Add Clone to MatrixIter and MatrixIterMut

[Shatur]

Could be needed for crates like https://docs.rs/logaddexp/latest/logaddexp/.
Other iterators in nalgebra also have Clone impl.

[Andlon]

I don't think it's sound for MatrixIterMut to implement Clone, since this could allow you to have two separate mutable iterators for the same mutable matrix entries, from which you could recover two mutable references to the same element in the matrix.

For MatrixIter I suppose it could be sound. But directly deriving might not be desirable since it will add a Clone bound for every generic parameter. In this case it's better to do a manual impl with only the bounds you actually need, I think.

[Shatur]

Makes sense, will do!

[Shatur]

But directly deriving might not be desirable since it will add a Clone bound for every generic parameter.

But we already have Clone for RowIter and ColumnIter.
Maybe do the same for MatrixIter by adding additional parameter to the iterator! macro?

[Andlon]

But we already have Clone for RowIter and ColumnIter.

Ah, I didn't realize that. That's quite unfortunate, and that's essentially a mistake. We should rather fix those Clone impls than perpetuate the same mistake for MatrixIter IMO.

[Shatur]

Are you sure that it's a mistake? What issues current Clone derive causes?

[sebcrozet]

Directly deriving Clone sounds fine. All the other type parameters already implement Clone anyway, and T: Scalar already implies T: Clone. So I would be in favor of keeping the clone derive, and relaxing the requirement late if we actually find a use-case that doesn’t fit.

But, yeah, Clone should not be implemented for MatrixIterMut.

[Shatur]

@sebcrozet thanks for letting me know!

Should I add additional parameter to iterator! and pass Clone, Debug for MatrixIter and just Debug for MatrixIterMut?

[sebcrozet]

@Shatur Yeah, I think that would make sense.

[Shatur]

@sebcrozet done!

[sebcrozet]

Thanks!