Obscure error building tree if fg solvable not set right everywhere

[dehann]

Tree elimination is bad, and therefore get incomplete cliques. This issue was caused by inconsistent solvable flags between variables and factors. In this case fg had 8 variables with 9 factors. All factors were solvable=1 but 3 variables were solvable=0.

The nonsense issue encountered in this case was

ERROR: mcmcIterationIDs -- unaccounted variables
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:33
  [2] mcmcIterationIDs(cliq::IncrementalInference.TreeClique)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:1366
  [3] mcmcIterationIdsOrdered(cliq::IncrementalInference.TreeClique)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:1395
  [4] setCliqMCIDs!(cliq::IncrementalInference.TreeClique)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:1460
  [5] buildCliquePotentials(dfg::LightDFG{SolverParams, DFGVariable, DFGFactor}, bt::MetaBayesTree, cliq::IncrementalInference.TreeClique; solvable::Int64)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:1480
  [6] buildTreeFromOrdering!(dfg::LightDFG{SolverParams, DFGVariable, DFGFactor}, p::Vector{Symbol}; drawbayesnet::Bool, solvable::Int64)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:737
  [7] prepBatchTreeOLD!(dfg::LightDFG{SolverParams, DFGVariable, DFGFactor}; eliminationOrder::Nothing, variableOrder::Nothing, eliminationConstraints::Vector{Symbol}, variableConstraints::Nothing, ordering::Symbol, drawpdf::Bool, show::Bool, filepath::String, viewerapp::String, imgs::Bool)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:793
  [8] #buildTreeReset!#132
    @ ~/.julia/dev/IncrementalInference/src/JunctionTree.jl:880 [inlined]
  [9] solveTree!(dfgl::LightDFG{SolverParams, DFGVariable, DFGFactor}, oldtree::MetaBayesTree; timeout::Nothing, storeOld::Bool, verbose::Bool, verbosefid::Base.TTY, delaycliqs::Vector{Symbol}, recordcliqs::Vector{Symbol}, limititercliqs::Vector{Pair{Symbol, Int64}}, injectDelayBefore::Nothing, skipcliqids::Vector{Symbol}, eliminationOrder::Nothing, variableOrder::Nothing, eliminationConstraints::Vector{Symbol}, variableConstraints::Nothing, smtasks::Vector{Task}, dotreedraw::Vector{Int64}, runtaskmonitor::Bool, algorithm::Symbol, multithread::Bool)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/SolverAPI.jl:339
 [10] solveTree!(dfgl::LightDFG{SolverParams, DFGVariable, DFGFactor}, oldtree::MetaBayesTree) (repeats 2 times)
    @ IncrementalInference ~/.julia/dev/IncrementalInference/src/SolverAPI.jl:280
 [11] top-level scope
    @ REPL[13]:1

---

At first glance it looks hard to just define a simple test that works in all cases. We definitely need the case of solvable=0 to work, so just looking at the variable list is not good enough.