For an overview of how to use the phys_quan library,
see the file main.cpp.  The comments there explain
how to use it.

Design criteria:

- Basic types (length_t, mass_t, speed_t, etc.) must be
provided by the library.

- Expressions with phys_quans must be as simple and clear
as any other expressions, i.e. they must use normal
operator overloading.

- It must be syntactically impossible to write a legal
expression that assigns a value to a physical quantity
without specifying the units.  E.g. "width = 3 meters" and
"width = 3 feet" are ok, but "width = 3" must be caught
as an error by the compiler.

- Any attempt to assign (or pass as an argument) a phys_quan
value to a phys_quan variable of different dimension must
be caught as a compile-time error.

- Conversions to and from any compatible units (i.e. of the
same dimension, such as meters and feet) at any time must be
possible and use a clear, simple syntax.

- Both SI and Imperial units must be available. There
should be no problems created if the units are mixed.

- phys_quans should not be limited to m-kg-s,
but must also be able to handle A-K-mol-cd.

- It should be possible to define physical constants
(like the speed of light) clearly and efficiently.

- It must be possible for the user to define new types
(e.g. density_t), units (e.g. furlong), and constants.
These user-defined entities must use the same syntax and
be just as efficient as those supplied by the library.

- Any calculations that can be done at compile time should be.
This implies that the run-time overhead of using the phys_quan
library (as compared to just using raw numbers) will normally
be zero.

The result of expressions should be as follows.  (phys_quan1
and phys_quan2 denote physical quantities of different dimensions).
"ERROR" means caught at compile-time

- phys_quan = number -> ERROR
- number = phys_quan -> ERROR
- phys_quan = phys_quan -> phys_quan &
- phys_quan1 = phys_quan2 -> ERROR

- phys_quan + number -> ERROR
- number + phys_quan -> ERROR
- phys_quan + phys_quan -> phys_quan
- phys_quan1 + phys_quan2 -> ERROR

- phys_quan - number -> ERROR
- number - phys_quan -> ERROR
- phys_quan - phys_quan -> phys_quan
- phys_quan1 - phys_quan2 -> ERROR

- phys_quan * number -> phys_quan
- number * phys_quan -> phys_quan
- phys_quan1 * phys_quan2 -> phys_quan3 (or number)

- phys_quan / number -> phys-quan
- number / phys_quan1 -> phys_quan2
- phys_quan / phys_quan -> number
- phys_quan1 / phys_quan2 -> phys_quan3

- phys_quan relop number -> ERROR
- number relop phys_quan -> ERROR
- phys_quan relop phys_quan -> bool
- phys_quan1 relop phys_quan2 -> ERROR

- ostream << phys_quan -> ostream &

The user must *not* have to specify the result type
of such expressions; it must be deduced by the
compiler automatically.