
 Comparing version 3.4 to 5.1, one difference on the printout is a change
of 3989 to 3990:  signif(3989.5) rounds down in 5.1, up in 3.4 (at least
on my home Linux machine it rounds thus).

----
 Comparing the 5.1 fits to the 6.0 ones with the new user-written feature

   1. Some roundings again: 2980.5 becomes "2980" in one and "2981" in the
other.  Now the routine is using formatg() to round them.

   2. The CP tables are larger  -- we no longer remove "almost tied"
lines of the output.

   3. The printout for class regression has changed slightly.  In print,
two blanks have diasppeared, and the probabilities now print as only 2
digits if there are more than 3 of them (to save space).  Digits are
also changed in summary.
     However, this printout is now localized into simple functions in
rpart.class.s, so users can readily change things if they wish.

   4. The tests for weighting (treble*.s) had to be changed.  There were
a two exact ties for goodness of split; and in the current code round
off error would cause one to be chosen first in the weighted and the other
in the unweighted splits.  Thus, the all.equal test would fail even though
the routines were ok.  For the cars, changed which was the y variable; and
for the other data set set the max depth of the tree to avoid this.






