[bug] window-end breakage

Has anyone noticed that `window-end' and `window-start' are almost completely hosed? (`window-end' is extremely useful when you want to operate on the visible portion of the window, for whatever reason.) Specifically: * (window-end) at the end of buffer returns (1+ (point-max)), which breaks any code that tries to use it sensibly. I now use (min (window-end) (point-max)). * The GUARANTEE argument makes it return a different value. Specifically, (window-end) returns the start position of the line after the clipped (bottom-most) line. OTOH, (window-end nil t) returns the end position of the end of the line *before* the clipped (bottom-most) line. I don't know which of the two is more correct, but I really think they should not differ. * `window-start' doesn't have a GUARANTEE argument. What use is it to be able to find window's end position without forcing a redraw without being able to do the same for its start position? I would really appreciate some help by the redisplay gurus.