Change in behavior of & operator between 83 and 9x?

[graham@clsi.COM (Paul Graham)]

If an array type A has component type B which is itself an array type, is it
possible to concatenate two arrays of type B to produce an array of type A?
For instance:

    package foo is
	type my_string is array(1 .. 3) of character;
	type my_string_vector is array(positive range <>) of my_string;
	c1 : my_string_vector := "abc" & "def";
    end;

In this example, I want the value of the concanenation to be equivalent to

	("abc", "def")	-- my_string_vector(1 to 2)

But it can also be interpreted as

	"abcdef"	-- string(1 to 6), my_string(1 to 6), ???

which does not match the type of c1.  

The 83 RM (4.5.3(4)) states "if both operands are one-dimensional arrays,
the result of the concatenation is a one-dimensional array whose length is
the sum of the lengths of the operands ...".  Clearly the concatenation of
two strings is always a string, and never an array of strings.

But the 9x RM defines the result of a concatenation in terms of its result
type.  In the above example the result type is known to be my_string_vector.
Thus the "&" should be "&"(x, y : my_string) return my_string_vector.

Is my understanding wrong, or is this an incompatibility between the 83 and
9x versions of Ada?

Paul
--
Paul Graham 	graham@compass-da.com	Compass Design Automation, Inc.
(speaking only for myself)	"Cekoslovakyalilastiramadiklarimizdanmissiniz."