How to model definitions and declarations of a programming language?

[Rikard Land]

I try to model a data definition language in XML. It can be seen as C
without any executable statements other than variable assignments. I
want to allow for:
(1) type declarations ("structs" in C), which can be nested, and where
each member can be (optionally) associated with
a) a comment (as a help to the user who will enter data) and
b) a unit such as kg, cm, etc.
(2)variable declarations of ordinary types such as float, int, string,
etc., as well as the types defined in (1), and
(3)assignments of values to variables.

Examples (in C-like syntax) would be:
(1)Type definitions:
struct Point {
float x, comment("X coordinate"), unit(m);
float y, comment("Y coordinate"), unit(m);
}
struct Z {
int a, comment("The amount of...");
float b, comment("The bla bla..."), unit(kg);
Point point, comment("The point ...");
}

(2)Variable declarations:
Point p;
Z z;

(3)Assignments
p.x = 1.1;
p.y = 2.2;
z.a = 1
z.b = 3.3
z.point.x = 4.4
z.point.y = 5.5

I want to keep these in separate files to have a clean separation of
these kinds/levels of data. (It must e.g. be possible to declare
several instances of the same struct.) There will also be arrays.
However, the problem is not to model this in XML syntax (that is
straightforward ), but my question is rather how to model the
interrelationsh ips between the files? At first sight, I thought it was
obvious that (1) should be defined as a schema for (3), but there are
some challenges that I cannot resolve:
* I want explicit variable declarations, mainly so that several
instances of the same type can be defined. This gives by hand the
inclusion of (2).
* I want to use the comment and unit in (1) for e.g. helping the user
in a GUI. Isn't a schema only a template for an XML file? Or can I put
things in the schema (1) that are needed there, but _shall not_ be
present in the XML file (3)?
* I want to resolve the names in (2), so that only declarations of
types defined in (1) (or basic types, float etc) are valid.
* I also want to make sure the assignments are made correctly, so that
e.g. "z.point.x = 4.4" is valid because z is of type Z which contains
a member point of type Point, which in its turn contain a float named
"x".

I realize that I can do these things programmaticall y (people have
been building compilers for some years now), but I first want to look
for solutions utilizing the power of XML. I cannot find anywhere on
the net someone who has attempted this – maybe I have used the wrong
search terms or it is just a stupid thing to attempt...

Any help in solving any or part of the problems is highly appreciated.
Regards,
Rikard Land