$Id: Notes,v 1.4 1997/10/20 19:38:59 jt Exp $

20 Oct 97

Just had another fail-to-compile, because it couldn't find sources
that it hadn't built yet.  I think the Makefile structure is way too
complicated, and will suggest an alternative way too complicated
approach:

Put the source building makefile stuff into the file GNUmakefile,
and build it with something like 'make sources'.  Then cvs commit
all those sources every time one of the template changes.  That's
probably not too often.  It is possible with this approach that
we can even avoid the Makefile.ami and just work with Makefile.am


==================================================================
19 Nov 96

1/ could simplify the basic tasks required of the writer of
   new random number generators by
   a/ rewriting zuf() using integers, and then having the
      uniform() function defined in the xxx-state.xc file
   b/ using an if(inityet) scheme to ensure that calls to
      gsl_ran_randomInt() first call the seed() function with
      a seed of 1.  It is hard to imagine that this will slow
      down the speed at all.

2/ some routines, like 'fortune' want to use a different seed
   each time it is run.  we could provide a scheme to set the
   seed with the clock, or the pid, in order to achieve this 
   effect more easily.

3/ **FIXED**
   by the way, i had trouble linking against libgslrandom.a
   should make sure that it includes the appropriate default
   generator

4/ what about efficient ways of generating random integers
   from 0 to N-1, where N is something like 7?  is there a
   better way than floor(7*uniform()) ?  Note that this
   scheme requires the uniform() is always strictly less
   than 1.

4a/similarly, what about efficient random bits?

5/ should have a standard policy of whether gsl_ran_random()
   ever returns a zero.  probably safest to assume that it
   can, and in the special cases where it cannot, return one
   less than what would have been returned, and also decrement
   max by one in that case.

6/ is there any good reason gsl_ran_max() is a double even
   though gsl_ran_random() is an unsigned long ??
   i was thinking about max=2^32 even though true max is
   2^32-1.  on the other hand, if you want the range of 
   uniform's to be [0,1), then you want to divide not by
   max but by max+1, in which case you'll need to convert
   to double anyway if max=2^32-1.


1/ could simplify the basic tasks required of the writer of
   new random number generators by
   a/ rewriting zuf() using integers, and then having the
      uniform() function defined in the xxx-state.xc file
   b/ using an if(inityet) scheme to ensure that calls to
      gsl_ran_randomInt() first call the seed() function with
      a seed of 1.  It is hard to imagine that this will slow
      down the speed at all.

2/ some routines, like 'fortune' want to use a different seed
   each time it is run.  we could provide a scheme to set the
   seed with the clock, or the pid, in order to achieve this 
   effect more easily.

3/ **FIXED**
   by the way, i had trouble linking against libgslrandom.a
   should make sure that it includes the appropriate default
   generator

4/ what about efficient ways of generating random integers
   from 0 to N-1, where N is something like 7?  is there a
   better way than floor(7*uniform()) ?  Note that this
   scheme requires the uniform() is always strictly less
   than 1.

4a/similarly, what about efficient random bits?

5/ should have a standard policy of whether gsl_ran_random()
   ever returns a zero.  probably safest to assume that it
   can, and in the special cases where it cannot, return one
   less than what would have been returned, and also decrement
   max by one in that case.

6/ is there any good reason gsl_ran_max() is a double even
   though gsl_ran_random() is an unsigned long ??
   i was thinking about max=2^32 even though true max is
   2^32-1.  on the other hand, if you want the range of 
   uniform's to be [0,1), then you want to divide not by
   max but by max+1, in which case you'll need to convert
   to double anyway if max=2^32-1.