Tested on i686-linux, committed on trunk

Previously, -gnatdI only removed serialization if a name started with
an upper case letter followed by digits, but now this sequence is
converted (replacing the digits by three dots) anywhere within a name.

Given the test program:

pragma Restrictions (No_Exception_Handlers);
procedure a is
   function f return string is
   begin
      return "abc";
   end;
   s : string (1 .. 3);
begin
   s := f;
end a;

The -gnatG output has lines like:

        [subtype a__B3b__S7b is string (R8b .. R9b)]

Before the patch, compiling with -gnatdI gave:

        [subtype a__B3b__S7b is string (R...b .. R...b)]

with the patch, we get:

       [subtype a__B...b__S...b is string (R...b .. R...b)]

This helps in building base lines for regression tests

2007-08-14  Robert Dewar  <dewar@adacore.com>

	* debug.adb: Improve -gnatdI to cover all cases of serialization
	Add documentation of dZ, d.t

	* sprint.ads, sprint.adb: Improve -gnatdI to cover all cases of
	serialization.
	(Sprint_Node_Actual): Generate new output associated with implicit
	importation and implicit exportation of object declarations.