From: "Robert Ramey" <ramey@rrsd.com>
Terje Slettebø wrote:
What's a dimension library? I.e. how might it look like, and be used (preferably a code example)? I've searched the Boost list for "dimension library"/"dimensions library", but your posting is the only one coming up.
This whole thread was started by question about ...\libs\mpl\doc\tutorial\dimensional-analysis.html which uses dimensional analysis as an example of a possible application of mpl. In my view, this tutorial is makes a great case for the utility of such a library and explanation of how it might be implemented
Agreed. A couple of years ago, I played with a dimensional analysis/quantity library, myself, using Loki's typelists to store the dimensions. However, the combination of them were rather involved, and not something like the elegant solution of using mpl::transform (which didn't exist at the time, at least not as a Boost library). It was just the terminology, "dimension library", that I wondered about, and I guessed it was about the dimensions, without attaching any particular meaning to them (which was the case). It has been said in this thread that this could sidestep some of the issues of a unit library, but there still seems to be a number of them for a dimensional analysis library, such as the number of dimensions, and whether they should be fractional or not. As has been noted, a number of dimension/quantity libraries have been developed and proposed over the years, and I guess one useful test of a dimension library would be if it's possible to write a useful quantity library for it (or more). As Andy and others who have been involved in these discussions are all too aware of, there are many tricky issues involved in making a quantity library (such as his PQS library), such as "kg" being the only SI base unit with a prefix, and if you have 1000 * 1 kg, you won't get 1 k(kg), but 1 Mg; the different temperature scales, and perhaps even support for non-SI units, like feet, mile, etc. As well as how to present these quantities, internationalisation, and so on.
For example, if it only takes care of dimensional analysis (i.e. dimension calculations), how many dimensions should it have, if it's independent of units? If the answer is "user-configurable through typelists/vectors", have anyone done any compile-time timing of dimension calulation-heavy code using this approach, compared to a fixed set of dimensions?
I would expect that number of dimensions would be fixed a compile time.
Then it also depends on how that is done. I'm not saying there's a problem; I haven't measured, but I think it should be a consideration. I would think that a simple class template with N integral parameters (or rationals, somehow), where they are combined to form new types using simple +, -, * and /, could be a heck of a lot faster at compile time, than one using mpl::vector, and mpl::transform, and then again I may be wrong, and also the difference may not be enough to be a problem.
Won't this bring us right back to the "how many dimensions"-discussion of a unit/quantity library?
if it were user configurable, I wouldn't think so. I would expect that such a library would have a base implementation along with a higher layer specifying a common set of dimensions. Anyone not happy the the chosen set, he could just use the base layer and use his own set of dimensions.
I wonder what that base functionality would be, when it won't know about the number of dimensions, and it won't know their meaning. It seems a little like "stone soup", :) but I'm willing to be convinced. ("Ok, so we can hardly agree on anything in this library, so let's not make _anything_. Can we agree on that? Yes!") :) Regards, Terje