This package implements a generalized interface for array-like objects. The idea is that we should be able to index an object as (xref object 1 2 3) to retrieve or set an element. This would be easy to do it aref was a generic function, but it is not. I describe a simple interface with a few generic functions (xdims queries dimensions, xtype element type, etc), any objects that has these methods is called "xrefable". The nice part is that views on an array can also be xrefable. For an two-dimensional array *a*, (slice *a* :all 4) selects the 4th column of an array.[1] This view is also xrefable, so you can use (setf (xref (slice *a* :all 4) 2) 9) to set element 2 to 9. Changes will be made in the original array *a*, since this is a view. But for functions that just expect an array, they can use the interface (xref, xdims, etc) and not care where it is coming from. [1] The slice interface is similar to my affi package, but now arbitrary index vectors are allowed, much like R. I have also implemented permutations (generalized transpose), and row-major projections. If you want to collect the elements of a view in an array, use (take object), which delivers a CL array. There are convenience functions that for easy array/vector creation, inspired by R: (carray* '(2 3) 1 2 3 4d0 5 6) ;; => #2A((1.0d0 2.0d0 3.0d0) (4.0d0 5.0d0 6.0d0)) guess the type so that all of the elements would fit in. See also cvector and carray (manually specified type) and cvector*. Roadmap: 1. unit testing. supremely important. 2. integrate linear algebra routines, probably from GSLL. It should be easy to rig an xrefable interface to GSLL arrays. 3. upper- and lower-triangular matrices, etc. xrefable elements can be read-only, xref-writeable-p is an interface to test if an element is writeable, it was included specifically for this. 4. decent printing for xrefable objects, I currently convert them to array. 5. integrate sparse matrices from cl-sparsematrix. 6. specialized subclasses for certain cases. eg views on matrices, a transpose-view would be much simpler (and faster, maybe?) than the generalized permute. 7. certain views can be directly accommodated by LAPACK/GSLL (eg a matrix with a stride). Minor possibility for speedup/memory savings. Tamas K Papp tkpapp@gmail.com