o) we have had some problems, but they are solved now
     ---> tests/wgt-himed.R etc	  {MM, Dec.2005}

2) I think its wrong in principle to only work with *integer* weights
      for 'whimed'.
      Further note, that *because* of the integer weights, we need type
      'long long' {because otherwise integer overflow kills everything}.

   Using C Macro, we now have both, integer and double weights;
   and I have tests of consistency in ./tests/
   --> using wgt.himedian() as an R function {MM, Dec.2005}

   Note however that a more general function (weighted quantiles) might be
   of interest, see file ./TODO

3) Data Sets --- Valentin Todorov has several of Rousseeuw's
	     in his "rrov" package

   --> got them, and dropped all *.x and *.y data sets
       -- as part of the loaded data -- but kept a definition
        <foo>.x <- as.matrix(.....)
      in the examples section.	 {MM & VT,  Dec.2005/Jan.2006}

4) covMcd( * ):  replace 'print.it'  by  'trace'  which is more appropriate
   	         and common in S

5) renamed  tolellipse() to tolEllipsePlot()
   all these plotting functions may *NOT* set  par(mfrow = c(1,1)) !

6) considerable simplification in the  correction factor computations

7) glmrob(): trailing "..." argument is now passed to glmrob<FOO>control()

8) make tests/wgt-himed-xtra.R an order of magnitude faster (less
   simulations)

9) man/fitted.nlrob.Rd merged into man/nlrob.Rd

10) glmrob() now works when the internal glm.fit() returns NA coefficients
    by dropping the corresponding columns (as Martin's "robGLM1" did).

11) 'family = binomial' is a problematic default, when OTOH,
  glm() has  'family = gaussian' as a default.
  ==> have *NO* default, so users need to set family explicitly.

12) Instead of modsel.glmrob() [ R/modsel.glmrob.R ]
    Andreas has provided anova.glmrob()

13) covOGK() is now consistent per default
    because it is used with scaleTau2() which has now a constant
    consistency factor by default.
