The point is to print one logical line.  Printing a fragment of the logical line is clearly not very helpful...I have found myself a number of times looking for the bug in the fragment printed, and (quickly, but still...) then remembering that the traceback just prints the last physical line and the problem might be earlier.

Providing a caret is probably impossible.  I don't think the bytecode contains enough information to even make a guess as to where a caret should go.  I'm not even sure we can get the logical line.