Remove convert methods which cause a large number of method invalidations.

These methods appear to be used from before we switched to MutableArithmetics.

In the future, instead of overloading Base.convert, we should use a new function
like `to_constant(::AffExpr)` which avoids the invalidation. Alternatively,
future versions of Julia may be clever enough to avoid the invalidations.

Author: odow

src/aff_expr.jl

@@ -311,13 +311,6 @@ end
 function Base.convert(::Type{GenericAffExpr{T,V}}, v::_Constant) where {T,V}
     return GenericAffExpr{T,V}(convert(T, _constant_to_number(v)))
 end
-# Used in `JuMP._mul!`.
-function Base.convert(::Type{T}, aff::GenericAffExpr{T}) where T
-    if !isempty(aff.terms)
-        throw(InexactError(:convert, T, aff))
-    end
-    return convert(T, aff.constant)
-end
 
 # Alias for (Float64, VariableRef), the specific GenericAffExpr used by JuMP
 const AffExpr = GenericAffExpr{Float64,VariableRef}

src/quad_expr.jl

@@ -324,13 +324,6 @@ function Base.convert(::Type{GenericQuadExpr{C, V}}, v::Union{_Constant,Abstract
     return GenericQuadExpr(convert(GenericAffExpr{C, V}, v))
 end
 GenericQuadExpr{C, V}() where {C, V} = zero(GenericQuadExpr{C, V})
-# Used in `JuMP._mul!`.
-function Base.convert(::Type{T}, quad::GenericQuadExpr{T}) where T
-    if !isempty(quad.terms)
-        throw(InexactError(:convert, T, quad))
-    end
-    return convert(T, quad.aff)
-end
 
 function check_belongs_to_model(q::GenericQuadExpr, model::AbstractModel)
     check_belongs_to_model(q.aff, model)