Erik Tollerud

Ah, point taken this is a feature request. Will re-label!

> Can you please clarify? Did someone send you a pickled object instead of the JWST data product? Some examples: 1. Data streamed from an online source such that you...

(to be clear: in all of those cases, one can simply write to disk and then load that data, but that's pretty annoying and sometimes problematic from a storage perspective....

> For 1, streaming is currently not supported. It is an entirely different ticket/epic/whatever. Sorry, miscommunication there. I mean something like: ``` >>> res = requests.get('http://something') >>> image = PrimaryHDUOrWhatever.from_bytes(res.read())...

The snippet in the docs indeed are missing the metadata. so you are understanding that right (although there's surely a way to do that, it's annoying that I have to...

(Oh, and I'm not insisting on specific objects to support, but rather trying to understand how to plug in support for specific objects - i.e., like how unified I/O is...

> generic custom I/O for arbitrary image formats that's a good point, but just to be clear I'm realizing I was being a bit confusing in my wording above so...

Ah, that might be a good direction to consider, @pllim . I think the key, though, is ensuring it carries enough contextual information over. I think I worked out the...

Alright, I checked and at least for my science case the change I'm suggesting in #3279 makes an NDData object that seems to have no copying (`copy=False` by default in...

A few thoughts: While we might not need it now, it's possible we will need to pay for some github services in learn. A separate org makes that a lot...