1572644 - Support "import" in worklet scripts

Status: ASSIGNED

(Core :: DOM: Core & HTML, enhancement, P3)

Description

[Karl Tomlinson (back Jan 31 :karlt)]

+++ This bug was initially created as a clone of Bug #1473463 +++

ScriptLoader supports fetching module script graphs but currently has a number of main-thread-only characteristics. The approach proposed in bug 1311726 would permit using this from worklet code.

Jon said in https://bugzilla.mozilla.org/show_bug.cgi?id=1290021#c34
"It would be really great if we could reuse some of the code in the script
loader (and would be useful for other efforts, e.g. bug 1308512). This is not
trivial unfortunately. I think it would require factoring out the module
loader into a separate class with a caller-provided a fetch interface passed
in."
This would be bug 1311726.

Andrea said in https://bugzilla.mozilla.org/show_bug.cgi?id=1472324#c1
'we must unify WorkletFetchHandler,
workerinternals::Load/LoadMainScript and mozilla::dom::ScriptLoader.'

Ben said in https://bugzilla.mozilla.org/show_bug.cgi?id=1442776#c1
"Would it not be possible to make worklets and workers completely separate?
Decoupling them seems like it would be nice for maintenance since they are
going to be fairly different."
That probably means that any shared code should be in or near ScriptLoader,
rather than in dom/worklers.

Houdini WG said in
https://github.com/w3c/css-houdini-drafts/issues/506#issuecomment-343276228
"RESOLVED: dynamic import() in worklets is a no-op that returns a rejected
promise (or equivalent after passing thru JS people)"

Web platform tests expect static import to work.
https://github.com/w3c/css-houdini-drafts/issues/506#issuecomment-342796390

Comment 1

[Paul Adenot (:padenot)]

It would be great if we could revisit priorities here. People expect imports to work in AudioWorklet, that we just enabled in Nightly (it's supposed to work per spec, and it works in Chromium). I'm seeing some websites that rely on this. AudioWorklet usage is still not widespread however.

https://superpowered.com/webbrowserlatency
https://mtg.github.io/essentia.js/examples
https://googlechromelabs.github.io/web-audio-samples/audio-worklet/

Comment 2

[Hsin-Yi Tsai (she/her) (back on Feb 5) [:hsinyi]]

(In reply to Paul Adenot (:padenot) from comment #1)

It would be great if we could revisit priorities here. People expect imports to work in AudioWorklet, that we just enabled in Nightly (it's supposed to work per spec, and it works in Chromium). I'm seeing some websites that rely on this. AudioWorklet usage is still not widespread however.

https://superpowered.com/webbrowserlatency
https://mtg.github.io/essentia.js/examples
https://googlechromelabs.github.io/web-audio-samples/audio-worklet/

Thanks for raising this. Would take to our roadmap planning discussion. :)

Comment 3

[Paul Adenot (:padenot)]

https://ide.csound.com/

Comment 4

[Anne (:annevk)]

Also bringing this to the attention of Adam and Jens.

Comment 5

[johannes.feulner]

(In reply to Hsin-Yi Tsai [:hsinyi] from comment #2)

(In reply to Paul Adenot (:padenot) from comment #1)

It would be great if we could revisit priorities here. People expect imports to work in AudioWorklet, that we just enabled in Nightly (it's supposed to work per spec, and it works in Chromium). I'm seeing some websites that rely on this. AudioWorklet usage is still not widespread however.

Here is one more Website using WASM in AudioWorklet: https://ssc.scorio.com/localtest.html

Comment 6

[hjack]

I'm part of a team using WASM in AudioWorklets, via Superpowered, and would greatly appreciate this feature! We are trying to add Mozilla support and facing complications because import is not supported.

http://samples.landr.com/

Comment 7

[Jack Atherton]

I'm not sure if this is the right place for this.

Webchuck (ChucK in the browser) could also benefit from the ability to "import" in the AudioWorklet.

This version works in Chrome but not Firefox, exhibiting the "Module resolve hook not set" error message.

As suggested in 1636121, I've made a version that uses --extern-post-js to avoid the use of any imports. This faces the new bug "Module metadata hook not set", for which I haven't found any useful workarounds / bug reports.

This version works in Chrome but not Firefox, exhibiting the new bug. I'm not sure if this is completely unrelated, or if I've done something wrong in my attempt to eliminate import statements / ES6. Perhaps I can't use the -s MODULARIZE=1 option in Emscripten, even though I'm not using it alongside import statements?

Comment 8

[Jack Atherton]

It turns out I made a very silly mistake and did not remove the flag that was enabling ES6 support. I was able to flatten all of the imports, etc. into a single extern-post-js file, and things seem to work now. Here's the command that seems to be working for ChucK in Firefox, in case it helps anyone else facing this bug:

host_web/webchuck/js/webchuck.js: $(EMSCRIPTENSRCS)
	emcc -O3 -s DISABLE_EXCEPTION_CATCHING=0 \
        [many flags related to ChucK internals]
	-Wformat=0 \
        $(EMSCRIPTENSRCS) -o host_web/webchuck/js/webchuck.js \
        -s EXTRA_EXPORTED_RUNTIME_METHODS='["ccall", "cwrap", "getValue", "setValue", "addFunction", "removeFunction", "UTF8ToString", "stringToUTF8"]' \
        -s LINKABLE=1 -s MODULARIZE=1 -s 'EXPORT_NAME="ChucK"' \
        -s ALLOW_MEMORY_GROWTH=1 \
        -s --extern-post-js host_web/chucknode-postjs.js \
        -s RESERVED_FUNCTION_POINTERS=50 -s FORCE_FILESYSTEM=1

Comment 9

[Anne (:annevk)]

Clearing needinfos in favor of bug 1311726 comment 2 as that bug is the main thing that needs fixing here as I understand it.

Comment 10

[Maxime Chevalier-Boisvert]

Hello. I got here because I also stumbled upon this bug while trying to build a modular-synth-in-a-browser app that uses AudioWorklet. ES6 support in AudioWorklet would be much appreciated :)

Comment 11

[guest271314]

Still an issue on Firefox 96.

Error: Module resolve hook not set self-hosted:2022:31
    CallModuleResolveHook self-hosted:2022
    InnerModuleLinking self-hosted:2235
    declarationInstantiation self-hosted:2205
Uncaught (in promise) DOMException: The operation was aborted. 

Comment 12

[Hauke Krüger]

Hi, is there any recipe how to get wasm audioworklets to work with Firefox these days? It seems that even the links from above,

https://superpowered.com/webbrowserlatency
https://googlechromelabs.github.io/web-audio-samples/audio-worklet/

do not work with Firefox 101.

I tried so many options in the emcc command but none works. And it is always the same error message in the JS console. A proper error report or an explanation which options to use to get any example working would be greatly appreciated!

Comment 13

[Lukas Lihotzki]

rollup can be used to bundle worklet modules before publishing, so they can be loaded without imports.

rollup also runs in the browser. This can be used to implement a polyfill to work around this issue purely on the client side: https://gist.github.com/lukaslihotzki/b50ccb61ff3a44b48fc4d5ed7e54303f

Comment 14

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 1: Move WorkletFetchHandler to its own file.

This is a refactoring to move WorkletFetchHandler to its own file, so
later other classes can include its header.

Comment 15

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 2: Init WorkletModuleLoader when WorkletGlobalScope is created.

Comment 16

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 3: Add StartModuleLoadRunnable.

Add a Runnable class called StartModuleLoadRunnable, which is dispatched
from main thread to worklet thread to call
ModuleLoadRequest::StartModuleLoad.

Comment 17

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 4: Add WorkletLoadContext.

Add WorkletLoadContext to delegate WorkletFetchHandler.

Comment 18

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 5-2: Implement WorkletModuleLoader::StartFetch with WorkletFetchHandler::StartFetch method.

Move the fetch code into WorkletFetchHandler::StartFetch, which will be
called by WorkletModuleLoader::StartFetch.

Comment 19

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 6: Set CORS and referrer on the Request.

Comment 20

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 7-4: Add FetchCompleteRunnable to call OnFetchComplete.

Add a class called WorkletScriptHandler to handle the loading of each
module script.

Add a Runnable class called FetchCompleteRunnable to be dispatched from
main thread to worklet thread to call
ModuleLoadRequest::OnFetchComplete.

Comment 21

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 8: Compile fetched module.

The compile options is from ExecutionRunnable::ParseAndLinkModule of the
previous patch, or could be found in
https://searchfox.org/mozilla-central/rev/fb9a504ca73529fa550efe488db2a012a4bf5169/dom/worklet/Worklet.cpp#386-389

Comment 22

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 9: Create static import.

Comment 23

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 10: Implement WorkletModuleLoader::OnModuleLoadComplete.

Comment 24

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 11: Add GetResolveFailureMessage and override it in WorkletModuleLoader.

Comment 25

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 12: Using StructuredCloneHolder to pass a JS::Value from a worklet thread to main thread.

Comment 26

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 14: Update ini files in wpt/.

There are still some failed test cases in audio-worklet-referrer.https.html.ini,
which is due to https://bugzilla.mozilla.org/show_bug.cgi?id=1808189.

Comment 27

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 5-1: Remove mErrorStatus and mURL from WorkletFetchHandler.

Unify to reject the promise with AbortError per spec requirement.

Also WorkletFetchHandler will fetch not only the top level module script
but also its dependent modules, so remove the mURL as well.

Comment 28

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 7-2: Add WorkletScriptHandler.

Add a class called WorkletScriptHandler to handle the loading of each
module script.

Comment 29

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 7-1: Add DispatchFetchCompleteToWorklet.

Dispatch error to worklet thread.

Comment 30

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 7-5: Add WorkletLoadContext into CC.

To fix the compilation error when traversing mHandler in
WorkletLoadcontext.

Comment 31

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 7-3: Add a hash table to map from nsIURI to ModuleLoadRequest.

Add a nsRefPtrHashtable<nsURIHashKey, ModuleLoadRequest> called mFetchingRequests
to map from nsIURI to ModuleLoadRequest.

When WorkletModuleLoader::StartFetch is called, it will insert an entry
into the table. Later when the main thread finishes the fetching of the module
script, we will use the URI to get the ModuleLoadRequest on worklet
thread and call ModuleLoadRequest::OnFetchComplete.

Comment 32

[Yoshi Cheng-Hao Huang [:allstars.chh][:allstarschh][:yoshi] OOO ~ Feb. 3]

Attachment: Bug 1572644 - Part 13: Get the localized message on the main thread.

On Worklet thread, we just return the error message with "{0},{1}",
where {0} is the ResolveErrorInfo message and {1} is the specifier.

Later when we are on the main thread, we extract the error info and
specfier and use them to get the localized error message.