Re: SIunits and quantity library concerns

From: Eric Ford (eford_at_[hidden])
Date: 2001-10-04 19:41:24

Glad to see all the discussion about quantities, dimensions, units,
and qualifiers.  I like the idea of a units library like SIunits, but
with a qualifier library on top.  I whipped up a beast and uploaded it
to the qualified_unit folder.  I hope people will take a look.

Basically, it provides strong typing for qualified units.  By default
there's strong typing of qualifiers.  If a user wants to allow adding
apples and oranges to give fruits, then they can specialize the the
promote_unit_qualifier traits-like class.  They still can't assign
apples to oranges (or apples to fruits).  However, iff the units
library allowed assigning a mass in kg to a weight in pounds (one unit
library might, another might not), then they could assign a
qualified_unit<pounds,apples_qualifier> to a
qualified_unit<kg,apples_qualifier>.  

The test program includes it's own single-unit unit library, mainly
just to perform the most basic of tests.   If you provide a real units
library, you'll need to change the typedef for type in
promote_unit<T1,T2,multiplies_op> and promote_unit<T1,T2,divides_op>
to give the type that that library returns when it multiplies/divides
quantities with units T1 and T2.  Other than that, I hope it will
work, but I can't get SIunits to compile, so I there are probably
things I did wrong.

It still has a ways to go.  For example, presently, it's not smart
about "canceling" qualifiers.  E.g. if you divide 2 liters of coke by
2 liters of coke, it would give 1 [unitless] but with a unit qualifier
of (coke/coke) .  

Also, it doesn't have a way to deal with power of qualifiers.  In
principle this is probably necessary, although it can certainly wait. 
Both these shortcomings would probably need to be addressed by using a
different data structure (currently a binary tree) to hold the
qualifier information. 

Comments?