The class RandomLongSteam is for representing generators of (independent)
uniformly distributed random machine integers in a given range; the range is
specified when creating the generator (and cannot later be changed). See also
RandomBoolStream for information about generating random bools, and
RandomZZStream for information about generating random large integers.
An alternative way of generating random values is to use a RandomSource.
There are three ways of creating a new RandomLongSteam object:
RandomLongStream RLS1(lo,hi); // seeded with 1 by default
RandomLongStream RLS2(lo,hi, n); // seed generator with abs(n)
Each generator will produce values uniformly distributed in the range from
lo to hi (with both extremes included). An ERR::BadArg
exception is thrown if lo > hi; the case lo == hi is allowed.
The third argument is for seeding the generator. If you create more than
one RandomLongStream object with the same seed (and range), they will
each produce exactly the same sequence of values. In particular, to obtain
different results each time a program is run, you can for instance seed the
generator with the system time (e.g. by supplying as argument
time(0)); this is likely desirable unless you're trying to debug a
randomized algorithm.
Once you have created a RandomLongStream you may perform the following
operations on it:
*RLS // get the current value of RLS (as a signed long).
++RLS // advance to next value of RLS.
RLS++ // advance to next value of RLS **INEFFICIENTLY**.
sample(RLS) // advance RLS and then return new value; same as ``*++RLS``
out << RLS // print some information about RLS.
RLS.myIndex() // number of times RLS has been advanced,
same as the number of random values generated.
Note that a RandomLongStream supports input iterator syntax.
You may assign or create copies of RandomLongStream objects; the copies
acquire the complete state of the original, so will go on to produce exactly
the same sequence of bits as the original will produce.
The idea is very simple: use the pseudo random number generator of GMP to
generate a random machine integer in the range 0 to myRange-1 (where
myRange was set in the ctor to be 1+myUpb-myLwb) and then add that
to myLwb. The result is stored in the data member myValue so that
input iterator syntax can be supported.
There are two "non essential" data members: mySeed and myCounter.
I put these in to help any poor blighter who has to debug a randomized
algorithm, and who may want to "fast forward" the RandomLongSteam to
the right place.
The data member myState holds all the state information used by the GMP
generator. Its presence makes the ctors, dtor and assignment messier than
they would have been otherwise.
The advancing and reading member functions (i.e. operator++ and operator*)
are inline for efficiency, as is the sample function.
myGetValue is a little messy because the value generated by the GMP function
gmp_urandomm_ui cannot generate the full range of unsigned long values.
So I have to call gmp_urandomb_ui if the full range is needed.
The data members myLwb, myUpb and myRange are morally constant,
but I cannot make them const because I wanted to allow assignment of
RandomLongStream objects.
It might be neater to put ++myCounter inside myGenValue, though
this would mean that myCounter gets incremented inside the ctor.
Should sample advance before or after getting the value?
Is the information printed by myOutputSelf adequate? Time will tell.
Discarded idea: have a ctor for RLS which take a ref to a RandomSource,
and which uses that to obtain randomness. I discarded the idea because of
the risks of an "invisible external reference" (e.g. a dangling reference,
or problems in multithreaded code). Instead of passing a reference to a
RandomSource to the ctor, you can use the RandomSource to create an initial
seed which is handed to the ctor -- this gives better separation.