Asynchronously get ShaderModule creation result and compilation info

_release-content/migration-guides/async_shader_loading.md

@@ -0,0 +1,13 @@
+---
+title: "Asynchronous `ShaderModule` loading"
+pull_requests: [25352]
+---
+
+Because getting shader compilation info needs to be asynchronous on wasm and can't be blocking,
+`ShaderCache` no longer loads and stores `wgpu::ShaderModule` now, which is moved to inside `PipelineCache`.
+
+Changes:
+
+- `PipelineCache::block_on_render_pipeline` is not available on `wasm32` now.
+- `ShaderCache` no longer has `ShaderModule` and `RenderDevice` generic parameters.
+- `ShaderCache::get` is changed to return processed shader source code instead of `wgpu::ShaderModule`.

crates/bevy_core_pipeline/src/upscaling/mod.rs

@@ -47,6 +47,10 @@ fn clear_view_upscaling_pipelines(
 
 fn prepare_view_upscaling_pipelines(
     mut commands: Commands,
+    #[cfg_attr(
+        target_arch = "wasm32",
+        expect(unused_mut, reason = "`block_on_render_pipeline` is unused on wasm32")
+    )]
     mut pipeline_cache: ResMut<PipelineCache>,
     mut pipelines: ResMut<SpecializedRenderPipelines<BlitPipeline>>,
     blit_pipeline: Res<BlitPipeline>,
@@ -98,6 +102,7 @@ fn prepare_view_upscaling_pipelines(
 
             // Ensure the pipeline is loaded before continuing the frame to prevent frames without
             // any GPU work submitted
+            #[cfg(not(target_arch = "wasm32"))]
             pipeline_cache.block_on_render_pipeline(pipeline);
 
             commands

crates/bevy_render/Cargo.toml

@@ -85,6 +85,12 @@ image = { version = "0.25.2", default-features = false }
 # When the 'atomics' feature is enabled `fragile-send-sync-non-atomic` does nothing
 # and Bevy instead wraps `wgpu` types to verify they are not used off their origin thread.
 wgpu = { version = "30", default-features = false, features = [
+  # `std` is required: wgpu's error scopes are tracked per-thread via
+  # `std::thread::ThreadId`. Without this feature wgpu falls back to a no_std
+  # stub where all threads share a single (global) error-scope stack, which
+  # panics with "Mismatched pop_error_scope call" when pipelines are compiled
+  # concurrently on multiple threads (async pipeline compilation).
+  "std",
   "wgsl",
   "dx12",
   "metal",
@@ -116,6 +122,7 @@ bitflags = "2"
 itertools = "0.14"
 weak-table = "0.3"
 tracing = "0.1"
+async-lock = { version = "3.0", default-features = false }
 
 [dev-dependencies]
 proptest = "1"