Description
Filing for future reference (would be too much of an interruption to fix now):
julia> a = rand(UInt8, 5)
5-element Array{UInt8,1}:
0x08
0xcd
0x4a
0x5d
0x1f
julia> b = rand(UInt8, 5)
5-element Array{UInt8,1}:
0x7c
0xf4
0x2c
0x86
0xef
julia> dot(a, b)
0x9b
The fix is less trivial than it seems: the r_promote
infrastructure in reduce.jl
seems promising, but it doesn't handle the product of elements, only the sum of the accumulator. We might need a widen_op
analogous to promote_op
to prevent these kinds of overflows.