GZD 4.4.2 and devbuild crash on Paradise MAP04 and MAP05

Sat Sep 05, 2020 12:58 pm

I've been having crashes on MAP04 and 05 from the GZDoom mapset "Paradise" (found here: https://www.doomworld.com/forum/topic/1 ... 3-4122020/). I have not seen these sorts of problems reported before, so I have to assume that there's something wrong with my own setup that is causing these issues. They are also not 100% consistently reproducible crashes - there are "problem areas" in these maps where they are more likely to happen, but they are not guaranteed to happen. Extremely frustrating when doing no-save run attempts.

I have tried these maps with GZDoom v4.4.2 and v4.3.3. They both exhibited crashes of the generic
Code:
*** Fatal Error ***
Address not mapped to object (signal 11)
Address: 0x50

variety, which don't really give any useful information unless I get a debug build and backtrace it.

However, I then tried it with the latest master of the GZDoom dev git, (g4.5pre-131-g91a4e6aed) and I got a slightly more specific error on crash:
Code:
gzdoom: gzdoom_build/gzdoom/src/common/objects/dobject.h:253: PClass* DObject::GetClass() const: Assertion `Class != nullptr' failed.
Aborted

I could go ahead and compile in debug mode and get the backtrace of the crash for you if needed. Just let me know. Wanted to get the initial report out of the way.

As usual, system info:
- OS: Debian Sid (GNU/Linux) amd64, kernel 5.7.0-3
- GPU: NVIDIA GTX 760
- CPU: i5 4440
- nvidia-driver: 440.100-2
- gzdoom: g4.5pre-131-g91a4e6aed
- renderer: Vulkan (it also happens on OpenGL)
- compatibility flags: Boom (strict) preset

Re: GZD 4.4.2 and devbuild crash on Paradise MAP04 and MAP05

Sun Sep 06, 2020 3:53 am

PhysixCat wrote:I could go ahead and compile in debug mode and get the backtrace of the crash for you if needed.

Yes, we need a backtrace, preferably with '+vm_jit 0' added to command line.