With this patch the compiler now handles properly the use of @ as a prefix
of a reference to a discriminated record component and to its bounds.

The following must compile quietly in gnat2020 mode:

---
procedure Discrs is
begin
   declare
    -- Discrim-dependent subtypes
    subtype Index is Integer range 0 .. 123;
    type R1 (D1 : Index := 0) is record F1 : String (1 .. D1); end record;
    type R2 (D2 : Index := 0) is record F2 : R1 (D1 => D2); end record;
    X : R2;
   begin
     for I in 1 .. Index'Last loop
        X := (I, (I, (others => 'A')));
        X.F2.F1 (1) := 'B';
        X.F2.F1:= 'C' & @ (@'First .. @'Last - 1);
        X.F2 := (@.D1,  'D' & @.F1 (@.F1'First .. @.F1'Last - 1));
        X := (D2 => @.D2,
              F2 =>
                (D1 => @.D2,
                 F1 => 'E' & @.F2.F1 (@.F2.F1'First .. @.F2.F1'Last - 1)));
        pragma Assert
          (X.F2.F1 = (if I <= 4
                      then String'("EDCB")(1..I)
                      else "EDCB" & (1 .. I-4 => 'A')));
     end loop;
   end;
end;

Tested on x86_64-pc-linux-gnu, committed on trunk

2017-04-25  Ed Schonberg  <schonberg@adacore.com>

	* sem_ch5.adb (Analyze_Assignment): Reset Full_Analysis flag on
	the first pass over an assignment statement with target names,
	to prevent the generation of subtypes (such as discriminated
	record components)that may carry the target name outside of the
	tree for the assignment. The subtypes will be generated when
	the assignment is reanalyzed in full.
	(Analyze_Target_Name): Handle properly class-wide types.