missing one(::Mat)

[stevengj]

Currently this works but produces a dense matrix.

[kshyatt]

Is the right thing to do here use MatDuplicate and then MatSetValues or is there a better/easier way?

[stevengj]

I don't think we want to duplicate the nonzero structure, so maybe just allocate a fresh matrix rather than calling similar.

Note that there is also a MatDiagonalSet, though I doubt it has any advantage over MatSetValues in this case.

[kshyatt]

Wait - do you want ones(M), or eye(M)? I thought ones preserved the non-zero structure but set all values to one(T)?

[stevengj]

I want one(M) = eye(M), not ones(M).

[stevengj]

I should also mention diag(M) and diagm(v) via MatDiagonalSet/Get, while it is on my mind.

[kshyatt]

Aha! Then in that case, I agree that MatDuplicate isn't at all what we want :). I think I wrapped diag(M) using MatDiagonalGet but I didn't do diagm.

[stevengj]

ones(M) does not preserve the non-zero structure, it just makes a matrix of all 1's. I don't think it is particularly useful in the case of PETSc. Also, I think it already works, because the default implementation probably calls fill!?

[kshyatt]

Oh god, I was confused between one and ones.

[stevengj]

Hmm, looks like ones(M::Mat) gives garbage.