[boost] Dimensional Analysis Feature Summary [was Dimensional Analysis Interest?]

From: David Richardson (dgr24_at_[hidden])
Date: 2003-10-15 08:55:04

There is a new dimensioned.tar.bz2 on yahoo.  It now works with 
gcc 3.3.1 and icc 7.0 (both on Linux), adds a sqrt function, and 
fixes a bug involving fractional powers not being reduced correctly.

Hopefully the changes to get it to compile under icc 7.0 will let 
it compile on more compilers.  Could people with access to other 
compilers please check?

Summary of Current Features:

* Units defined in terms of Mass, Length, Time
* Handles fractional powers for units
* Provides typedefs for commonly used units
* Provides mechanism for users to make new units
* Has functions for exponentiation (sci::pow) and square roots 
(sci::sqrt)
* Arithmetic operators +, +=, -, -=, *, *=, /, =
* Comparison operators ==, !=, <, <=, >, >=
* Numeric type may be specified by user
* May be extended to use new numeric types by creation of a traits 
class
* Absolutely zero runtime cost for validating dimensional correctness
* All invalid expressions are detected at compile time and prevent 
compilation

Rough Edges:

* Does not do type promotion when underlying numeric types are 
different.  Does anyone know of existing code to do this other 
then what is in Generative Programming/GMCL?

* Number of dimensions hard coded to three.  I'm working learning 
the MPL so I can fix this.

* Dimension semantics are hard coded to Mass, Length, Time.  How 
many people will want to use this for situation where they need 
different semantics?

* sci::pow calling syntax is a bit awkward.  Does anyone have any 
suggestions for improving this?

Possible Features:

* Formated I/O
* Automatic Conversion
* Multiple Units Systems (e.g. English, Metric)
* Uncertainty Propagation.  This is easily added by using an 
interval as the numeric type (and boost already has an interval 
class).

Are there any other features people are interested in?  One design 
goal that I have is making it so that people can always use the 
dimensional analysis part of the library at zero runtime cost. 
There are many domains (e.g. molecular simulation) where any 
runtime penalty makes the library useless.  If things like 
conversion and I/O formatting get implemented, I think it will 
need to be on top of the dimensional correctness class.  That way 
people can not use the I/O and conversion functionality if they do 
not need it.

Dave