[racket] Units/measures library

Neil Van Dyke wrote at 11/17/2011 05:49 PM:
> Shriram Krishnamurthi wrote at 11/17/2011 09:43 AM:
>> DSSSL (-:
>>
>> (Since I believe Bigloo and/or Gambit implemented DSSSL, lurking in
>> their implementations is code that does this...)
>
> Also, the Curl programming language had units of measurement built-in:
> http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.131.6818
>
>> 2.4.1.3 Quantities
>>
>> Presentation-oriented applications frequently deal with
>> measurements of space and time. Curl accordingly includes
>> “quantity types,” which are numeric data types with units, such as
>> Distance and Time. These types are distinct from the
>> conventional dimensionless numeric types such as float and
>> double. There are also quantity literals, such as 2.54cm,
>> 72pt, 50ms, 1week, and even 800dpi and 32(ft/(s^2)).
>> By using quantities, programmers can let Curl handle conversions
>> between units and can write more descriptive (and checkable)
>> interfaces to graphical APIs. There is also a pixel-denominated
>> unit px, which can be converted to a Distance by dividing by
>> the spatial resolution of a display device. Quantity types are not
>> found in other languages for creating Web content, but they are
>> used pervasively in the Curl GUI APIs and are tremendously
>> useful. in application programs.
>
> It seems easy for someone to add to Racket as a library, so long as 
> the library user doesn't mind having to explicitly ``cast'' these 
> objects to normal Racket numbers when they want to interoperate with 
> Racket libraries that only work with numbers. Maybe use Typed Racket 
> or contracts info to add in the type coercions manually, or maybe 
> having to do the explicit casts is actually good programming practice.

Typo. I meant not "manually", but "automatically across module boundaries".

>
> I am doing something like this with an (as yet unreleased) time 
> library, and have thought of doing it for financial apps as well.
>
> One thing I haven't decided is whether and how to represent precision 
> (or resolution) information, and what to do with that information. For 
> a simple example, what is the answer to the difference "(2 days) - (3 
> minutes)"? I'm wondering whether the unit of measure alone can 
> indicate the precision for 99.9% of practical purposes, or we have to 
> be able to support more flexible specification of precision (e.g., "(2 
> days) - (3 minutes to a precision of 15 seconds)").
>