Summarizing the previous discussion:

• awheel would like to declare itsoptionalJS dependency once imported as package
• we, thePyScript team, could offer an indirect solution through theconfig but that would not scale for suchwheel
• thewheel should not statePyScript is needed, so that such package, with its JS dependencies, could work elsewhere too
• the proposal is to have abrowser_env.json that dictates dependencies from the browser neededbefore the package executes
• my idea is thatenv.json would be a better place to have abrowser field that contains bothmain andworker related fields, to disambiguate the package need in case it's running from the main thread or a worker
• from now on ... it's discussion board/field, as long as we realize this is not aPyScript only request, this is something that should benefitpip or any publiched Python package to date in the future ... so please be careful with variants around this proposal, thank you!

Use-case

A cross-platform library implements functionality in a cross-platform way, across Windows, Mac, Linux, and browsers. It delegating to the appropriate platform-specific APIs and dependencies.
For example,funaudiolibrary supports audio playback across Windows, Linux, Mac, and browser using appropriate platform-specific APIs. In browser environments,funaudiolibrary leans on a.js shim/module which must be loaded into the browser environment.

Consumers offunaudiolibrary cansynchronouslyimport funaudiolibrary ; funaudiolibrary.play_sine_wave(440) without extra pyscript configuration.

Strawman proposal

Suppose a wheel can contain 2x files:

browser_adapter.js

// JS code that is loaded in the main thread or a worker thread async *before* the user's entrypoint .py begins executionclass EfficientJsTalksToWebAudioApi {  // ...}

browser_environment.json

// A manifest describing necessary bootstrapping tasks to support this python module in browser environment{  "main_modules": [    "./browser_adapter.js" // relative path within the VFS  ],  "worker_modules: [] // scripts to load in worker thread instead of main thread}

Browser runtime environments can bootstrap themselves like this:

• Ask micropip/pypy/pip to fetch dependencies (this is async downloading)
• Load them all into the VFS
• asynchronously load required JS shims into main or worker thread

fordepinall_deps:ifVFS.exists(dep.path+"/browser_environment.json"):browser_env_manifest=parse_json(VFS.read(dep.path+"/browser_environment.json"))forjs_pathinbrowser_env_manifest.main_modules:load_js_into_main_thread(VFS.read(dep.path+js_path)

• execute user's entrypoint .py

So when the user doesimport funaudiolibrary the library is able to synchronously access the JS library and call it immediately.

Examples in the wild: pyglet

pyglet does not current support browser runtimes, but they're interested in supporting them eventually. So they might use this proposal.

Audio

For example, pyglet supports multiple audio backends for different platforms. Roughly, it has a list of each backend. It iterates, attempts to load each in atry: block, stops at the first one which loads successfully.

I understand that pyodide/pygbag already supports some audio APIs, so maybe pyglet's existing OpenAL backend will "just work." Or perhaps pyglet will implement an additionalwebaudio backend which relies on a companion.js library.

Here are links to the relevant code:
https://github.com/pyglet/pyglet/blob/c374de9c654dac5af7e46b9c403af2f417f49bbb/pyglet/__init__.py#L53-L56
https://github.com/pyglet/pyglet/blob/c374de9c654dac5af7e46b9c403af2f417f49bbb/pyglet/media/drivers/__init__.py#L18-L19
https://github.com/pyglet/pyglet/blob/c374de9c654dac5af7e46b9c403af2f417f49bbb/pyglet/media/drivers/__init__.py#L62-L73

Font rendering

Another example: font loading, font rendering.

Here, pyglet usespyglet.compat_platform to decide which implementation to load. Each usesctypes to bind to native platform APIs.
https://github.com/pyglet/pyglet/blob/c374de9c654dac5af7e46b9c403af2f417f49bbb/pyglet/font/__init__.py#L33-L54

I can imagine pyglet adding another "web" platform, conditionally using the browser'sFontFace andCanvas APIs to load fonts. This could be done directly from python using FFI, but it might be more performant or easier to write the backend in .js using .js dependencies.