Don't rely on reduce being fold-left

[stevengj]

As you mentioned today, the code

mdo_desugar(exprIn) = reduce(mdo_desugar_helper, :(), reverse(exprIn.args))

assumes that the associativity of reduce is fold-left, which is true of the current implementation but is undocumented, and may change (JuliaLang/julia#4046).

In any case,

function mdo_desugar(exprIn)
    r = :()
    for i = length(exprIn.args):-1:1
        r = mdo_desugar_helper(exprIn.args[i], r)
    end
    r
end

has the additional virtue of being faster (since it avoids the call to reverse and the allocation of a new array), not that it probably matters to you.

[pao]

virtue of being faster

This is so clearly not relevant to any of this code :)

I think I used to have it written something like that at one point during its evolution, but the higher-order-function approach was appealing for reasons that are completely impractical, but that fits well with the context. But in light of JuliaLang/julia#4046 I agree that it'll need to be changed. I'll roll it in when I update all my packages around 0.2-rc1 time.

[jfarid27]

Closing since it's old

[stevengj]

The bug in mdo_desugar (relying on associativity of reverse) is still there in the current code. It should just call foldl.