Task: A diverging design from Future and Scalaz Task

task-proposal.md

Monix Task: A diverging design

This document is about Task, an alternative for Scalaz's Task or Scala's Future. Note this is work in progress and this document suffered multiple modifications already:

• Jan 26, 2016
• Dec 23, 2015
• Dec 22, 2015
• Dec 21, 2015
• Dec 20, 2015

UPDATE - May 30, 2016 - This document is now obsolete, replaced by the Monix Task documentation:

https://monix.io/docs/2x/eval/task.html

@tpolecat re question from cats gitter: "is there a better place to discuss this?"

@adelbertc re WIP branch

Both points were brought up by @non in https://github.com/non/cats/issues/32#issuecomment-141079117, but that was a while ago. Specifically:

I think using this issue to brainstorm, ask questions, and help try to flesh out the requirements of a task in Cats would definitely be helpful.

and

Thus, feature/task (in its current state) is not eligible to be merged.

Just as a suggestion, perhaps the old issue has got stale and rather than try and reboot it, it might be worth launching a V2 issue?

👍 @inthenow.

I have updated the document. It now contains a list of what this implementation does better than the Task in Scalaz, as I think @milessabin wanted. See the "What's better than Scalaz?" section.

Cool. Before I/we do this, there are two issues regarding moving the current Future instances from Cats to Alleycats - non/alleycats#26 and https://github.com/non/cats/issues/589.

Given this - would Alleycats be the better place to open the new issue and possible first implementation?

Changed again (Dec 30).

scalaz.concurrent.Task is not very good. Using just a reasonable IO type, it should be possible to subsume Task's use cases with some uses of MVar-like atomic-references. For example:

scalaz/scalaz@series/7.3.x...puffnfresh:feature/concurrent-io

I highly recommend focusing on doing a good job at I/O and then using that to talk about concurrency, rather than trying to do the opposite.

@puffnfresh: 👍

Hi all,

You might be interested to look at my matterhorn experiment: https://github.com/aloiscochard/matterhorn

It is basically a specialized interpreter for IO actions: https://github.com/aloiscochard/matterhorn/blob/master/core/src/main/scala/rts/Interpreter.scala

There is some tests and benchmarks showing how it work:
https://github.com/aloiscochard/matterhorn/blob/master/core/src/test/scala/CoreSpec.scala
https://github.com/aloiscochard/matterhorn/blob/master/bench/src/main/scala/pure.scala

The approach to concurrency is to use a STM, I just did plug a library for having rapid prototyping.

Good luck

@aloiscochard, @puffnfresh thanks, will take a look.

scalaz.concurrent.Task is not very good. Using just a reasonable IO type, it should be possible to subsume Task's use cases with some uses of MVar-like atomic-references. For example:

I agree, and that is the route we are going with FS2's Task type (which I suppose could be renamed to 'IO'), which implements MVar-like references and then uses that to implement async (and lots of other operations, like parallelTraverse, as well as all the concurrent stream operations of FS2 itself): https://github.com/functional-streams-for-scala/fs2/blob/topic/redesign/core/src/main/scala/fs2/Async.scala#L75