[boost] Re: Re: Dimensional Analysis

From: Andy Little (andy_at_[hidden])
Date: 2005-02-01 18:23:03

"Robert Ramey" <ramey_at_[hidden]> wrote in message
news:ctojd3$obh$1_at_sea.gmane.org...

> So lets make a minimalist dimensional analysis library whose function is
> to check arithmetic operations at compile time for consistency.

I will attempt to write a Concept definition keeping as close as possible to
David Abrahams example. ie with an interface based in mpl
Of course there are some subtleties that will raise eyebrows, but there aint
much I can do except explain the rationale. For instance I would treat a
*raw* set of dimensional-exponents as having no particular name, because one
set can represent more than one quantity. (e.g torque and energy both have
the same set of dimensional-exponents).  the Units library must inform the
dimensional analysis library regarding this to try to forestall backward
compatibility issues.

FWIW I'd put this in boost::pqs
// eg
typedef boost::pqs::abstract_quantity<
     boost::mpl::vector<x,y,z>,  // set of dimensional exponents
     boost::pqs::energy_tag,  //    to distinguish from torque , also
anonymous_tag for results of operations
 > energy;

// Not
 typedef boost::mpl::vector<x,y,z> energy ;

(This is basically using the semantics and interface of pqs but implementing
in boost style:
http://www.servocomm.freeserve.co.uk/Cpp/physical_quantity/operations-semantics.html
)

> I'm not even convinced that fractional dimensions are needed.

It is possible they might occur in temporaries results of calculations in
some cases. Its marginal but there are some eg opamp noise parameter that
are specified as nanovolt per sqrt Hz.

>  If the really are, maybe
> the library can take a template paramter the specifies either  an integer
> or compile time rational type.

AFAIK you may even be able to mix with boost::mpl::numeric_cast

  This would cleave off another tricky
> piece -
> compile time rationals.

There are already several compile time rationals about (One I used is based
on Matthias Schabels code).It should not be too hard to convert this to use
the mpl style. This should enable Concept compatibility with whatever else
might emerge as rational or fraction (e.g from Cromwell Enage). IOW A
dimensional-exponent is a Concept essentially with an  interface eg using
mpl::plus etc.

docs Might look like so:
   Concept DimensionalExponent // eg int_ , rational_c
wish
    implements :
    mpl::plus<Dimensional ExponentL, Dimensional ExponentR>  qa*qb
    mpl::minus<Dimensional ExponentL, Dimensional ExponentR>  qa/qb
    mpl::multiplies<DimensionalExponent, Rational >  pow<Rational>(q)
    mpl::equal<DimensionalExponentL,DimensionalExponentR>  // dimensional
analysis operator

   Concept DimensionalExponents // eg mpl::vector
    implements :
    mpl::plus<Dimensional ExponentsL, Dimensional ExponentsR>  qa*qb
    mpl::minus<Dimensional ExponentsL, Dimensional ExponentsR>  qa/qb
    mpl::multiplies<DimensionalExponents, Rational >  pow<Rational>(q)
    mpl::equal<DimensionalExponentsL,DimensionalExponentsR>  // dimensional
analysis operator

   Concept AbstractQuantity //  eg  struct abstract_quantity
    implements :
    mpl::plus<AbstractQuantityL, AbstractQuantityR>  qa*qb
    mpl::minus<AbstractQuantityL, AbstractQuantityR>  qa/qb
    mpl::multiplies<AbstractQuantity, Rational >  pow<Rational>(q)
    mpl::equal<AbstractQuantityL,AbstractQuantityR>  // dimensional
analysis operator

    negate, reciprocal also for convenience.

regards
Andy Little