Can't run valgrind. always segfaults in _glfwInitGLX

I run “make valgrind” and it loads up and promptly aborts in the open GL setup. What am I doing wrong? oh - running ubuntu 18 under VMWare. Rack runs fine by itself and in dbg.

==19464== Process terminating with default action of signal 11 (SIGSEGV)
==19464==  General Protection Fault
==19464==    at 0x13153162: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x13160505: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x131609B6: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x129EC9F2: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x12EFEE50: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x12DC7811: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x12DC28DA: ??? (in /usr/lib/x86_64-linux-gnu/dri/
==19464==    by 0x116C3AF6: ??? (in /usr/lib/x86_64-linux-gnu/
==19464==    by 0x116B00AD: ??? (in /usr/lib/x86_64-linux-gnu/
==19464==    by 0x116AB753: ??? (in /usr/lib/x86_64-linux-gnu/
==19464==    by 0x116ABC8C: ??? (in /usr/lib/x86_64-linux-gnu/
==19464==    by 0x367334: extensionSupportedGLX (glx_context.c:208)

maybe give “export LIBGL_ALWAYS_SOFTWARE=1” a try - this forces the mesa opengl library to use software rendering only - it will be very slow, but maybe its enough for debugging.

best wishes - hexdump

tx! I’ll try that.

Yes, that works. It is hideously slow, but may be useful. tx!

(no, too slow to be useful)

Address sanitizer gave much better luck, but you can’t run very far in rack before you start getting access after delete violations in Rack itself. Don’t know if they are spurious or not, but logged a github issue about it: