I'm guessing that this lecture covered Question 1, and left Question 2 open for later discussion, as just sampling triangle coverage doesn't necessarily determine correctly which triangles should be occluded. Maybe we could just draw primitives starting from those farthest from the camera to deal with occlusion as well? I believe raster pipelines often do this with a depth buffer, possibly with some other tech to query occlusion early on to avoid running fragment shaders that won't ever get rendered.
@zbp Yep, we'll talk about occlusion later on. Sorting is not a great solution because (i) you need to have all your geometry in memory at once, (ii) it can fail for intersecting primitives, and (iii) it can even fail for cyclic arrangements of non-intersecting primitives, e.g., part of triangle A is in front of B, part of B is in front of C, part of C is in front of A. In other words, "is in front of" does not necessarily give an ordering. On the other hand, sorting is kind of the only way to go for rasterizing scenes with partially transparent polygons; though in recent years techniques for order independent transparency have started to pick up steam.