Alfredo, Thank you for your answers. To be more pragmatic here is an archetype of function that I have in my problem : any_quantity<force, length> myfunction(energy_q arg1) { if (arg1 > 1.0 * jouls) return 2.2 * newtons; else return 3.3 * meters; } That may sound weird but it's a real case. I'm solving a problem with two quantities that have unrelated units and one quantiy or the other is choosen at runtime depending on some state variable values. Ultimately those quantities are stored in a single matrix. For now I break the dimensionnal analysis at this step using plain doubles.
I think this goes against the library design which is to be zero overhead. (But it may be useful some times although my pattern above is a good competition against your requested feature)
What I was working on is something like:
any_quantity<force_dimension>
that can hold any quantity with force dimension, this makes more sense to me because at least I know the all force quantities are potentially convertible to one another, regardless of the unit system. In my proposal there is no runtime checking at all, but only runtime conversions.
Runtime conversion are also something I am interrested. Are you following the lead of : http://www.boost.org/doc/libs/1_46_0/doc/html/boost_units/Examples.html#boos... I think that we could layer things this way : LAYER 3: any_quantity<heterogeneous_units> runtime dimensional analysis and conversion LAYER 2: any_quantity<homogeneous_dimension_units> compile time dimensional analysis and runtime conversion LAYER 1: quantity<unit> the actual library Each extension layers could be build on top of the others. For performance reason, the runtime dimensional analysis could be switched off for release build.
In any case I would like to know of any solution you can come up with.
Of course.