(As requested by @ScreenDream)

There are two options available:

Option A: User-defined setting

1. The user defines the maximum length of the result buffer and must either accept truncated results or tolerate a portion of the buffer being unused (filled with separator points). Although this may sound inefficient, in practice, 90% of the performance cost lies in drawing the points or lines. In most cases, separators are not as problematic as they might seem.

Option B: Automatic buffer resizing

Count the required number of points, read the count back to the CPU (which takes a few frames), allocate a new buffer with the exact size, and copy the result into it.

As expected, this is non-trivial and may have performance implications. A similar approach was implemented for DrawMeshChunks, but the results were not ideal.

Option A may already be sufficient for many use cases.