TODO

TODO

[] opSpot class: move funhandle and matrix props, which are def'd in
   many (most?!) of the subclasses, into opSpot. (These are used too
   often.) Then can have opSpot/double.m do a simple check if
   ~isempty(op.matrix).

[] opBlockDiag.m: Avoid replicating operator explicitly?  How to
                  display repeated blocks:
                  opBlockDiag(ones(10,1),opDCT(4))?  make sure help
                  comment mentions weight vector in combination with a
                  single operator as well.

[] @opSpot/diag.m; add documentation

[] opInverse.m: Should we allow operators such as opInverse on
   non-linear operators? NO!!! Check linear flag. LIKEWISE for PCG,
   LSQR, ...

[] Add inverse implementation for opBlockOp
[] Clean up multiplication implementations, see e.g., opCurvelet
[] opWavelet.m: Help comment
[] opToeplitz.m, opToepGauss.m, opToepSign.m: Help comment
[] Comments in the @opSpot files
[] opSurfacelet.m: Haven't got a clue about how to use? Still have to test.
[] For each operator, check if complexity, linear flag needs to be set or not
[] opSubsRef.m, opSubsAsgn.m, opExcise.m: Add proper help
[] Check constructor comment block in each file
[] DCT, DCT2, FFT, FFT2, HAAR, HAAR2, WAVELET, WAVELET2; THESE ARE
DERIVED FROM opOrthogonal.
[] Redundant wavelet transform: RWavelet, RWavelet2.
[] Is there anything we can to for printing opBlockDiag(20,opOnes(4,4),2)
   nicely?

Precedence levels
1 - Operators that never need brackets, such as opDCT, opPower, opTranspose
2 - Unary minus
3 - Multiplication
4 - Subtraction, addition

DOCUMENTATION
-------------

[] Note that the random operators depend on old-style calls to
randn('state'). These are not compatible with the latest stream-based
calls. Need to document this.