Implement OpenGL Backend For Unix Platforms

[zicklag]

Connections
Requires: gfx-rs/gfx#3340 ( merged )
Should be merged first: #910
Works towards: #450

Description
Integrates the GFX GL backend for Unix platforms

Testing
Runs the basic cube and triangle examples for wgpu-rs, but not without rendering artifacts on the cube.

[monocodus]

This is an autogenerated code review.New suggestions: 8

[zicklag]

OK, I haven't nailed down the dependency versions yet, but the rest of the code should be ready for review.

The most controversial organizational change that I've included in this ( sorry, maybe should have been a separate PR ) was I used the cfg_aliases crate to replace the backends! macro throughout the code. I think it is much more readable to use plain Rust cfg checks instead of a macro in all of the cases that we previously used backends! and now you can use the same cfg syntax anywhere that the backends! macro could not handle such as in struct field definitions.

[kvark]

Note that wgpu always works with released gfx-rs versions. So we can't merge this until we get something usable in gfx-rs 0.6 branch and publish the GL backend.

[kvark]

I agree the configuration change should land independently prior to GL stuff

[zicklag]

I agree the configuration change should land independently prior to GL stuff

OK, I'll extract it out to a separate PR that should be merged before this one.

[zicklag]

Note that wgpu always works with released gfx-rs versions. So we can't merge this until we get something usable in gfx-rs 0.6 branch and publish the GL backend.

Yeah, I figured. I'll put the GFX git commit hashes in there to fix the deps just so we can verify that CI will pass.

[zicklag]

OK, the two commits have been separated. Once #910 is merged this PR will only have the second commit in it.

After separating the commits you can see that this was actually really simple from the WGPU side thanks to the abstraction provided by GFX, which is really awesome.

[zicklag]

Just found that the port to the newer version of GFX broke the Vulkan backend on Linux at least. I'm working on fixing that. It's some panic on the unwrap for get_surface_mut.

Edit: Figured it out: I had left out the x11 feature on the Vulkan backend.

[zicklag]

Couple of implementation questions.

[kvark]

gfx-rs GL backend is now published. Time to revive this!

[zicklag]

Yay! I'll see about getting this back up to date tomorrow probably. 😃

[zicklag]

There you go! It looks like surfman is failing to compile on iOS, unfortunately. Not sure why we hadn't hit that until now.

Also the cube example doesn't run:

thread 'main' panicked at 'called `Option::unwrap()` on a `None` value', /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/gfx-backend-gl-0.6.0/src/command.rs:1174:30
stack backtrace:
   0: rust_begin_unwind
             at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/std/src/panicking.rs:483:5
   1: core::panicking::panic_fmt
             at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/core/src/panicking.rs:85:14
   2: core::panicking::panic
             at /rustc/ffa2e7ae8fbf9badc035740db949b9dae271c29f/library/core/src/panicking.rs:50:5
   3: core::option::Option<T>::unwrap
             at /home/zicklag/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/option.rs:386:21
   4: <gfx_backend_gl::command::CommandBuffer as gfx_hal::command::CommandBuffer<gfx_backend_gl::Backend>>::bind_graphics_descriptor_sets
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/gfx-backend-gl-0.6.0/src/command.rs:1172:40
   5: wgpu_core::command::render::<impl wgpu_core::hub::Global<G>>::command_encoder_run_render_pass_impl
             at /home/zicklag/git/zicklag/wgpu-workspace/wgpu/wgpu-core/src/command/render.rs:1010:29
   6: wgpu_core::command::render::<impl wgpu_core::hub::Global<G>>::command_encoder_run_render_pass
             at /home/zicklag/git/zicklag/wgpu-workspace/wgpu/wgpu-core/src/command/render.rs:414:9
   7: <wgpu::backend::direct::Context as wgpu::Context>::command_encoder_end_render_pass
             at ./src/backend/direct.rs:1417:9
   8: <wgpu::RenderPass as core::ops::drop::Drop>::drop
             at ./src/lib.rs:2340:13
   9: core::ptr::drop_in_place
             at /home/zicklag/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:175:1
  10: <cube::Example as cube::framework::Example>::render
             at ./examples/cube/main.rs:412:9
  11: cube::framework::start::{{closure}}
             at ./examples/cube/../framework.rs:297:17
  12: winit::platform_impl::platform::sticky_exit_callback
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/winit-0.22.2/src/platform_impl/linux/mod.rs:698:5
  13: winit::platform_impl::platform::x11::EventLoop<T>::run_return
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/winit-0.22.2/src/platform_impl/linux/x11/mod.rs:312:21
  14: winit::platform_impl::platform::x11::EventLoop<T>::run
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/winit-0.22.2/src/platform_impl/linux/x11/mod.rs:390:9
  15: winit::platform_impl::platform::EventLoop<T>::run
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/winit-0.22.2/src/platform_impl/linux/mod.rs:645:35
  16: winit::event_loop::EventLoop<T>::run
             at /home/zicklag/.cargo/registry/src/github.com-1ecc6299db9ec823/winit-0.22.2/src/event_loop.rs:149:9
  17: cube::framework::start
             at ./examples/cube/../framework.rs:230:5
  18: cube::framework::run
             at ./examples/cube/../framework.rs:307:5
  19: cube::main
             at ./examples/cube/main.rs:419:5
  20: core::ops::function::FnOnce::call_once
             at /home/zicklag/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/function.rs:227:5
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
[0.222763 ERROR]()(no module): Heaps not completely freed before drop. Utilization: MemoryHeapUtilization { utilization: MemoryUtilization { used: 917504, effective: 263424 }, size: 18446744073709551615 }
[0.222821 ERROR]()(no module): Heaps not completely freed before drop. Utilization: MemoryHeapUtilization { utilization: MemoryUtilization { used: 16777216, effective: 262996 }, size: 18446744073709551615 }
[0.222881 ERROR]()(no module): Not all allocations from LinearAllocator were freed
[0.222901 ERROR]()(no module): Memory leak: SizeEntry(768) is still used
[0.222924 ERROR]()(no module): Memory leak: SizeEntry(6144) is still used
[0.222942 ERROR]()(no module): Memory leak: SizeEntry(49152) is still used
[0.222959 ERROR]()(no module): Memory leak: SizeEntry(256) is still used
[0.222992 ERROR]()(no module): Memory leak: SizeEntry(262144) is still used
[0.223013 ERROR]()(no module): Memory leak: SizeEntry(65536) is still used
[0.223039 ERROR]()(no module): DescriptorAllocator is dropped

Is this potentially related to your recent image changes @kvark ? The triangle example works, though.

Otherwise the changes should pretty much be in order I think.

[kvark]

Thank you for the amazing work!
A few small things here

[kvark]

Oh, for iOS - we shouldn't be building OpenGL for iOS at all. So if it falls under unix, we should change build.rs to make it so it's unix but not iOS.

[zicklag]

Oh, for iOS - we shouldn't be building OpenGL for iOS at all. So if it falls under unix, we should change build.rs to make it so it's unix but not iOS.

Oh, yeah, of course. Easy fix.

[zicklag]

OK, now I'm a little confused why surfman is being pulled in at all. The only thing that has a dependency on surfman should be gfx-backend-gl and is already excluding the surfman dep on iOS:

[target.'cfg(all(unix, not(target_os = "ios")))'.dependencies]
surfman = { version = "0.3", features = ["sm-raw-window-handle"], optional = true }

Could clippy be being weird about this and including the dependency when it isn't supposed to?

[kvark]

@zicklag you are looking at the master branch for gfx but using the 0.6 released version, which is a bit different:

[target.'cfg(not(target_arch = "wasm32"))'.dependencies]
surfman = { version = "0.3", features = ["sm-raw-window-handle"] }

Really, we shouldn't even try to build the GL backend on iOS, so this line needs to be changed:

[target.'cfg(unix)'.dependencies]
gfx-backend-gl = { version = "0.6" }

[zicklag]

Oh, yes, that's right. Thank you. I don't know how I missed that. 😆 Fixed it ( assuming CI passes, now ).

[kvark]

bors r+

[kvark]

Thank you for all the amazing work!

[bors]

Build failed:

• Ubuntu Nightly

[zicklag]

Glad I could help!

Unfortunately it looks like we've got to wait for Canonical to fix their servers or something. The apt repos are failing on CI everywhere at the moment.

[kvark]

bors retry

[bors]

Build succeeded:

• Android Stable
• iOS Stable
• MacOS Nightly
• MacOS Stable
• Ubuntu Nightly
• Ubuntu Stable
• Windows Nightly
• Windows Stable