On Tue, 18 Jul 2006, Kirk, Benjamin (JSC-EG) wrote:
> this->_element being deleted is a bigge, so that clearly needs to be
> The other problem, however, should be an easy one... The fallback loop
> probably *should* cover all elements, but what is needed is a search of
> all children when a point is located in an inactive element. If a point
> is contained in a parent but none of the children then we probably want
> to trap for that. The only time I would expect this to happen is if the
> boundary nodes are moved to better represent the geometry.
Don't forget Derek's smoothing; that can move child elements outside
their parent too.
We've already got some implicit "a parent element's closure coincides
exactly with the union of the closure of its child elements"
assumptions in System::project_vector(), but if we can avoid adding
them elsewhere in the code we probably should. So what we ought to be
doing is searching through only active elements in that fallback loop.
I'm more worried about the underlying Tree and List implementations,
though; the fallback loop is an easily fixed two line bug, but the
lack of AMR-awareness pervades the rest of the PointLocator code too.