@ntoll this is a before / after in our CI ... I have 12 secs before and 6 secs after so it's aligned with almost twice as fast Pyodide bootstrap and usage which is closer to native (on main) performance.

FYI this has been published on npm as both https://cdn.jsdelivr.net/npm/@pyscript/[email protected]/dist/core.js and https://cdn.jsdelivr.net/npm/@pyscript/[email protected]/dist/core.css

@ntoll I've moved the ugly way to use the experimental worker boost from global script to the config which feels like more appropriate.

Description

if experimental_ffi_timeout is specified and it's a number (0 or more but not -1, which is the default) then the reflectedffi* will bootstrap with smart cache enabled so we can better battle test that feature beyond just our twice as fast worker tests in Pyodide.

Ideally, we should make it 0 (or more) by default but let's make it opt-in and gradual.

OK, in case we want to release this (we should) here I am writing down what this latest update would bring in:

Release Notes

• added and tested a new PyScript Bridge helper companion, to import from JS Python modules
• fixed a bug in <label> Fix a bug in <label> handling where 'for_' attribute should be 'htmlFor' on underlying HTML element. #2352
• updated polyscript dependency to its latest which brings in:
  • a new packages_cache = "passthrough" option, beside "never", to avoid lazy micropip initialization and parallelize Pyodide bootstrap when network connection is good and offline ability is not a concern. This should help speeding up bootstrap Fix #137 - Allow the passthrough option for the cache polyscript#138
  • added a debug = True option in config to help debugging Pyodide Adding debug config option to help improving debugging in Pyodide polyscript#135 - please note this degrades Pyodide performance
  • updated coincident orchestration to bring back the service-worker attribute without needing @pyscript/service-worker at all and tested in multiple devices, including iOS and iPad OS
  • updated also coincident ffi logic via a dedicated project able to improve performance a lot, specially with all the JS primitives where performance matters, including TypedArray of all kinds and ArrayBuffer
  • coincident ffi also has a smart cache feature that is able to make Pyodide bootstrap and performance nearly 1:1 with the main thread. Roundtrips can be cut off entirely for most JS primitives while DOM nodes or arrays from the main thread are not cached to preserve their constantly mutable state and preserve correctness.
  • coincident ffi also exposes some extra feature that is reflected through pyscript.ffi, specially direct(ref) to pass directly values instead of holding these on the main thread and assign to update many fields at once from a worker. Both utilities work seamlessly on either main or worker but on worker these will have a boost.