@jepler @v923z This should allow for importing submodules of ulab.

I've linked @jimmo to it and I bet @dpgeorge would be interested too.

So the claim is that

from ulab import fft

fft.fft(...)

would work with this fix, right? If so, it is really great, thanks for taking care of it!

@dpgeorge Damien, do you think this could be merged into micropython proper on a reasonable time scale? With that we could maintain a single version of ulab, ant that would save some resources on our end.

@tannewt thanks for the note, I'll look into incorporating this or similar. I didn't yet look in detail but this might be related: micropython#4731

See micropython#5701 for the equivalent change upstream. Thanks for suggesting this, I think it's a much better approach than micropython#4731

Some minor notes on this PR:

• Might need to check when looking at the globals dict that the result is actually a module (i.e. type == mp_module_type).
• You can avoid the change in objmodule.c by just setting a flag in builtinimport, then avoid calling mp_store_attr at all.

Thanks @jimmo and @dpgeorge. Will keep an eye on upstream and plan on replacing this PR next time we merge from upstream.

@jimmo Good idea on the type check. I've added it and confirmed we can't import microcontroller.cpu which is not a module. My preference is to keep the set_attr because I'm assuming a module will track whether it's members have been imported in the future rather than having import manage it.

@jepler I think this is ready to merge.

@jepler Removed! Please take another look!