Support if

[tshort]

It would be nice to support the following along with other uses of if:

a = [1:10]
b = rand(10)
@devec c = a .< 4 ? b : -b

[tshort]

Just found blend (like R's ifelse).

[nalimilan]

blend is useful, but it would also be interesting to have a vectorized switch-like function that would allow testing more conditions if the first one fails. More generally, the possibility to construct complex if... else blocks would really be nice to solve JuliaData/DataFrames.jl#369.

[lindahua]

One of the design goal here is the compatibility with the Julia base, that is, even omitting the @devec macro, the expression should still work as expected (but possibly slower).

For example, when you write @devec r = a + b .* c, and when you remove @devec, r = a + b. * c still works. In this way, people can first ensure the correctness of the implementation and then add @devec to accelerate part of the codes.

But this is not the case for c = a .< 4 ? b : -b -- this expression is not a valid one in itself in Julia.

What about we introduce a new function blend, and allow writing c = blend(c .< 4, b, -b) and prepending it with @devec.

[lindahua]

Oh, I forgot I already included blend in the package.

[nalimilan]

How about having a .? operator that would work like blend, for similarity with vectorized operations like .*?

Regarding if... else vectorized blocks, this would indeed be incompatible with base Julia. Maybe we should find a way to make them work in base first, and only then try to devectorize them?

[lindahua]

Unfortunately, .? is not a valid symbol / expression in Julia. Julia will raise error when you write :(a .? b : c) -- this means that this can not pass the basic syntax check even before the macro works on it.

[lindahua]

But we can try to propose to allow .? in Julia first.