4/22/97: in nodesplit.c  int *which,   to  int *which;
                             **sorts;	   long **sorts;
                                    
5/15/97: in post.rpart.s  printit=F to print.it=F
5/27/97: fixed more int <--) long problems in the following:
	 choose_surg.c gini.c graycode.c mysort.c pred_rpart.c 
	 rpart.c rpart.h rpart.s rpcountup.c rplabel.c rpmatrix.c
	 s_to_rp.c s_xpred.c surrogate.c xval.c

6/12/97: fix bug with csplit, line 57
6/20/97: small changes to help files (`quotes')
7/15/97: small changes to Install, First.lib.S
8/28/97: small changes to help files
9/3/97:  update tech report, add in shorter version of tech report.
         made small changes to help files, modified functions per
	 B. Ripley suggestions, including fix to predict.rpart.

10/16/97:fixed new bug in snip.rpart with csplit 

11/21/97:fixed bugs in insert_split.c, mysort.c. Problem found by 
	 A. Prasad - endless loop when all of a group except 1 have 
	 the same value 

2/98:    Many changes prompted by helpful comments/critique of Brian Ripley.

1. All routines that call "S_alloc" now include rpartS.h.  The lines in this
   file differ between versions of S (Bell version 4 is the odd one at present).

2. The C code has all been converted to ANSI prototypes.  This helps with
   argument checking, and besides, it was about time to upgrade.

   The include file rpartproto.h contains prototypes of all of the central
   functions, i.e., not  anova, gini, or poisson.

   Also, remove the "register" declarations.  Modern compilers are
   smart enough without my help (and likely smarter) about registers.

3. Add an argument to free_tree.  If the second arg is 1, then free the
   node, all children, and all contents, if 0 free the children.  Now it can
   be called from rpart as well, for the inital parent node.

4. Fix bugs in pred_rpart.c and partition.c pointed out by Brian Ripley.

5. Change the dummy args in post and prune, per Brian Ripley.  This also
   affects prune.rpart.

6. Small improvement in predict.rpart.s, per Brian.

7. Changes in post.rpart.
   a. Agree with Brian, put args in the same order as post.tree (as much as
      possible). 
   b. Add a little trickery so that the arg can be "title" without getting
      a warning message from the title function.
   c. There are differences in the postscript functions between platforms!
      Rather than have two post.rpart functions, just drop the args that were
      causing problems and add a "...".  

8. Summary.rpart has changes needed in Splus 4.0, contributed by Brian.

9. The plotcp function was contributed by Brian.

10. Add in ordered factors.  Treat them as continuous vars for splitting
    purposes, then change the output structure so that they print as factor
    variables would.
    a. Essentially Brian's code, with some fixes.  (He ignored the case where
       there are ordered factors but no unordered ones, some things
       are then missing).
    b. Does have one problem: missing categories print as L/R rather than "-"
       This is a nice peice of information lost, but the resulting 
       prediction may be more consistent than if I did it the other way.

11. Several changes to the .d files, compliments of Brian.

12. The path.rpart function was contributed by Anantha Prasad (see path.tree)
