[mulimoen]

Simplified to use NnzIndex directly without going through a view

@rodrigodev Do you have more tests which should be added to this PR?

[rodrigodev]

We hit this on 0.11.5, aarch64-apple-darwin, Rust 1.98. Your test covers the exact case from my report, and I checked it: it fails on master and passes on your branch. Nothing more is needed for #410.

For reference, the reproducer as a test:

#[test]
fn issue_410() {
    use crate::TriMat;
    let mut t = TriMat::new((3, 3));
    for (r, c, v) in [
        (0, 0, 1.0),
        (0, 1, 4.0),
        (1, 1, 2.0),
        (1, 2, 5.0),
        (2, 2, 3.0),
    ] {
        t.add_triplet(r, c, v);
    }
    let mut m: CsMat<f64> = t.to_csc();

    m.set(1, 1, 20.0);
    assert_eq!(Some(20.0), m.get(1, 1).copied());
    assert_eq!(Some(4.0), m.get(0, 1).copied());
}

Before and after on the same machine:

master (1525319)

before: (1,1) = Some(2.0), (0,1) = Some(4.0)
after:  (1,1) = Some(2.0), (0,1) = Some(20.0)

issue_410 (5d2332e)

before: (1,1) = Some(2.0), (0,1) = Some(4.0)
after:  (1,1) = Some(20.0), (0,1) = Some(4.0)

The one test I would add is the same thing in CSR, since the fix routes both storage orders through nnz_index and only CSC is covered. It also fails on master and passes on your branch:

#[test]
fn issue_410_csr() {
    use crate::TriMat;
    let mut t = TriMat::new((3, 3));
    for (r, c, v) in [
        (0, 0, 1.0),
        (0, 1, 4.0),
        (1, 1, 2.0),
        (1, 2, 5.0),
        (2, 2, 3.0),
    ] {
        t.add_triplet(r, c, v);
    }
    let mut m: CsMat<f64> = t.to_csr();

    m.set(1, 2, 50.0);
    assert_eq!(Some(&50.0), m.get(1, 2));
    assert_eq!(Some(&4.0), m.get(0, 1));
}