Can't register erf as :erf

[mewilhel]

I was trying to register erf as a work around for issue #1803 and ran into the following behavior. If I try to register the function using :erf with the below script then I receive an error.

using SpecialFunctions
using JuMP

model = Model()
@variable(model, x)

JuMP.register(model, :erf, 1, x -> erf(x), autodiff = true)
@NLconstraint(model, c1, erf(x) <= 0.0)

d = NLPEvaluator(model)
MOI.initialize(d, Symbol[:Grad])

out = zeros(1)
values = Float64[2.0]
MOI.eval_constraint(d, out, values)

Namely, this error:

ERROR: LoadError: UndefVarError: erf not defined
Stacktrace:
 [1] eval_univariate at C:\Users\wilhe\.julia\packages\JuMP\e0Uc2\src\_Derivatives\forward.jl:411 [inlined]
 [2] forward_eval(::Array{Float64,1}, ::Array{Float64,1}, ::Array{JuMP._Derivatives.NodeData,1}, ::SparseArrays.SparseMatrixCSC{Bool,Int64}, ::Array{Float64,1}, ::Array{Float64,1}, ::Array{Float64,1}, ::Array{Float64,1}, 
::Array{Float64,1}, ::Array{Float64,1}, ::JuMP._Derivatives.UserOperatorRegistry) at C:\Users\wilhe\.julia\packages\JuMP\e0Uc2\src\_Derivatives\forward.jl:188
 [3] _forward_eval_all(::NLPEvaluator, ::Array{Float64,1}) at C:\Users\wilhe\.julia\packages\JuMP\e0Uc2\src\nlp.jl:503
 [4] macro expansion at C:\Users\wilhe\.julia\packages\JuMP\e0Uc2\src\nlp.jl:571 [inlined]
 [5] macro expansion at .\timing.jl:233 [inlined]
 [6] eval_constraint(::NLPEvaluator, ::Array{Float64,1}, ::Array{Float64,1}) at C:\Users\wilhe\.julia\packages\JuMP\e0Uc2\src\nlp.jl:569
 [7] top-level scope at C:\Users\wilhe\Dropbox\My PC (DESKTOP-P6322LG)\Desktop\Specials_reg.jl:16
 [8] include_string(::Function, ::Module, ::String, ::String) at .\loading.jl:1088
in expression starting at C:\Users\wilhe\Dropbox\My PC (DESKTOP-P6322LG)\Desktop\Specials_reg.jl:16

If I register this function using a different name, like below, everything works out fine.

using SpecialFunctions
using JuMP

model = Model()
@variable(model, x)

JuMP.register(model, :random_name, 1, x -> erf(x), autodiff = true)
@NLconstraint(model, c1, random_name(x) <= 0.0)

d = NLPEvaluator(model)
MOI.initialize(d, Symbol[:Grad])

out = zeros(1)
values = Float64[2.0]
MOI.eval_constraint(d, out, values)

Version and package info below:

julia> versioninfo()
Julia Version 1.5.1
Commit 697e782ab8 (2020-08-25 20:08 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i9-10900 CPU @ 2.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-9.0.1 (ORCJIT, skylake)
Environment:
  JULIA_EDITOR = "C:\Users\wilhe\AppData\Local\atom\app-1.52.0\atom.exe"  -a
  JULIA_NUM_THREADS = 10

 [c52e3926] Atom v0.12.24
  [6e4b80f9] BenchmarkTools v0.5.0
  [336ed68f] CSV v0.7.7
  [7057c7e9] Cassette v0.3.3
  [e2554f3b] Clp v0.8.1
  [a93c6f00] DataFrames v0.21.7
  [864edb3b] DataStructures v0.17.20
  [163ba53b] DiffResults v1.0.2
  [0c46a032] DifferentialEquations v6.15.0
  [bb8be931] EAGO v0.5.0 `C:\Users\wilhe\Desktop\Package Development\EAGO.jl`
  [587475ba] Flux v0.11.1
  [f6369f11] ForwardDiff v0.10.12
  [4b11ee91] Gaston v1.0.2
  [7073ff75] IJulia v1.22.0
  [d1acc4aa] IntervalArithmetic v0.17.5
  [d2bf35a9] IntervalRootFinding v0.5.4
  [b6b21f68] Ipopt v0.6.3
  [4076af6c] JuMP v0.21.4
  [e5e0dc1b] Juno v0.8.4
  [b964fa9f] LaTeXStrings v1.2.0
  [33e6dc65] MKL v0.2.0 `https://github.com/JuliaComputing/MKL.jl#master`
  [b8f27783] MathOptInterface v0.9.17
  [53c679d3] McCormick v0.9.0 `C:\Users\wilhe\Desktop\Package Development\McCormick.jl`
  [2774e3e8] NLsolve v4.4.1
  [91a5bcdd] Plots v1.6.10
  [d330b81b] PyPlot v2.9.0
  [189a3867] Reexport v0.2.0
  [295af30f] Revise v3.1.4
  [276daf66] SpecialFunctions v0.10.3
  [90137ffa] StaticArrays v0.12.4
  [c3572dad] Sundials v3.9.0
  [d1185830] SymbolicUtils v0.5.1
  [6aa5eb33] TaylorSeries v0.10.8
  [d621b6e3] ValidatedNumerics v0.11.0

[odow]

Closing as duplicate of #1803. The issue is that JuMP recognizes the erf symbol from Calculus so ignores the registered function. We should either add SpecialFunctions as in #2369, or remove them entirely.