Merge branch 'btdhbfsZ' of git+ssh://scm.mulcyber.toulouse.inra.fr//var/lib/gforge/chroot/scmrepos/git/toulbar2/toulbar2 into btdhbfsZ

to be consistent with BTD

2) Add check for logz mode for one assert
3) Start to code BTD-C
WARNING : Variable Elimination results in bugging -zub=2
WARNING : Variable Elimination and BTD in logz mode is bugging
WARNING : Mean Field lower bound does not work with ore than pairwise MRF/CFN

Merge branch 'cpd' of git+ssh://scm.mulcyber.toulouse.inra.fr//var/lib/gforge/chroot/scmrepos/git/toulbar2/toulbar2 into cpd

Conflicts:
	toulbar2/src/tb2enumvar.cpp

Take care of overflow. It happens a lot of time... We may need to go for float128
BTD incoming...

PS: I add a little check in cpd.cpp
if (!isspace(current_char)) rot2aa_var.push_back(current_char);
Please verify if it does change anything just remove the if statement

overflow bit in the cpu). This relies on built-in functions available
in gcc/g++ (versions above 5.0) and clang > 3.5. So it should run on
Linux, MacOSX and Windows (as far as these compilers are used). See eg.
http://clang.llvm.org/docs/LanguageExtensions.html#checked-arithmetic-builtins

 - declaration of bool Add/Sub/Mul(Cost,Cost,*Cost) that add,
   subtract, multiply two Costs, leaves the result in the last Cost
   pointed Cost and return True only in case of overflow.

 - using them in few crucial project/extend but probably more need to
   be replaced.

I see no performance changes using g++ (make test).

It's a good idea to test clang++ also. I tested version 4.0 that
apparently generates faster code than g++ (single test, on a CELAR
instance cpu time went from 17.2" to 16.8") and also plenty of
warnings !

Rather use MAX_COST.

( Working on adding BTD and go to float128 instead of long double)

HBFS counting is ready and working.
For now, it lunch by default HBFS Counting with sigma=0.
Maybe make an option for Z* ? As -zstar ?

Change help for better understanding of logepsilon

Now use -logz -espilon=[integer] for Z*

Add some counting problem in "validation/counting"
Objectif: Test when compiling tb2

Strat coding HBFSZ and MeanField (but not finish), these function does not change anything to tb2 (for now).

it works now

Now you can use -subz=1000 to have the 1000 first energie enumerate by tb2 in LogZ

[bug] When the upper bound is low, pruning occurs while the model is built and the domain sizes change. Use initial domain size for sanity test.

-logz is OK
-ztmp is OK
-zub is OK
-epsilon is OK but Correction from atol to stold
-gum is not working ---> remove the option for now
-subz is not working ---> remove the option for now
--run ---> remove the option

Correction in the output from log10like to Energy only for UAI and LG file

  astyle --style=kr --attach-namespaces --indent=tab --pad-oper  --unpad-paren --pad-header --align-pointer=name --keep-one-line-blocks <files>