Ilhan Polat

It has nothing to do with backends. NumPy code can still live in the same Python file. The point is to remove the cythonization of this code not array API...

The PR introduces a Python class in pure Python. It is a new Python file.

That indeed smells like a bug. Since U and V are just block diags of the individual parts, probably the bug is living in the D generation part. Since you...

`cython_blas`, `cython_lapack`, and `cython_special` serve the purpose to access low level functions from Python/Cython. These are/were particularly useful since the fortran linkage is typically tricky. But now that `scipy.special` is...

If I understand what you want to do then this is a 2x2 linear system perturbation $$ \left( \begin{bmatrix} CC & CS \\ CS & SS \end{bmatrix} + \epsilon I...

You can also solve the system; $$ \begin{bmatrix} CC & CS \\ CS & SS \end{bmatrix} \begin{bmatrix} x_1 \\ x_2 \end{bmatrix} = \begin{bmatrix} y_1 \\ y_2 \end{bmatrix} $$ So if...

> this will be obvious to users straight away, and they'll have to learn more about using Generator to get it right. This is what I was trying to get...

If we can get a few examples crystal clear and comparing with old `np.random.seed(1324134)` I think users will appreciate it. The problem is that there is no hand-holding so far....

I didn't have a thorough analysis in the codebase but the typical usecase is to provide an int when you see the word `seed`. Anything else that goes into `seed`...

Like I said, I don't have any opinion but if we want to make people adopt the new methodology we have to take people with us. So far, it is...