glfwTerminate causes segfault, at XCloseDisplay on master

[chalecampb]

Hi guys,

Versions - Ubuntu 16.04LTS, with CMake 3.5.1, git 2.7.4, GCC 5.4, NVidia 384.98

I just came from the nanogui repo, I found that their 2 year old example repository (nanogui-test) works but when GLFW is updated to latest, it starts segfaulting on exit. Wanted to bring it to your attention.

The glfw submodule in nanogui-test is pegged at bde2fa0. Syncing to that commit resolves the issue.

I checked where the segfault occurred; the last break I set was at

/glfw/src/x11_init.c:783

Shortly after that the line XCloseDisplay(_glfw.x11.display); causes a segfault.

I saw that there are some issues entered which could cause a segfault if terminate were called twice, but you can inspect the code at nanogui-test, the application only calls glfwTerminate is called only once on exit (from nanogui::shutdown()). Also, in gdb there was no second call to terminate, it only ever failed at the first one.

I'll try to find specifically when the commits stop working. If you have any recommendations let me know.

Thanks!
-Alex

EDIT: Changed NVidia driver version to 384.98

[elmindreda]

Are you asking GLFW to create the context using EGL?

[chalecampb]

Thanks for your quick response.

All the code interacting with GLFW is in https://github.com/wjakob/nanogui/blob/master/src/screen.cpp

It states that the default is an OpenGL 3.3 context (line 121).

The example code does not pass the major and minor so the context should be 3.3.

In any case, EGL is not found anywhere in the calling code so I have to say, probably not, unless there is something tricky going on. I would have expected glfwGetEGLDisplay somewhere if it were using EGL. Admittedly I am not a GL buff, but I am happy to do any digging.

[elmindreda]

Thanks!

A call stack for the segfault would be invaluable.

If you're able, also try running the current glfwinfo and see if that also segfaults. It has switches that lets you create any kind of context and framebuffer (see glfwinfo -h).

The output of ldd $(which glxinfo) might also be of use.

[chalecampb]

Hi again,

Stack trace is

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff2283a90 in ?? ()
(gdb) bt
#0  0x00007ffff2283a90 in ?? ()
#1  0x00007ffff2029e2a in ?? () from /usr/lib/nvidia-384/libnvidia-glsi.so.384.98
#2  0x00007ffff2032f45 in ?? () from /usr/lib/nvidia-384/libnvidia-glsi.so.384.98
#3  0x00007ffff62df642 in XCloseDisplay () from /usr/lib/x86_64-linux-gnu/libX11.so.6
#4  0x00007ffff7b5a07e in _glfwPlatformTerminate () from /home/alex/Projects/Source/nanogui-test/ext/nanogui/libnanogui.so
#5  0x00007ffff7b53b78 in glfwTerminate () from /home/alex/Projects/Source/nanogui-test/ext/nanogui/libnanogui.so
#6  0x0000000000401a81 in main (argc=<optimized out>, argv=<optimized out>) at /home/alex/Projects/Source/nanogui-test/test.cpp:62

[chalecampb]

I was able to get a list of all glfw calls, just started the application, immediately closed, got a segfault, and processed the output.

https://gist.github.com/chalecampb/0e2b95ff6201779685e26dd5b912aacb

And based on those calls, internally something is routing to the GLFW calls to EGL. So that does actually sound like you guessed the issue correctly. I am not sure how that is happening, I will dig more. I am interested in understanding why that could cause the failure, even then.

[elmindreda]

And based on those calls, internally something is routing to the GLFW calls to EGL.

It seems from the gdb call stack like you're not using a debug build (-g -O0), so the trace tool you're using may be guessing wrong. I don't see any calls to _glfwChooseVisualEGL or _glfwCreateContextEGL which would be there if EGL was used for context creation.

[chalecampb]

Is this behavior expected under EGL?

…

On Tue, Nov 28, 2017, 8:18 AM Camilla Löwy ***@***.***> wrote: And based on those calls, internally something is routing to the GLFW calls to EGL. It seems from the gdb call stack like you're not using a debug build (-g -O0), so the trace tool you're using may be guessing wrong. I don't see any calls to _glfwChooseVisualEGL or _glfwCreateContextEGL which would be there if EGL was used for context creation. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1146 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AInNKuropCSAgL7Mt1Opr91aZiM11Bbqks5s7AgigaJpZM4QrIU5> .

[elmindreda]

There was an issue with recent Nvidia EGL that was fixed yesterday with 9e6c0c7.

[chalecampb]

Oh. I can check that tonight and see what happens. Will get back to you.

…

On Tue, Nov 28, 2017, 8:26 AM Camilla Löwy ***@***.***> wrote: There was an issue with recent Nvidia EGL that was fixed yesterday with 9e6c0c7 <9e6c0c7> . — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1146 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AInNKhz-VFqDHkgA4BXrd4vCTk52Vt9yks5s7AoZgaJpZM4QrIU5> .

[chalecampb]

Hi,

I wasn't able to checkout that reference, but I did see that it was a 1 line move in x11_init.c where the segfault was happening. So I just made the change and can confirm that it resolves the issue.

Thanks!
-Alex

[elmindreda]

Yay! Thank you for the follow-up!