Infinite loop in `multiply` when applied to NxN and N shaped arrays

[Caellian]

In browser environment, when I used multiply, it got stuck on NDIter in an infinite loop.

• The cause is that NDArray arity isn't checked by the multiply function and it assumes it will work with two matrices. If one tries to multiple a matrix ([[...],[...],[...]]) with a vector [..], the c2 value is undefined.
• This causes a subsequently constructed result matrix to have 0 size, i.e. undefined length.
• Which isn't checked either when NDIter is constructed, resulting in an infinite loop because index > undefined is true (because undefined is cast to 0).

Solution

• Check if vector is provided as second argument and turn it into a matrix.
  • Handle vectors from NDIter?

[mateogianolio]

Hey, thanks for creating an issue! Multiply is not as versatile as numpy's dot in python. As described in the docs, it "Multiplies two matrices x and y of matching dimensions. Accelerated with BLAS ?gemm.".

If you want to multiply a matrix with a vector, you need to first make the vector a column/row matrix and then perform the multiplication.

I think the more correct way to address this shortcoming is to make dot behave like numpy's dot.

EDIT: And also make multiply handle the infinite loop. It should instead throw an error.

[Caellian]

Thanks for clarification. I ended up using gl-matrix because it was more in line with what I need algrebra for (WebGL), but it's much less ergonomic because it operates on TypedArrays directly.

I believe it would be good to note this in docs as well and point to dot function. I assume a.size.length !== b.size.length is enough to check for this case, but yeah, whatever doesn't make the whole browser freeze is great haha.

I'd like to point out that ability to distinguish between dot and multiply probably depends on previously used libraries. In my case, I'm coming from nalgebra which overrides the multiply operator so I'm probably expecting dot when I see multiply.