Unlike Direct3D 9, the full extent of the resource view is always cleared. Viewport and scissor settings are not applied.
The solution Microsoft proposes is actually that used by the Chromium/ANGLE projects at Google.
However, I actually think the way GMS gets around this problem may be the fact that each view might have its own render target. This means the render target of each view is already the size of the view, hence no additional clipping during clear is necessary. The documentation isn't clear (pun again) whether they create you a surface for each view if you don't assign one (it sounds like they don't since that would be expensive).
it will return -1 if no surface has been assigned