Redefining a method used in a generated function

General Usage
1

I understand that generated functions are generated “on demand” and therefore all used methods have to be defined before:

julia> @generated g() = (@show f(1); nothing)
g (generic function with 1 method)

julia> f(x) = x
f (generic function with 1 method)

julia> g()
ERROR: MethodError: no method matching f(::Int64)

However, why it is not possible to redefine methods like this?

julia> f(x) = x
f (generic function with 1 method)

julia> @generated g() = (@show f(1); nothing)
g (generic function with 1 method)

julia> f(x) = x-1
f (generic function with 1 method)

julia> g()
f(1) = 1

Is there a way around this or is it just fundamentally impossible? I am writing a library, where I would like to generate functions based on type traits and this would really come handy.

2

No, this is not safe. Although, it often works, there are no guarantees. Some discussion here: https://github.com/mauro3/SimpleTraits.jl/issues/40

1 Like
3

It is garanteed to not work. If it does then it’s a bug.

2 Likes