Thanks for the thorough and considerate response.

I do think your original recommendation of reverting the broken feature is the best approach at this point. That is, keep resources.files with the fallback shim and eliminate support for loaders supplying that behavior. That will avoid users relying on that protocol but enable the files feature for Source and Zip importers.

That will buy time for the remaining functionality, mainly the provider interface, to mature and possibly evolve further, for eventual adoption in a future Python release.

Some planning will need to be dialed back, but I don’t have confidence in the implementation to say that it’s the best one. Better to defer that effort.

I’ll put together a patch for 3.9 to remove the loader support (backward incompatible with b1/2 but compatible with 3.8 but only for custom loaders), and put together a hot fix for master so it’s no longer broken.