The core of Mondrian is not independent of the cache implementation. The logic of Mondrian
rely on the cache to implement its semantics. This is obviously wrong.

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
bounds
    "Answer the bounds of the receiver."
    "the bounds is has an absolute origin"
    "Note that the bounds computed above, may have (and it is likely to) a different origin.
The reason is that the layout is in charge to position the nodes properly"
    | basicBounds |

    boundsCache ifNotNil: [ ^ boundsCache ].

    "We check if  the shape if present"
    self shapeBoundsAt: self shape ifPresent: [ :b | ^ boundsCache := b ].

    basicBounds := self shape computeBoundsFor: self.
    self shapeBoundsAt: self shape put: basicBounds.
    ^ boundsCache := basicBounds
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
translateAbsoluteCacheBy: aPoint
    absoluteBoundsCache ifNil: [ ^ self ].
    absoluteBoundsCache := absoluteBoundsCache translateBy: aPoint
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=

Thanks to Santiago Vidal for pointing this out!