Is there any reason why there aren't many aliases to ArcArray?

[varchasgopalaswamy]

All the other Array types have the dynamic and 1-6D aliases, while ArcArray only has 1 and 2 dimensional aliases. Was this an intentional omission? Easy enough for me to make my own aliases, but I wanted to check to make sure I wasn't missing anything!

[akern40]

I just noticed this today! To my knowledge, there's no particular reason - perhaps just that ArcArray is less-used. Would be happy to include the usual 6+dyn aliases, if you're looking to open a PR.

[varchasgopalaswamy]

Should be easy enough. But I guess while I have you here, a follow up question. If I have an Array<A,D> in one struct, and I want to have another struct hold a reference to it, I guess I need to use Arc?

[akern40]

ArcArray is used when you want to have multiple arrays pointing to the same data on the heap in a thread-safe manner. If you just want to hold a reference to an array in a struct, you can do something simpler, like

struct MyStruct<'a, A, D> {
  arr: &'a ArrayRef<A, D>
}

[varchasgopalaswamy]

The difference between Arc<Array> and ArcArray is whether the metadata + data is Arc (Arc<Array>) or just the data(ArcArray)?

In my use case above, the struct holding the reference (call it RefStruct) has ArrayD members, since it could hold 2-4D arrays (depending on the source of the data). All the struct's operations are elementwise or reductions, so it's easy enough to write code without actually caring about the dimension of the arrays. However, the struct that owns the arrays (call it OwnStruct) does care about the dimension of the arrays - basically it's a 1,2 or 3D simulation result, and some stuff does need to be handled differently. I would have made RefStruct generic in dimension, but pyo3 doesn't support exposing generics to python, and I need both RefStruct and OwnStruct to be exposed to python.

I would ideally also like RefStruct to be a field in OwnStruct. My understanding is this makes it a self referential struct, which I guess Rust isn't a big fan of. I am also not sure what that would do if I expose both to python using pyo3. From my googling, it seems like I ought to have OwnStruct store Arc versions of the arrays, and then use Arc::clone to initialize RefStruct.

If I have a function that takes an ArrayD, but my struct has a fixed dimensional array, can I convert it without copying? That way I could have OwnStruct store fixed dimension arrays, and RefStruct store dynamic dimensional arrays.

This is getting a little off topic from the original issue, so I'm happy to open a new one if you'd prefer.

[akern40]

I guess I'm just not sure why RefStruct needs to be inside OwnStruct here? But yes, you could have them both be ArcArrays and you do understand the differences in Arc usage.

[akern40]

Closed by #1561.

Feel free to continue our conversation if necessary, I'm just cleaning up!