MersenneTwister output depends on 32/64-bit architecture

[rickhg12hs]

Xref: #5993

$ ./julia -e 'versioninfo()'
Julia Version 0.3.0-prerelease+1794
Commit 4dfbecc* (2014-03-01 09:22 UTC)
Platform Info:
  System: Linux (i686-redhat-linux)
  CPU: Genuine Intel(R) CPU           T2250  @ 1.73GHz
  WORD_SIZE: 32
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm
$ make test-random
    JULIA test/random
     * random
exception on 1: ERROR: test failed: rand(rng) - 0.07749284875576845 < 0.01
 in error at error.jl:21
 in default_handler at test.jl:19
 in do_test at test.jl:39
 in runtests at /usr/local/src/julia/julia/test/testdefs.jl:5
 in anonymous at multi.jl:629
 in run_work_thunk at multi.jl:590
 in remotecall_fetch at multi.jl:663
 in remotecall_fetch at multi.jl:678
 in anonymous at task.jl:1309
while loading random.jl, in expression starting on line 14
ERROR: test failed: rand(rng) - 0.07749284875576845 < 0.01
 in error at error.jl:21
 in default_handler at test.jl:19
 in do_test at test.jl:39
 in runtests at /usr/local/src/julia/julia/test/testdefs.jl:5
 in anonymous at multi.jl:629
 in run_work_thunk at multi.jl:590
 in remotecall_fetch at multi.jl:663
 in remotecall_fetch at multi.jl:678
 in anonymous at task.jl:1309
while loading random.jl, in expression starting on line 14
while loading /usr/local/src/julia/julia/test/runtests.jl, in expression starting on line 34

make[1]: *** [random] Error 1
make: *** [test-random] Error 2
$ ./julia -e '@show rand(MersenneTwister(0))'
rand(MersenneTwister(0)) => 0.8236475079774124
$ 

[jiahao]

Reopening as an issue about MersenneTwister not generating platform-independent sequences.

[andreasnoack]

Regarding @StefanKarpinski's comment, the discussion is here and his fix is here. The fix works as long as you set the seed through srand but not if you set it directly in MersenneTwister. I guess @Stefan's solution can be used also in the MersenneTwister constructor but maybe we should start by discussing how to set the seed when allowing different rngs.

[StefanKarpinski]

I think we should probably make all RNG constructors behave similarly. We can probably generalize the make_seed method to generate different kinds of seed arrays, or we can just reinterpret the ones it produces to feed different RNGs.

[jperla]

Can we get Travis CI to run all of the unit tests on 32-bit machines too?

[staticfloat]

@jperla unfortunately, that's a feature they're still working on.

[jperla]

What do you recommend as the best way for me to test out some changes I'm making? (I don't have a 32-bit machine). Can I force build this as 32-bit to test?

[rickhg12hs]

This seems to be working now.