This works, until something about the image violate your expectation. To use the image example, say you have an image editor that uses the common assumption tha

This works, until something about the image violate your expectation. To use the image example, say you have an image editor that uses the common assumption that a pixel is a 3-byte item, with 255 possible values. Lots of stuff worked this way, its sort of a base assumption around which you have to build all your other stuff. Violating it requires you to like change sizes everywhere, which may mean performance of your fancy vectorized algorithm plummets somewhere. Anyway, lots of stuff follows this 8-bit-per-channel api. Until you use RAW images or any number of higher color depth . Then you need 12 or 16 or 24 or 32 (or more) bits per channel.