re-passing function keywords break type inference

[stevengj]

Passing function keywords through a second function call seems to break type inference:

using StaticArrays, Base.Test
foo{n,T<:AbstractFloat}(f, a::SVector{n,T}, b::SVector{n,T}; norm=vecnorm) = f(norm(a + b) * n)
function foo(f, a::SVector{n,T}, b::SVector{n,S}; kws...) where {n,T<:Real,S<:Real}
    F = float(promote_type(T,S))
    return foo(f, SVector{n,F}(a), SVector{n,F}(b); kws...)
end

@inferred foo(abs, SVector(0.), SVector(1.))  # succeeds
@inferred foo(abs, SVector(0), SVector(1))    # FAILS!

If I delete the kws... in the second foo method then inference succeeds.

[JeffBezanson]

Yes, this is a pretty well-known pothole; see for example #9551 #21794 #21811.

[JeffBezanson]

Closing as dup.