Add at-par macro

docs/src/index.md

@@ -30,6 +30,18 @@ s = delayed(combine)(p, q, r)
 
 @assert collect(s) == 16
 ```
+
+The above computation can also be written in a more Julia-idiomatic syntax with `@par`:
+
+```julia
+p = @par add1(4)
+q = @par add2(p)
+r = @par add1(3)
+s = @par combine(p, q, r)
+
+@assert collect(s) == 16
+```
+
 The connections between nodes `p`, `q`, `r` and `s` is represented by this dependency graph:
 
 ![graph](https://user-images.githubusercontent.com/25916/26920104-7b9b5fa4-4c55-11e7-97fb-fe5b9e73cae6.png)

src/thunk.jl

@@ -75,6 +75,19 @@ end
 
 delayedmap(f, xs...) = map(delayed(f), xs...)
 
+"""
+    @par f(args...) -> Thunk
+
+Convenience macro to call `Dagger.delayed` on `f` with arguments `args`.
+"""
+macro par(ex)
+    @assert ex.head == :call "@par requires a function call as the argument"
+    f = ex.args[1]
+    args = ex.args[2:end]
+    # TODO: Support kwargs
+    :(Dagger.delayed(f)($(args...)))
+end
+
 persist!(t::Thunk) = (t.persist=true; t)
 cache_result!(t::Thunk) = (t.cache=true; t)
 

test/runtests.jl

@@ -6,6 +6,7 @@ using Dagger
 
 include("fakeproc.jl")
 
+include("thunk.jl")
 include("domain.jl")
 include("array.jl")
 include("scheduler.jl")

test/thunk.jl

@@ -0,0 +1,8 @@
+@testset "@par" begin
+    x = 2
+    a = @par x + x
+    @test a isa Dagger.Thunk
+    b = @par sum([x,1,2])
+    c = @par a * b
+    @test collect(c) == 20
+end