[Issue 8247] New: Inconsistent behaviour of randomSample depending on whether a random number generator is specified

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=8247

           Summary: Inconsistent behaviour of randomSample depending on
                    whether a random number generator is specified
           Product: D
           Version: D2
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Phobos
        AssignedTo: nobody at puremagic.com
        ReportedBy: joseph.wakeling at webdrake.net


--- Comment #0 from Joseph Rushton Wakeling <joseph.wakeling at webdrake.net> 2012-06-14 12:27:53 PDT ---
Created an attachment (id=1116)
Working minimal example illustrating the inconsistencies described.

The randomSample function in std.random can be called with or without
specifying a random number generator to use.

If no RNG is specified, then each lazy evaluation of the sample evaluates
differently, i.e. if you do

   sample1 = randomSample(iota(0, 100), 5);
   writeln(sample1);
   writeln(sample1);
   writeln(sample1);

you will get 3 different samples.

Conversely, if a random number generator is specified, you will get 3 times the
same result:

   sample2 = randomSample(iota(0, 100), 5, Random(unpredictableSeed));
   writeln(sample2);
   writeln(sample2);
   writeln(sample2);

Note that the seeding of the RNG is important, because if an already-existing
RNG is provided to create multiple different samples, they will evaluate
identically, e.g.

   sample3 = randomSample(iota(0, 100), 5, rndGen);
   writeln(sample3);
   sample4 = randomSample(iota(0, 100), 5, rndGen);
   writeln(sample4);
   sample5 = randomSample(iota(0, 100), 5, rndGen);
   writeln(sample5);

... will produce the same output 3 times.  This happens because the RNG passed
to randomSample is copied rather than used by reference.

These inconsistencies lead to a lot of potential confusion and sources of bugs.
 So, first of all, we need a firm decision on how the lazy evaluation of
RandomSample should behave -- should it

  (1) always evaluate to the same sample, or

  (2) always evaluate to a different sample?

... and depending on the answer, we then need to address how to specify and
seed an RNG for RandomSample.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------