Dieter Weber

On Linux the kernel seems to do a good job with prefetching. This would mostly be a topic for Windows to solve the performance issues with memory maps there.

- [ ] Refs #1089 -- include annular CoM in CoM example drive-by

I've played around a bit with that and made some progress for buffers. However, we assign to slices also in other places, most notably I/O code. All this code is...

It seems that manually checking is our best option for now. I've raised https://github.com/numpy/numpy/issues/19831 since checking within NumPy would probably be much easier and faster to implement. The validation code...

As a general remark, we should make it a habit to include version information and changelog info early-on. :-)

This is a surprisingly small change! 👍 How does it find the correct cache path for an external location? Will users always have to remember where they cached a DS...

About naming: `ClusterDS` and `CacheDS` to avoid confusion with dask.distributed?

Hi @Algoritter, thank you for the issue and detailed description! Could you perhaps translate it to English so that other stakeholders can understand it? https://www.deepl.com/translator can help you. Many thanks...

Ah I see, one could develop an UDF that calculates the histograms. Is this still current?

See also #427