Type error in coeff using Symbolics 5.5.1

[Johan-Gronqvist]

I see an error when running

@syms a b c d

expr = 2.0*(a+b*c^2-10d)

Symbolics.coeff(expr, b)

The error message is

TypeError: non-boolean (SymbolicUtils.BasicSymbolic{Bool}) used in boolean context

Stacktrace:
 [1] _all(f::typeof(iszero), itr::Vector{Any}, #unused#::Colon)
   @ Base ./reduce.jl:1287
 [2] all(f::Function, a::Vector{Any}; dims::Function)
   @ Base ./reducedim.jl:1007
 [3] all(f::Function, a::Vector{Any})
   @ Base ./reducedim.jl:1007
 [4] coeff(p::SymbolicUtils.BasicSymbolic{Number}, sym::SymbolicUtils.BasicSymbolic{Number})
   @ Symbolics ~/.julia/packages/Symbolics/BQlmn/src/utils.jl:321
 [5] top-level scope
   @ In[40]:5

I use Julia 1.9.0 with the following output of Pkg.status()

  [8e7c35d0] BlockArrays v0.16.36
  [a6e380b2] ControlSystems v1.8.0
  [587475ba] Flux v0.14.2
  [b99e6be6] Hypatia v0.7.3
  [d1acc4aa] IntervalArithmetic v0.20.9
  [4076af6c] JuMP v1.13.0
  [d8e11817] MLStyle v0.4.17
  [b8f27783] MathOptInterface v1.18.0
  [1ec41992] MosekTools v0.15.0
  [872c559c] NNlib v0.9.4
  [bac558e1] OrderedCollections v1.6.2
  [91a5bcdd] Plots v1.38.17
  [f27b6e38] Polynomials v3.2.13
  [92933f4c] ProgressMeter v1.7.2
  [295af30f] Revise v3.5.3
  [0c5d862f] Symbolics v5.5.1
  [e88e6eb3] Zygote v0.6.63
  [37e2e46d] LinearAlgebra
  [8dfed614] Test

[Johan-Gronqvist]

I am now using Symbolics v4.14.0 instead, which works for me.

[shashi]

@YingboMa

[Johan-Gronqvist]

The problem seems to appear if the expression is a multiplication and one of the terms contain coefficients that are themselves symbolic expressions. A check if those expressions are zero fails.

Removing the following if-clause in utils.jl and just keeping the else-clause seems to fix the problem.

        if all(iszero, coeffs)
            return 0
        else
            @views prod(Iterators.flatten((coeffs[findall(!iszero, coeffs)], args[findall(iszero, coeffs)])))
        end

A better fix may be to replace the call to iszero with a call to isequal that checks equality with zero. The call to iszero appears to call a method from base that uses ==, as warned against in the FAQ https://symbolics.juliasymbolics.org/stable/manual/faq/#Equality-and-set-membership-tests