[BUG] TaskFailedException due to domain error

[MilesCranmer]

It seems like there are still some domain errors in very rare edge cases (~one in a billion!). This is the traceback:

    nested task error: TaskFailedException
    Stacktrace:
     [1] wait
       @ ./task.jl:334 [inlined]
     [2] fetch
       @ ./task.jl:349 [inlined]
     [3] (::SymbolicRegression.var"#60#98"{Vector{Vector{Task}}, Int64, Int64})()
       @ SymbolicRegression ./task.jl:423
    
        nested task error: DomainError with Inf:
        cos(x) is only defined for finite x.
        Stacktrace:
          [1] cos_domain_error(x::Float32)
            @ Base.Math ./special/trig.jl:97
          [2] cos(x::Float32)
            @ Base.Math ./special/trig.jl:108
          [3] deg1_l1_ll0_eval(tree::Node, cX::Matrix{Float32}, #unused#::Val{1}, #unused#::Val{2}, options::Options{Tuple{typeof(*), typeof(/), typeof(+), typeof(-)}, Tuple{typeof(sin), typeof(cos), typeof(exp), typeof(log_abs)}, L2DistLoss})
            @ SymbolicRegression.../EvaluateEquation.jl ~/miniconda3/envs/pysr/share/julia/packages/SymbolicRegression/Z2pqJ/src/EvaluateEquation.jl:175
          [4] evalTreeArray(tree::Node, cX::Matrix{Float32}, options::Options{Tuple{typeof(*), typeof(/), typeof(+), typeof(-)}, Tuple{typeof(sin), typeof(cos), typeof(exp), typeof(log_abs)}, L2DistLoss})
            @ SymbolicRegression.../EvaluateEquation.jl ~/miniconda3/envs/pysr/share/julia/packages/SymbolicRegression/Z2pqJ/src/EvaluateEquation.jl:31
          [5] deg2_eval(tree::Node, cX::Matrix{Float32}, #unused#::Val{4}, options::Options{Tuple{typeof(*), typeof(/), typeof(+), typeof(-)}, Tuple{typeof(sin), typeof(cos), typeof(exp), typeof(log_abs)}, L2DistLoss})

In principle these should be caught before evaluating the operator (and inf loss returned), but it looks like there are some cases where it is not caught. Will need to debug on the Julia side.

[MilesCranmer]

Cross posted to MilesCranmer/SymbolicRegression.jl#71 - will focus on the issue there since the bug is on the Julia side.

[MilesCranmer]

This should be fixed with v0.7.9