Feature #4649: Adding parallel constructors to Ruby 2.0

• Status: Rejected
• Assignee: matz (Yukihiro Matsumoto)
• Target version: -

[rosenfeld (Rodrigo Rosenfeld Rosas)]

I was not sure if this was RubySpec or Feature type.

My request is to create some new syntax for easing the write of concurrent code with Ruby. The syntax could be something like this:

parallels do
task "get response from service A" do # optional description
a = IO.read("http://serviceA/request")
end
task { b = get_response_from_service_b }
end

at this point both tasks have finish, it's like a join in the threads at the end of the parallel block.

call_another_service_that_depends_on(a and b)

I am not sure, though, how to deal with scopes inside the parallel block. Traditionally in Ruby, this wouldn't work because both 'a' and 'b' would be local to the 'task' block and would not be accessible from the outside of the parallel block. If we don't want instance variables, I don't know what is the best approach for avoiding some "a=nil; b=nil" before the parallel block.

Maybe this could start as a gem, but having it implemented directly in Ruby would be awesome, specially for scripts that usually don't rely in external dependencies for simplifying distribution...

Any thoughts about this propose?

[vjoel (Joel VanderWerf)]

IMO this is not really about parallel computation, but about syntax and scope, and it can be solved without changing ruby.

One solution that seems fairly idiomatic to me is:

def get_response_from_service_a; "a response"; end
def get_response_from_service_b; "b response"; end

class Parallels
def initialize
@tasks = []
end
def task &task
@tasks << Thread.new(&task)
end
def values
@tasks.map {|task| task.value}
end
end

def parallels &task_block
par = Parallels.new
par.instance_eval &task_block
par.values
end

a, b = parallels do
task { get_response_from_service_a }
task { get_response_from_service_b }
end

p a, b

[rosenfeld (Rodrigo Rosenfeld Rosas)]

You're right Joel. Maybe the syntax I'm requesting to be included in standard Ruby should be something like:

concurrently do
task {...}
...
end

But, although the implementation is simple, having a standard idiom for that common requirement would be great. That means I wouldn't have to redefine this idiom not to rely on external gem while doing some scripting... Other than that, making it a language feature could allow some low level improvements in performance, but it is just a guess since I don't know how this could be achieved yet.

Furthermore, depending on the specs, maybe it wouldn't be necessary to do something like "a, b = concurrently do...". Being a standard idiom, someone looking at a code like this would know for sure what is happening since it is a language feature instead of an external dependency...

Also, my initial example was too simple, but I imagine possibly several variables inside each task that should be available at the end of the parallels/concurrently block.

[naruse (Yui NARUSE)]

• Tracker changed from Misc to Feature

[naruse (Yui NARUSE)]

• Project changed from Ruby master to 14
• Target version deleted (3.0)

[naruse (Yui NARUSE)]

• Project changed from 14 to Ruby master

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

[rosenfeld (Rodrigo Rosenfeld Rosas)]

• File feature-4649.odp feature-4649.odp added

Third wish!

[mame (Yusuke Endoh)]

Received.

You want just Kernel#concurrently and Kernel#task?
And, Kernel#concurrently returns an array that contains the result values returned by inside Kernel#tasks. Right?
The array may be shuffled, or in the same order as task definitions?

ary = concurrently do
task { 1 }
task { 2 }
task { 3 }
end
p ary #=> [1, 2, 3]

--
Yusuke Endoh [e-mail]

[rosenfeld (Rodrigo Rosenfeld Rosas)]

If we're going to care about the results of the tasks, which I think it is a good idea, although it was really yours and not mine, it only makes sense to me for the array to have the results in the same order as the tasks were defined, right?

[mame (Yusuke Endoh)]

• Status changed from Assigned to Rejected

Rodrigo Rosenfeld Rosas,

Sorry but this proposal was rejected at the developer meeting (7/21).

Please start it as a gem. Matz might reconsider this proposal if
your gem really becomes popular or even de facto.

--
Yusuke Endoh [e-mail]

[rosenfeld (Rodrigo Rosenfeld Rosas)]

Okay, thanks for considering it :)