Dieter Weber

> Hmm, possible, but maybe not necessary - I guess for the web API we will need some kind of registry, which can be implemented independent of namespacing. Ok, probably...

This is an ongoing task. Step by step!

The test case fails, actually: Despite using `process_partition()`, the tile has the shape specified when creating the memory data set. What is the correct behavior here? Feeding a tile that...

Ok, that's a good point. Probably a "partition ROI". Feels similar to #764, actually!

Just yesterday, I'm at the ESTEEM3 Electron Diffraction Workshop in Trondheim ATM. Looked very handy, a spin-off from the pyxem/diffsims universe to make this stuff reusable.

Support for Dask-based execution should probably remain as an option even after we have an advanced executor, since Dask allows to run on ipyparallel clusters and enables integration with Hyperspy...

@sk1p @matbryan52 This could be the right issue to coordinate the executor and UDFRunner redesign.

@matbryan52 Thank you for the diagram! It looks a bit intimidating. :-D It will be a great tool to understand the flow, we can work on refining that and then...

Discussions with @sk1p and @matbryan52: We want to support both live processing and offline processing, in particular on a Dask cluster with Dask arrays. In order to share as much...

For the current K2 prototype: The UDP receiving, ordering and decoding part could be separated from the actual processing. The decoder would then send the parts to the appropriate recipients...