-
-
Notifications
You must be signed in to change notification settings - Fork 35.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
WebGPURenderer: Fix global references in Node.js #29919
Conversation
📦 Bundle sizeFull ESM build, minified and gzipped.
🌳 Bundle size after tree-shakingMinimal build including a renderer, camera, empty scene, and dependencies.
|
react
react
Note – I've renamed the issue to "fix global references in Node.js". I think that's the issue here, not React. Next.js is a React-based framework that happens to execute code in both a browser environment and server-side in Node.js. |
@@ -64,7 +64,7 @@ class WebGPUBackend extends Backend { | |||
powerPreference: parameters.powerPreference | |||
}; | |||
|
|||
const adapter = await navigator.gpu.requestAdapter( adapterOptions ); | |||
const adapter = ( typeof navigator !== 'undefined' ) ? await navigator.gpu.requestAdapter( adapterOptions ) : {}; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@donmccurdy{}
is intended as a workaround for node but I'm not sure {}
is ideal here. Do you think this use case is handled appropriately or would you implement it in a different way?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Base on that function code, it should be null, otherwise, next step is exceptional?
"adapter.features"
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@whatisor are you executing this code in the Node.js environment and seeing runtime errors, or is this a compile-time error (like from TypeScript or Webpack)? Not clear to me from #29916.
I'm not sure that backend.init( renderer )
should be making efforts to avoid runtime exceptions, if running in an environment where neither WebGL nor WebGPU exists. I would support fixes that avoid build-time errors, and run-time errors in module scope. But if you are initializing and using a renderer in Node.js, a runtime exception is to be expected.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not sure that backend.init( renderer ) should be making efforts to avoid runtime exceptions, if running in an environment where neither WebGL nor WebGPU exists
I agree, I think we should be throwing an error here, it doesn't make much sense to me to proceed with an empty object as an adapter.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@donmccurdy it is compile error.
We also need to fix three\examples\jsm\capabilities\WebGPU.js with global and await.
So, we should return null and (exception is next step as current)?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
three\examples\jsm\capabilities\WebGPU.js
Capabilities should be optional, it should be possible to use WebGPURenderer
without global await. #29218
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can we do same thing with webgl to avoid await?
https://developer.mozilla.org/en-US/docs/Web/API/GPUCanvasContext
=> No, we still need to await for real adapter. I guess browser need to improve this part.
Use `null` as fallback when navigator isn't available.
@@ -113,13 +113,21 @@ class WebGPUUtils { | |||
// TODO: Remove this check when Quest 34.5 is out | |||
// https://github.com/mrdoob/three.js/pull/29221/files#r1731833949 | |||
|
|||
if ( navigator.userAgent.includes( 'Quest' ) ) { | |||
if ( typeof navigator !== 'undefined' ) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could we remove this? It is inside a function that will never be executed by the adapter return null in this case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I guess the issue is that it's a compiler error, not a runtime error? I wish we knew what in the Next.js stack is doing compile-time checks on dependencies, and what its requirements are. Avoiding all references to browser globals is a pretty tall order for a large library relying heavily on Web APIs. :/
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The compiler throwing an error for a global variable inside a function seems strange to say the least, since a variable can be declared at the global level after the function declaration.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, I'm not sure why... trying to reproduce the issue:
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think it is this one:
https://react.dev/reference/react/StrictMode
I have set up a test suite of various build environments, like Vite, with and without React, Next.js, R3F, and a specific one to check this PR: https://github.com/verekia/three-gpu-ecosystem-tests The PR does indeed fix Next.js crashing on reference errors 🙌 That being said, I am getting a Regular import (fixed by this PR): Dynamic import workaround: |
It seems the only blocking bit is the change in @verekia Can you please verify if build works without this change? If so, we can revert it ( see #29919 (comment)). |
Yes, it works without the |
Revert.
Thanks for letting us know! I have reverted the changes in |
|
@whatisor Could you provide a repro where My test is pretty basic, the only thing I import is: import * as THREE from 'three'
import { WebGPURenderer } from 'three/webgpu' |
This PR has nothing to do with React or its strict mode etc. We need a reproduction to know what's going on here or if this is something we can actually support. Next is notorious for overzealous compilation the ecosystem itself can't mitigate. |
I did not mean issue of Core. It is just simply call Webgpu.isAvailable() in your test to see compilation issue, but I will push PR to your test case |
@whatisor You are not supposed to call any method that's meant for browsers during SSR. These should live in a WebGPU.isAvailable() // ❌ Don't do that, it runs on the server during SSR
function MyComponent() {
WebGPU.isAvailable() // ❌ Don't do that, it runs on the server during SSR
useEffect(() => {
WebGPU.isAvailable() // ✅ No problem, runs only in the browser
}, [])
return // ...
} That's why the only concern for Three.js should be that importing in a Next.js project doesn't cause a crash due to global browser variable access and top-level browser API calls. If you call methods during SSR that should never run on the server, then it's expected for SSR to break. |
I added more test cases to my repo, including Next 14/15, React 18/19, Pages/App router, and a TSL call for all cases. Now that this PR is merged, things are looking really good for Next.js support in r171. Thank you.
The recent release of Next.js 15 + React 19 RC is expected to break things on the R3F side, so it's not a problem Three.js should worry about. |
I added into your 2 test cases and it is failed from BUILD step. Thank @Mugen87 because of removing self but still issue with nagivator and warning with await. |
I guess we can simply do the following? let isAvailable = ( typeof navigator !== 'undefined' && navigator.gpu !== undefined ); It does not matter that |
How does this warning look like? Is it still possible to compile your code with the below
|
It compiles with this warning:
I am on Node v22.10.0, and interestingly, I was not getting the ReferenceError because |
I think we can live with that warning for now. |
|
After another full day of testing and Dockerizing for more reliable reproductions, I have updated the repo with new findings. I have added an import of The main takeaway on the Three.js side is that importing top-level await modules such as You can test it easily:
Or Next.js:
|
Yeah, I still need to implement my own WebGPU async checking anyway. Because cannot risky update everything to latest. |
Fix #29916.
Description
Fix compilation issue on react.