I'm thinking about a second edition of Parallel and Concurrent Programming in Haskell. Any thoughts on what should be added (or removed)? What's missing, what updates are needed to reflect common usage? The obvious new things are async and Haxl. Should the chapter on distributed-process be kept? - ThreadSky

simonmar.bsky.social • 56 days ago

I'm thinking about a second edition of Parallel and Concurrent Programming in Haskell. Any thoughts on what should be added (or removed)? What's missing, what updates are needed to reflect common usage?

The obvious new things are async and Haxl. Should the chapter on distributed-process be kept?

Comments

lucasdicioccio.bsky.social•55 days ago

I think it would be great to have some (pun intended) compact reference on the compacting GC, compact regions, and updated RTS tooling would be neat.

Also, some discussion about the relevancy of hot-reload (with a demo?) vs. stateless-container architectures would be neat.

moritzhamann.com•56 days ago

This is going to be random and not really relevant but ... Even though I've never programmed more than a few lines of Haskell, by chance I've read the first edition over the course of a few rainy days on holiday once and I really enjoyed it!

noteed.bsky.social•55 days ago

A discussion about using async to spawn and reliably monitor processes (maybe using System.Process) would be interesting.

chandlerbarlow.blog•55 days ago

I have a colleague who loves your book so much that he passive aggressively quotes pages of it in prs like it’s a scripture.

An update would blow his mind.

chandlerbarlow.blog•55 days ago

On the distributed process note we have had problems with it and had to create a homemade solution. Seeing what alternatives/best practices are out there for haskell rpcs would be very helpful.

simonmar.bsky.social•51 days ago

Thrift is what we used at Meta. https://github.com/facebookincubator/hsthrift It's one of the things I call a "Big Hammer" - so useful you can't do without it, and once you buy into it for your software ecosystem, it becomes a single point of leverage. I'm in the process of putting the packages on Hackage.

simonmar.bsky.social•51 days ago

Haha! I'd love to know which tidbits of advice have risen to this level, it might be useful if I'm reorganising things.

diazcarrete.bsky.social•55 days ago

Perhaps the bounded/closeable versions of Chans and Queues from "stm-chans" could merit a mention. I've often found them useful in practice.

"concurrently" from async is used a whole lot, sometimes in versions with a bounded thread pool.

Best practices around async exception handling.

simonmar.bsky.social•55 days ago

Yeah, there's actually a bounded version of mapConcurrently that we ended up using at Meta for most simple IO parallelism use cases, I'm considering adding this to async https://github.com/facebookincubator/hsthrift/blob/main/common/util/Control/Concurrent/Stream.hs

gusbicalho.bsky.social•55 days ago

Ki https://hackage.haskell.org/package/ki seems to be gaining traction - I saw a couple people recommending it at https://discourse.haskell.org/t/dynamic-number-of-asyncs/11574/6

davidslt.es•53 days ago

Activated my account here just to say that I’m really excited about a second edition! Working with the first one was instrumental for me getting a job, so I have it in a really high esteem.

Really interested in the newest things Haskell can offer in this area.

tsmc.purely-functional.com•54 days ago

I would keep the distributed process section, as there’s not really an equivalent Actor library in Haskell. The section on exceptions in Haskell is wonderful. More coverage on GPUs and patterns for building concurrent applications.

robstewartuk.bsky.social•55 days ago

Tiny suggestion - I'm unsure whether the 1st edition included an example of recursive Eval monad parallelism. Performance-wise, `runEval` is free, recursive calls to it doesn't have overhead. See the attached example. Apologies if recursive `runEval` calls is discussed, I must've missed it.

robstewartuk.bsky.social•55 days ago

Any more reflections you have about Eval vs Par monads would be great. I still teach rpar/rseq, but I understand that @nishiorain.bsky.social taught using the Par monad. Could like-for-like code examples and their performance be given to demonstrate these pros/cons:

https://www.oreilly.com/library/view/parallel-and-concurrent/9781449335939/ch04.html#sec_par-monad-reflections

robstewartuk.bsky.social•55 days ago

When I advocate for the Eval monad in lectures, my analogy is it's parallel *breadth-first* search. E.g. the sudoku solver in your book, you want to eat away at lots of possible solutions rather than going down the rabbit hole of one: if you can fully embrace laziness, the Eval monad is for you.

robstewartuk.bsky.social•55 days ago

Which is to ask: could a 2nd edition of the book be as clear as saying: if you're not in the habit of representing your computations as lazy data structures and you don't wish to familiarise yourself with evaluation strategies, then use the simpler Par monad API?

robstewartuk.bsky.social•55 days ago

Perhaps expanding on your December 2018 views about distributed-process on the parallel-haskell mailing list would be offer a retrospective insight about that library.

https://groups.google.com/g/parallel-haskell/c/Hr6vkpz3FB4/m/RuiNR47ZCgAJ

simonmar.bsky.social•55 days ago

Thanks, I'd forgotten about that post! How insightful 😆

simonmar.bsky.social•55 days ago

Perhaps a chapter on Thrift would be a more useful replacement for the distributed-process chapter.

ktroxie.bsky.social•55 days ago

Loved the first edition and am so excited. Some ideas:

- more GPUs, specifically cuda, given relevance to deep learning
- thrift / RPC for distributed / multiprocess / interop with other languages
- structuring apps with both io and cpu-bound work (stuff in folly that requires multiple executors).

Comments

Posting Rules

Reply