broadcast is ambiguous with base

[Keno]

This method: https://github.com/Keno/StaticArrays.jl/blob/master/src/broadcast.jl#L7
is ambiguous with base:

MethodError: broadcast(::Optim.UnivariateProblems.##1#3, ::Float64) is ambiguous. Candidates:
  broadcast(f, a::Union{Number, StaticArrays.StaticArray}, b::Union{Number, StaticArrays.StaticArray}...) in StaticArrays at /global/cscratch1/sd/kfischer/celeste-pkgs/v0.6/StaticArrays/src/broadcast.jl:8
  broadcast(f, x::Number...) in Base.Broadcast at broadcast.jl:16

that's bad form

[timholy]

The problem here is that Base broadcast requires zero or more arguments whereas here it requires one or more. Presumably just need to get rid of the a argument and start straight with b, since Union{A,B} should always be less specific than A.

Overall, detect_ambiguities picks up 72 ambiguities when StaticArrays is loaded. The good news is that Base.Dates seems to no longer be a source of trouble.

[andyferris]

This is due to a recent PR by @tkelman: https://github.com/JuliaArrays/StaticArrays.jl/pull/126/files

I suspect we should role back that particular change for broadcast?

[marius311]

Just ran into this also. Can't say I'm following closely enough to understand why the two separate a,b args were needed in the PR mentioned above, but if you want to keep that, you could use the following trick,

function broadcast(f, a::Union{_, Number, StaticArray}, b::Union{_, Number, StaticArray}...) where {_<:StaticArray}

this will only match when there's at least one StaticArray present among the args, so there's not even a question of if its more specific.

[marius311]

(See also https://discourse.julialang.org/t/incorrect-ambiguity/2690 since this trick/hack doesn't always work, granted I just checked it does in this case)

[andyferris]

Thanks @marius311 - that's a very interesting observation.

[dlfivefifty]

Even exp.(1.0+2.0im) triggers this....

[dlfivefifty]

How would you feel about this (less than elegant) work around for low number of arguments:

@propagate_inbounds broadcast(f, a::StaticArray, b::Union{Number, StaticArray}...) = _broadcast(f, broadcast_sizes(a,b...), a, b...)
@propagate_inbounds broadcast(f, n1::Number, a::StaticArray, b::Union{Number, StaticArray}...) = _broadcast(f, broadcast_sizes(n1,a,b...), n1, a, b...)
@propagate_inbounds broadcast(f, n1::Number, n2::Number, a::StaticArray, b::Union{Number, StaticArray}...) = _broadcast(f, broadcast_sizes(n1,n2,a,b...), n1, n2, a, b...)

[andyferris]

Does #135 work for you @dlfivefifty?

[dlfivefifty]

Yep, it does!

…

On 3 Apr 2017, at 4:48 pm, Andy Ferris ***@***.***> wrote: Does #135 <#135> work for you @dlfivefifty <https://github.com/dlfivefifty>? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#132 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABLDqfvXcMVnzTQ7Uf9X_FQJavnkwaw7ks5rsJYygaJpZM4MwYUB>.

[c42f]

I assume this is sorted now out via #135.

By the way @marius311 that's a really nice trick. I've used in #136 which would otherwise seem intractable in general.