>  Hi David,
>  
>  >> Hmm... I agree that this should be changed so that we avoid
>  >> side-effects. What about if we changed the signature to:
>  >>
>  >>   node-set math:random(number)
>  >>
>  >> and have math:random() return a number of <number> elements whose
>  >> values are random numbers between 0 and 1 (the number indicated by the
>  >> argument). The random numbers that are returned should be drawn from a
>  >> consistent set, such that if you do:
>  >>
>  >>   math:random(20)
>  >>
>  >> then the first 10 of the numbers are exactly the same as the numbers
>  >> that you would get if you did math:random(10). (Alternatively we could
>  >> just say that two calls to math:random() with the same argument must
>  >> result in the same set of random numbers.)
>  >>
>  >> The processor should, I think, produce a different set of random
>  >> numbers on each run of the particular stylesheet.
>  >
>  > Sounds good, but how about adding an optional "seed" argument? That
>  > would make it easier for users to get more than one set of return
>  > values.
>  
>  OK. We need to make it clear, then, whether this second seed argument
>  is actually used as the seed of the random sequence (such that if you
>  use the same seed in different transformations, you always get the
>  same random sequence), or if it's combined with something more random
>  (e.g. the date) by the processor in order to create a random sequence
>  that changes each time the transformation is run.
>  
>  I don't have any particularly strong feelings about it, but I imagine
>  that people who want to use random sequences will want those sequences
>  to be different each time the transformation is run.

Seeded RNGs are very widely used in simulation, and in testing scenarios, 
among other cases.

+1 for optional seed argument, where that seed is used independently of any 
other factors.


>  > Dmitre Novatchev also has a great implementation in his FXSLT
>  > library, so we could see what he's doing as well.
>  
>  Dimitre's templates generate a number of random numbers, with an
>  optional seed and an optional start and end scale. Given a generator
>  of random numbers between 0 and 1, I think a scaler can be fitted on
>  top easily enough.
>  
>  The other one of Dimitre's templates that's really cool takes a list
>  of value/probability pairs and generates a sequence of random numbers
>  that follows that distribution. Again, I think that this can be
>  implemented on top of a basic random() function, so I think that
>  math:random() is what we should concentrate on for now.

Agreed.  There are all sorts of distributions we can implement once we have a 
basic RNG.


-- 
Uche Ogbuji                           Fourthought, Inc.
uche.ogbuji@[...].com           http://fourthought.com
http://4Suite.org                     http://uche.ogbuji.net
Track chair, XML/Web Services One (San Jose, Boston): http://www.xmlconference.com/
RDF Query using Versa - http://www-106.ibm.com/developerworks/xml/library/x-think10/index.ht
ml
WSDL and the Wild, Wild West - http://adtmag.com/article.asp?id=6004
XML, The Model Driven Architecture, and RDF @ XML Europe - http://www.xmleurope.com/2002/ktt
rack.asp#themodel


_______________________________________________
exslt mailing list
list@[...].org
http://www.exslt.org/list