Re: [boost] quantity libraries

From: George A. Heintzelman (georgeh_at_[hidden])
Date: 2001-08-30 10:52:10

> >   From: Anatoli Tubman <anatoli_at_[hidden]>
> >
> >Oh come on. The question is not what is more fundamental, but what is
> >more useful. Which makes it much trickier, because usefulness is
> >relative. Meter is also a derived unit. This does not make it
> >any less useful.
> 
> The metre is a base unit not a derived unit.

Actually I believe the meter is now defined in terms of a number of 
wavelengths of a particular frequency of light, and the second in terms 
of the speed of light (or maybe it's the second that is defined in 
terms of a number of vibrations and the meter in terms of the speed of 
light, I forget); the Coulomb is defined in terms of a quantity of 
electrons. I believe the standard kilogram is the only 'base unit' 
which is defined in terms of the qualities of a particular object 
anymore. Does that mean we should have all the units stored in those 
physical quantities? I think not.

Personally, I think storing all quantities in the same way is asking 
for trouble with dynamic range. For example, capacitance is measured 
for typical electronic components in units of picoFarads. If you store 
it in units of Farads (which is what you get with 'whole' SI units), 
you've just chopped 12 orders of magnitude off the upper range of what 
you can represent with a particular floating-point number. If you're 
using an underlying fixed-point type, the problem is even worse, and I 
think that that should be a possible to do with a units library. 
Likewise, in high-energy physics, one often works in a system of units 
where h-bar = c = e = G = 1. The conversion factors between this system 
and SI units can be very large.

I think a units library should allow the user to specify both the 
dimensionality of units of a particular value *and* a base system to 
use. The set of base systems would be extendable by the user, and would 
specify the value of its base constants in terms of the SI units. Then 
when operating on units with the same base system, no run-time overhead 
is incurred; when operating on units with different base systems, the 
compiler looks up the conversion constants and hopefully can reduce the 
conversion to a single multiplication at compile time, invisible to the 
user. Of course, a reasonable default for the base system to use would 
be SI meters-kilograms-seconds.

George Heintzelman
georgeh_at_[hidden]