Issues of code quality:
* MBR and MINIMUM-BOUND-OF no longer need the tree as an argument,
because we precompute rectangle data for a leaf entry.
* There are numerous instances of minimization loops where there is no
good upper bound for the value, so that the initial value for
comparison must come from the first iteration of the loop. This is
dealt with in a number of different ways throughout the code, none
of which is particularly elegant.
* The test harness is reasonable at the moment; it could be improved,
and of course more tests should be added.
* The CHECK-CONSISTENCY function should be used in the test suite.
However, a means needs to be developed to be able to say that the
X-tree, say, obeys all of the invariants of the R-tree except for
the maximum number of children of a node in certain circumstances.
This subtraction of invariants appears to be a poor fit with the
simple class hierarchy.
* The visualiser is a reasonable barometer onto the tightness of the
protocol; the current verdict is "not terribly". Aim to reduce
initmate knowledge of internals in the CLIM inspector in as much as
Issues of performance:
* MBR seems a little too low-level to be implemented as an (expensive,
at least in SBCL) generic function. A simple profile on 2004-11-30
showed it to be the most time-expensive function in running the
tests on R-trees. Maybe it should be replaced by an inline function
calling the leaf-node-entry accessor or the spatial-tree-node
SLOT-VALUE as appropriate.
* REDUCE <set> :KEY #'MBR (or the equivalent) takes a surprisingly
long time. The compiler's understanding of REDUCE could be
Issues of generality:
* The dynamic insertion algorithm for R+ trees is broken. It would be
nice to fix it.
* MORE TREES
* The visualiser should present and describe the object in a
leaf-node-entry, not just rectangles.