WGPU: `wgpuAdapterRequestDevice` inexplicably segfaults if anything relating to a limit is created

[hotcocoaNcode]

M2 MacBook Air running Sequoia 15.0.1, I am honestly at a loss here.

For some inexplicable reason whenever I create a WGPURequiredLimits or WGPUSupportedLimits, the call to wgpuAdapterRequestDevice causes a segfault.

This is such that

//WGPURequiredLimits reqs{};
//reqs = getRequiredLimits();
deviceDesc.requiredLimits = nullptr;

works perfectly fine, while even

WGPURequiredLimits reqs{};
//reqs = getRequiredLimits();
deviceDesc.requiredLimits = nullptr;

or

WGPURequiredLimits reqs{};
reqs = getRequiredLimits();
deviceDesc.requiredLimits = nullptr;

will not succeed.

Since I am using the C API with no reference of the C++ file in my code, I am assuming no weird constructor is being called.
To attempt to circumvent this, I have tried the following horrible thing.

char reqs[sizeof(WGPURequiredLimits)];
*reinterpret_cast<WGPURequiredLimits*>(reqs) = getRequiredLimits();
deviceDesc.requiredLimits = nullptr;

Note that after calling getRequiredLimits(), reqs is never referenced anywhere else.

That is just not how this API works apparently. This causes the program to hang on wgpuAdapterRequestDevice, according to my breakpoints in CLion (on Mac, using LLDB).

I am unbelievably lost. I don't know why this is happening.

[hotcocoaNcode]

There is also a very good chance this is not a problem created by you, I am just putting this issue here since I am using your tutorial series and the cmake in this repository.

[eliemichel]

What does the definition of getRequiredLimits() looks like? Could you share a more complete code sample? That would help!

[hotcocoaNcode]

// directly from your tutorial series, though i can post implementation if you'd like
void setDefaultLimits(WGPULimits &limits) { /* ... */ }

WGPURequiredLimits getRequiredLimits() {
    WGPURequiredLimits requiredLimits{};
    setDefaultLimits(requiredLimits.limits);

    requiredLimits.limits.maxBindGroups = 7;

    return requiredLimits;
}

[hotcocoaNcode]

note: the same error does still occur even if getRequiredLimits() is fully commented out

[hotcocoaNcode]

Here's the full implementation of createDevice()

void createDevice() {
    WGPUDeviceDescriptor deviceDesc;
    deviceDesc.label = "John Graphics";
    deviceDesc.requiredFeatureCount = 0;
    deviceDesc.requiredFeatures = nullptr;
    /* insert anything about WGPURequiredLimits here, referenced in the descriptor or not and it breaks */
    deviceDesc.requiredLimits = nullptr;
    deviceDesc.defaultQueue.nextInChain = nullptr;
    deviceDesc.defaultQueue.label = "John Queue";
    deviceDesc.deviceLostCallback = [](WGPUDeviceLostReason reason, char const* message, void*) {
        std::cout << "WGPU: Device lost,  " << reason;
        if (message) std::cout << " (" << message << ")";
        std::cout << std::endl;
    };;

    bool areWeThereYet = false;

    wgpuAdapterRequestDevice(adapter, &deviceDesc, [](WGPURequestDeviceStatus status, WGPUDevice d, char const * message, void * b) {
        if (status == WGPURequestDeviceStatus_Success) {
            device = d;
        } else {
            std::cerr << "Could not get WebGPU device: " << message << std::endl;
        }
        *reinterpret_cast<bool*>(b) = true;
    }, &areWeThereYet);

#ifdef __EMSCRIPTEN__
    while (!awty) {
    emscripten_sleep(100);
}
#endif // __EMSCRIPTEN__

    assert(areWeThereYet);

    if (device == nullptr) {
        throw std::runtime_error("WGPU: Failed to get device!");
    }

    wgpuDeviceSetUncapturedErrorCallback(device, [](WGPUErrorType type, char const* message, void*) {
        std::cout << "Uncaptured device error: type " << type;
        if (message) std::cout << " (" << message << ")";
        std::cout << std::endl;
    }, nullptr);
}

[eliemichel]

This is all very mysterious. Does it work if you run the accompanying code of the guide as-is, e.g. step025? If it also fails, you may try step100-next, which uses a more recent version of WebGPU implementation, but is not the default yet (because the API has been moving a lot lately, prior to release freeze). If it runs correctly on the guide's code, could you share your whole source tree somehow, so I can try to reproduce your setup on my end?

[hotcocoaNcode]

Hi, sorry I took so long to respond. I am currently a high school student and got caught up in work. This is my implementation, however there are quite a few things I should note.

First off, compiling with Emscripten for the web actually succeeds, for some reason. What actually fails in that case is requesting more than 4 bind groups, which is not something I should be doing anyways and is something very horrible and bad and stupid that carried over from my Vulkan graphics backend that I wrote for the same engine. There is no reason I should actually be requiring any limits.

However, just creating a WGPURequiredLimits with WGPURequiredLimits limits; and never updating or using it anywhere works on Emscripten but locally causes a segfault.

I will also try compiling some things from the examples you listed later today.