ProcessEvents() ()
#15 0x00005555559beaaa in NetUpdate() ()
#16 0x00005555559bf906 in TryRunTics() ()
#17 0x00005555559b8cb9 in D_DoomLoop() ()
#18 0x00005555559ba737 in D_DoomMain() ()
#19 0x000055555572d6ef in main ()

Execution 2:

Code: Select all

(gdb) bt
#0  0x00007ffff6158086 in malloc_consolidate (av=av@entry=0x7ffff64a9c20 <main_arena>) at malloc.c:4506
#1  0x00007ffff6159e08 in _int_free (av=0x7ffff64a9c20 <main_arena>, p=<optimized out>, have_lock=0) at malloc.c:4398
#2  0x00007ffff615e44e in __GI___libc_free (mem=<optimized out>) at malloc.c:3145
#3  0x0000555555c5d6e5 in PSymbolTable::ReleaseSymbols() ()
#4  0x0000555555c5d769 in PSymbolTable::~PSymbolTable() ()
#5  0x0000555555c63e5a in FTypeTable::Clear() ()
#6  0x00005555559d21b9 in PClass::StaticShutdown() ()
#7  0x0000555555749256 in call_terms() ()
#8  0x00007ffff610aec0 in __run_exit_handlers (status=0, listp=0x7ffff64a96f8 <__exit_funcs>, 
    run_list_atexit=run_list_atexit@entry=true, run_dtors=run_dtors@entry=true) at exit.c:83
#9  0x00007ffff610af1a in __GI_exit (status=<optimized out>) at exit.c:105
#10 0x000055555574c910 in ST_Endoom() ()
#11 0x0000555555bef680 in ?? ()
#12 0x0000555555bef516 in ?? ()
#13 0x0000555555cb0f2f in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#14 0x0000555555cb2317 in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#15 0x0000555555cbe941 in VMCall(VMFunction*, VMValue*, int, VMReturn*, int) ()
#16 0x0000555555be23da in DMenu::CallMenuEvent(int, bool) ()
#17 0x0000555555be4141 in M_Responder(event_t*) ()
#18 0x00005555559b62e8 in D_ProcessEvents() ()
#19 0x00005555559beaaa in NetUpdate() ()
#20 0x00005555559bf906 in TryRunTics() ()
#21 0x00005555559b8cb9 in D_DoomLoop() ()
#22 0x00005555559ba737 in D_DoomMain() ()
#23 0x000055555572d6ef in main ()
(gdb) 

Execution 3:

Code: Select all

(gdb) bt
#0  0x00007ffff6159804 in _int_free (av=0x7ffff64a9c20 <main_arena>, p=<optimized out>, have_lock=0) at malloc.c:4328
#1  0x00007ffff615e44e in __GI___libc_free (mem=<optimized out>) at malloc.c:3145
#2  0x00005555559d209c in PClass::StaticShutdown() ()
#3  0x0000555555749256 in call_terms() ()
#4  0x00007ffff610aec0 in __run_exit_handlers (status=0, listp=0x7ffff64a96f8 <__exit_funcs>, 
    run_list_atexit=run_list_atexit@entry=true, run_dtors=run_dtors@entry=true) at exit.c:83
#5  0x00007ffff610af1a in __GI_exit (status=<optimized out>) at exit.c:105
#6  0x000055555574c910 in ST_Endoom() ()
#7  0x0000555555bef680 in ?? ()
#8  0x0000555555bef516 in ?? ()
#9  0x0000555555cb0f2f in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#10 0x0000555555cb2317 in VMExec_Unchecked::Exec(VMFrameStack*, VMOP const*, VMReturn*, int) ()
#11 0x0000555555cbe941 in VMCall(VMFunction*, VMValue*, int, VMReturn*, int) ()
#12 0x0000555555be23da in DMenu::CallMenuEvent(int, bool) ()
#13 0x0000555555be4141 in M_Responder(event_t*) ()
#14 0x00005555559b62e8 in D_ProcessEvents() ()
#15 0x00005555559beaaa in NetUpdate() ()
#16 0x00005555559bf906 in TryRunTics() ()
#17 0x00005555559b8cb9 in D_DoomLoop() ()
#18 0x00005555559ba737 in D_DoomMain() ()
#19 0x000055555572d6ef in main ()

[drfrag]

But you already got a crash on exit with 3.3.0:
viewtopic.php?f=2&t=60020#p1047163
Have you tried with 3.3.2? Do you get the crash with vid_renderer 0 in the legacy branch?

[vanfanel]

@drfrag: That was on the Pi3. I am now getting it on the X86_64 laptop, GNU/Linux system running X11, and SDL2 running on X11, not KMSDRM or BCRM. In other words, I am now on "standard" Linux/X11 stuff, and I have found out that the cause for this ***Fatal Error*** seems to be using SDL2 built without OpenGL support.
Can you please build SDL2 without OpenGL and test?

Code: Select all

./configure --enable-video-opengles --disable-video-opengl

Also, I did clone your repository like this:

Code: Select all

git clone --depth 1 -b g3.3mgw https://github.com/drfrag666/gzdoom.git

So I am testing 3.4.1 version of the legacy branch, not 3.2.0 anymore.

[drfrag]

No sorry. AFAIK there were a couple of SDL related commits since 3.3.0: 'Merged list of video modes for Cocoa and SDL backends' and 'skip rendering when application is not active'. None of the two seem related to your issue.

[vanfanel]

No sorry. AFAIK there were a couple of SDL related commits since 3.3.0: 'Merged list of video modes for Cocoa and SDL backends' and 'skip rendering when application is not active'. None of the two seem related to your issue.

I don't quite understand your answer: the "No sorry" part. Does that mean I have to clone a different branch? Will you do the test and try to replicate the issue? (It's specially funny because I believe we are both from Spain)

[drfrag]

You believe? So you're not sure if you're from Spain...
I don't have my new linux computer up and running yet so i haven't even built a linux version myself and about building SDL2...

[vanfanel]

You believe? So you're not sure if you're from Spain...
I don't have my new linux computer up and running yet so i haven't even built a linux version myself and about building SDL2...

Yes, I am sure I'm from Spain (living in a small village in Cáceres now but my english is a bit strange sometimes).
The building process on X86 GNU/Linux is very easy: I can help you with step-by-step instructions if you want. I will assume you're using some debian-derivative such as Ubuntu/Lubuntu... or Debian itself.

-Install libopenal, which is used by GZDoom for audio:

Code: Select all

sudo apt-get install libopenal-dev

-Remove your distro's SDL2 installation. Since most distros have SDL2 as a dependency for desktop enviroments, you can remove it manually and then reinstall it after you finish the experiments:

Code: Select all

sudo rm /usr/lib/x86_64-linux-gnu/libSDL2*

-Get the latest stable SDL2 sources, uncompress them, build them and install them:

Code: Select all

wget https://www.libsdl.org/release/SDL2-2.0.8.tar.gz
tax xvzpf SDL2-2.0.8.tar.gz
cd SDL2-2.0.8
 ./configure --prefix=/usr --enable-video-x11 --disable-video-mir --disable-diskaudio --disable-oss --disable-pulseaudio --disable-dummyaudio --disable-video-dummy --enable-video-opengles --disable-video-opengl
make -j6
sudo make install

Now let's go for your GZDoom branch:

-Clone your sources:

Code: Select all

git clone --depth 1 -b g3.3mgw https://github.com/drfrag666/gzdoom.git

-cd into the clone repo, create a building dir, and configure using cmake:

Code: Select all

cd gzdoom
mkdir b3
cd b3
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j6

...And that's pretty much it!

[drfrag]

So you're the guy from the ZX-One forum? That dondiego guy from Granada is in TestingZone? (ZDP) and we both could be the same person (or not). In any case as i said before i still don't have a working linux computer right now.

[vanfanel]

@drfrag: Yes! I am Vanfanel on the great ZX-UNO forum, and an inconditional FPGA retro-computing lover. I don't know who dondiego is, I am not in Zona de Pruebas. Totally offtopic: do you have a ZX-UNO too?
And about the thread... do you have any idea on what could be happening? Does GZDoom dynamically load/unload GLES/OpenGL libs somewhere? I can't find such a thing, but the problem seems to be related to that.

[drfrag]

Yeah ZX-Uno is great but i'm a very poor guy so i couldn't afford one, i can live with the ZesarUX emulator tough.
No idea but you should really try 3.3.2 to find if the problem is related to a later change (unlikely).

[vanfanel]

@drfrag: No luck.... On every version I try, if I use SDL2 with the "opengles2" or "opengles" videodriver and no opengl support built-in (--disable-video-opengl) I get the same fatal error. GZDoom is assuming something about opengl on SDL2 that shouldn't be assumed...

[drfrag]

Sorry but i'm afraid i cannot be of any help here, it's not a recent change and it's not something specific to my legacy branch. It seems OpenGL is still required even if you use GLES. But does it work now on the Pi? Should be the same as 3.3.x.

[vanfanel]

Sorry but i'm afraid i cannot be of any help here, it's not a recent change and it's not something specific to my legacy branch. It seems OpenGL is still required even if you use GLES. But does it work now on the Pi? Should be the same as 3.3.x.

The game DOES work: the **Fatal Error** happens on quit only!