Example hello fails on MS Windows

[tigerwings]

Description
Run hello on Microsoft Windows 10

$ set WGPU_BACKEND=dx12
$ cargo run --example hello
    Finished dev [unoptimized + debuginfo] target(s) in 0.27s
     Running `target\debug\examples\hello.exe`
Error loading EGL entry points.
error: process didn't exit successfully: `target\debug\examples\hello.exe` (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)

while cargo run --example cube works.

Repro steps
After tracing down the issue, it seems this line is the root cause. And the stack trace is the following:

<unknown> 0x0000000000000000
[Inlined] khronos_egl::impl$98::eglQueryString(*mut khronos_egl::Dynamic<libloading::safe::Library,khronos_egl::EGL1_4>,*mut core::ffi::c_void,i32) lib.rs:2032
khronos_egl::Instance<khronos_egl::Dynamic<libloading::safe::Library,khronos_egl::EGL1_4> >::query_string<khronos_egl::Dynamic<libloading::safe::Library,khronos_egl::EGL1_4> >(enum2$<core::option::Option<khronos_egl::egl1_0::Display> >,i32) lib.rs:987
wgpu_hal::gles::egl::impl$14::init(*mut wgpu_hal::InstanceDescriptor) egl.rs:756
wgpu_core::instance::impl$0::new::init<wgpu_hal::gles::Api>(wgpu_hal::gles::Api,*mut wgpu_types::InstanceDescriptor) instance.rs:87
wgpu_core::instance::Instance::new(str,wgpu_types::InstanceDescriptor) instance.rs:104
wgpu_core::global::Global<wgpu_core::identity::IdentityManagerFactory>::new<wgpu_core::identity::IdentityManagerFactory>(str,wgpu_core::identity::IdentityManagerFactory,wgpu_types::InstanceDescriptor) global.rs:36
wgpu::backend::direct::impl$7::init(wgpu_types::InstanceDescriptor) direct.rs:570
wgpu::Instance::new(wgpu_types::InstanceDescriptor) lib.rs:1733
wgpu::impl$21::default() lib.rs:1720
hello::run::async_fn$0(core::pin::Pin<ref_mut$<enum2$<hello::run::async_fn_env$0> > >,core::future::ResumeTy) main.rs:5
core::future::from_generator::impl$1::poll<enum2$<hello::run::async_fn_env$0> >(core::pin::Pin<ref_mut$<core::future::from_generator::GenFuture<enum2$<hello::run::async_fn_env$0> > > >,*mut core::task::wake::Context) 0x00007ff62af72724
pollster::block_on<core::future::from_generator::GenFuture<enum2$<hello::run::async_fn_env$0> > >(core::future::from_generator::GenFuture<enum2$<hello::run::async_fn_env$0> >) lib.rs:128
hello::main() main.rs:26
core::ops::function::FnOnce::call_once<void (*)(),tuple$<> >(*mut ) 0x00007ff62af710cb
[Inlined] core::hint::black_box(tuple$<>) 0x00007ff62af72c3e
std::rt::lang_start::closure$0<tuple$<> >(*mut std::rt::lang_start::closure_env$0<tuple$<> >) 0x00007ff62af733d1
[Inlined] core::ops::function::impls::impl$2::call_once() 0x00007ff62bec86de
[Inlined] std::panicking::try::do_call() 0x00007ff62bec86d3
[Inlined] std::panicking::try() 0x00007ff62bec86d3
[Inlined] std::panic::catch_unwind() 0x00007ff62bec86d3
[Inlined] std::rt::lang_start_internal::closure$2() 0x00007ff62bec86d3
[Inlined] std::panicking::try::do_call() 0x00007ff62bec86d3
[Inlined] std::panicking::try() 0x00007ff62bec86d3
[Inlined] std::panic::catch_unwind() 0x00007ff62bec86d3
std::rt::lang_start_internal() 0x00007ff62bec86d3
std::rt::lang_start<tuple$<> >(*mut ,i64,*mut *mut u8,u8) 0x00007ff62af733aa
main 0x00007ff62af7260c
[Inlined] invoke_main() 0x00007ff62bef0eb8
__scrt_common_main_seh() 0x00007ff62bef0e96
<unknown> 0x00007ff8f8c87614
<unknown> 0x00007ff8fa8a26b1

Expected vs observed behavior
The example should run successfully on Windows.

Extra materials
Screenshots to help explain your problem.
Validation logs can be attached in case there are warnings and errors.
Zip-compressed API traces and GPU captures can also land here.

Platform
Information about your OS, version of wgpu, your tech stack, etc.
OS: Microsoft Windows 10
wgpu: trunk

[teoxoy]

The error comes from here: https://chromium.googlesource.com/angle/angle/+/4b94d7bb4fbed9018d198dfdc50c3aa5f9be3ddf/src/libEGL/libEGL.cpp#47.

I'm not sure why though and it doesn't seem there is a way to catch and deal with this error.
@cwfitzgerald any ideas?

[cwfitzgerald]

I think we've seen this before, if you get a weird version of angle in your path first you'll get this problem. I'm really not sure why it happens, but putting a normally built angle in path fixes it.

[teoxoy]

It's also odd that the cube example works but not the hello. @tigerwings do the other examples work?

I think we've seen this before, if you get a weird version of angle in your path first you'll get this problem. I'm really not sure why it happens, but putting a normally built angle in path fixes it.

It might have to do with the ANGLE_USE_EGL_LOADER define in angle. EnsureEGLLoaded is a noop if it's not defined. I feel like it should not be defined since as far as I understand we already load the dll ourselves?

[teoxoy]

Stepping a bit back:

The examples don't require any additional feature flags and the angle feature is not enabled by default.
Looking into this further, I see that we enable the gles feature on all platforms by default.

What we probably want to do is enable gles by default only on linux & android and map the angle feature to gles for windows & macOS.

[tigerwings]

@tigerwings do the other examples work?

under v0.16 and in those I tried:
cube, water, and halmark work only if the env var is set as WGPU_BACKEND=dx12 ; hello, hello-triangle and wgpu-info show the same error.

under v0.17: none of them work. the new error shows:

error: failed to run custom build command for `nv-flip-sys v0.1.1`

[cwfitzgerald]

Could you show the whole issue - likely you're missing a C compiler?

[tasogare3710]

I was able to reproduce it, this error occurs when libEGL is not found when using the GL backend.

The cause of this problem is that the environment variable WGPU_BACKEND is not respected, wgpu::Backends::all() is used in the following line, which initializes all backends.

https://github.com/gfx-rs/wgpu/blob/858d7d590ef9ebfe1b91f6f0925aacde15cce714/examples/hello/src/main.rs#L5C30-L5C30

[tasogare3710]

reproducible minimum source code.

set RUST_LOG=debug

async fn run() {
    let instance = wgpu::Instance::new(wgpu::InstanceDescriptor {
        backends: wgpu::Backends::GL,
        flags: wgpu::InstanceFlags::default(),
        dx12_shader_compiler: wgpu::Dx12Compiler::default(),
        gles_minor_version: wgpu::Gles3MinorVersion::default(),
    });
    instance
        .request_adapter(&wgpu::RequestAdapterOptions::default())
        .await
        .unwrap();
}

fn main() {
    env_logger::init();
    pollster::block_on(run());
}

important line is DEBUG wgpu_core::instance] Instance::new: failed to create Gl backend in the log, so do not forget to set RUST_LOG=debug.

[teoxoy]

@tigerwings do the other examples work?

under v0.16 and in those I tried: cube, water, and halmark work only if the env var is set as WGPU_BACKEND=dx12 ; hello, hello-triangle and wgpu-info show the same error.

The cause of this problem is that the environment variable WGPU_BACKEND is not respected, wgpu::Backends::all() is used in the following line, which initializes all backends.

https://github.com/gfx-rs/wgpu/blob/858d7d590ef9ebfe1b91f6f0925aacde15cce714/examples/hello/src/main.rs#L5C30-L5C30

This explains why some of the examples were working and some were not.

[teoxoy]

With #4185 merged, the GLES backend is now disabled by default on Windows.
So, the loader won't try to load some random libEGL on the system.

I think we can close this issue and open a new one to track moving the examples over to use the example framework (which takes into account env variables).

@cwfitzgerald thoughts on this?

[cwfitzgerald]

The hello examples all don't use the framework intentionally to make them easier to grok.

[teoxoy]

We are done here then 🎉

[TheYahton]

I have the same problem even after you merged #4185
As far as I understand, I don't have libegl and idk how to install it.

[cwfitzgerald]

What is the exact error you're getting?

[TheYahton]

What is the exact error you're getting?

error: process didn't exit successfully: `target\debug\wgpu-examples.exe hello` (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)

My gpu does not support DirectX 12 or Vulkan. To be sure that OpenGL was used instead of any other API, I tried replacing wgpu::Backends::all() with wgpu::Backends::GL in examples\src\hello\mod.rs and the error was the same.

[cwfitzgerald]

Could you post a stack trace on a new issue, using latest trunk?