Iterating a tuple taints :terminates_locally

[LilithHafner]

A for loop over a tuple of known length does not infer as "terminates" unless it is annotated with @assume_effects :terminates_locally. This seems like a reasonable thing for effect analysis to figure out.

julia> function f(x...)
           a = 0
           for i in x
               a += i
           end
           a
       end
f (generic function with 1 method)

julia> Base.infer_effects(f, NTuple{2, Int})
(+c,+e,!n,!t,+s,+m,+u,+o,+r)

julia> Base.@assume_effects :terminates_locally function g(x...)
           a = 0
           for i in x
               a += i
           end
           a
       end
g (generic function with 1 method)

julia> Base.infer_effects(g, NTuple{2, Int})
(+c,+e,!n,+t,+s,+m,+u,+o,+r)

julia> @b f(1,2,3)
1.984 ns

julia> @b g(1,2,3)
1.134 ns

Iterating a Vector is also typically guaranteed to terminate locally. However it will be harder to determine when it is not (e.g. push! in the loop), and it may take many more iterations than with a tuple.

[Keno]

We don't really have any SCEV capabilities at all on the julia side. As soon as a loop is involved, we taint termination conservatively.