`@assume_effects` and `@nospecializeinfer` together produce wrong result

I've encountered a situation where the result of a function call changes if both `@assume_effects :foldable` and `@nospecializeinfer` are used. Here is a MWE: ``` using Base: @assume_effects, @nospecializeinfer @nospecializeinfer function g(T::Type{<:Tuple{Any,Any}}) @assume_effects :foldable @nospecialize promote_type(map(f, fieldtypes(T))...) # promote_type(f(fieldtypes(T)[1]), f(fieldtypes(T)[2])) end h(t::T) where T <: Tuple{Any,Any} = g(T) t = ('x', 'y') T = typeof(t) f(::Type{Char}) = Int8 @show g(T), h(t) f(::Type{Char}) = Bool @show g(T), h(t) ``` The output is ``` (g(T), h(t)) = (Int8, Int8) (g(T), h(t)) = (Bool, Int8) ``` So the function `g` notices that `f` has changed, but `h` doesn't. (Or is this expected because of some world-age implications?) If `@nospecializeinfer` or `@assume_effects` is removed, then I get the expected result: ``` (g(T), h(t)) = (Int8, Int8) (g(T), h(t)) = (Bool, Bool) ``` Also, if the call to `promote_type` is replaced by the line that is commented out, the result is again correct. Tested with Julia 1.10.11, 1.11.9, 1.12.6 and 1.13.0-rc1.