Rafael Schouten

@meggart what do you think of this as general change in approach? I was thinking when the type doesn't change we can just use the same allocated memory and skip...

So `lu` works now by accident, but then lu! doesn't work on a factorization holding LU DimArrays. \ works fine but doesn't return a `DimArray`

Maybe, can you explain the data a little better? I guess I'm not clear (even reading your link) on how you/DD would actually use the data, or what behaviours it...

Ok perfect. It's very similar to rotated X/Y dimensions in that you need both to resolve the indices. But in this case time is fine on its own, but X...

Yeah we need to add a third kind of `DependentUnallignedDimension`. I guess rotations could use it as well and just depends both ways Edit: or maybe just one Unaligned is...

Yeah, I think it's just a few more lines in the dim splitting code.

Yeah, that's kind of what `Transformed` is. One problem with being that generic is you can't subset the Matrix when you subset the dims, and you can't write a netcdf

> But it would require that multidimensional lookups no longer are strictly one dimensional Ah so its unaligned and multidimensional. That might mess with our type heirarchy

Yep. We might need to make one of those things a trait

`hasmultipledimensions` ? Idk