I've got a question: I see that triu_matrix got renamed to tril_matrix. Was it incorrect?
AFAIU triu_matrix zeroes the upper part of the matrix instead of the lower (assuming row-major-order), or is it column-major implementation?

This is just what git diff does - really what happened here is that I added triu_matrix - `tril_matrix continues to live below!

Here's an implementation of triu_matrix in the current main branch:

And here's triu_matrix from your branch:
https://github.com/mhvk/numpy/blob/502bfb27968746f70155130b5fa9579e4d9a3a85/numpy/linalg/umath_linalg.cpp#L987-L1000

It looks like the implementation of triu_matrix changed here, right?

What I mean: tril_matrix is the one that is new and triu_matrix is the existing one, and I wonder why the existing one changed implementation. I might be missing something.

I think this is a typo in the main branch version. In def cholesky there's assert(uplo == 'L'); but later triu_matrix(params.A, params.N); is called.
So I think the change here is correct (assuming row-major order), but the release note should say that triu_matrix implementation changed and tril_matrix was added.

Ah, yes, sorry! I did change the implementation, as indeed for a triangular-upper matrix, one should be zeroing the lower part. I now recall puzzling over this too, and changing the comment above to try to make clear what was actually done.

Ok! I've got one more question: Here triu_matrix is zeroing upper triangular if one assumes the input is row-major order and lower triangular if one assumes the input column-major order, right?

Then maybe we could have just a docstring that explains why:

triu_matrix(typ *matrix, size_t n)
{
    size_t i, j;
    for (i = 0; i < n-1; ++i) {
        for (j = i+1; j < n; ++j) {
            matrix[j] = numeric_limits<typ>::zero;
        }
        matrix += n;
    }
}

looks like it zeroes upper triangular but the function's name says triu? Just a comment that matrix is column major (I think so).
WDYT?

Yes, makes sense!

OK, added the extra comment (really good to have indeed!)