why does inlining introduce a variable here?

[JeffBezanson]

julia> code_typed(!=,((),()))
1-element Array{Any,1}:
 :($(Expr(:lambda, {:x,:y}, {{:_var0},{{:x,(),0},{:y,(),0},{:_var0,Bool,18}},{}}, :(begin  # operators.jl, line 27:
        _var0 = x::() == y::()::Bool
        return box(Bool,not_int(_var0::Bool))::Bool
    end::Bool))))

This seems pretty unnecessary. cc @vtjnash

[vtjnash]

i guess since == is not effect_free, it might be pessimistic and move it into an argument. I'm not sure why though

[JeffBezanson]

Yeah that's pretty weird. There is only one occurrence. This is almost the simplest case possible.

[vtjnash]

box(XXX, ...) is not effect_free when XXX is a symbol. effect_free in inference.jl::1873 needs to return true if XXX is a symbol representing a global immutable constant (which unfortunately doesn't actually help here because Bool is mutable, but it is perhaps a step in the right direction)

[vtjnash]

not a bug

[vtjnash]

julia> code_typed(!=, Tuple{Tuple{},Tuple{}})
1-element Array{Any,1}:
 LambdaInfo for !=
:(begin  # operators.jl, line 27:
        GenSym(0) = (_2::Tuple{} == _3::Tuple{})::Bool # bool.jl, line 16:
        return (Base.box)(Base.Bool,(Base.not_int)(GenSym(0)))
    end::Bool)

julia> code_llvm(!=, Tuple{Tuple{},Tuple{}})

define i1 @"julia_!=_32319"() {
top:
  %0 = call i1 @"julia_==_32320"()
  %1 = xor i1 %0, true
  ret i1 %1
}