'par' - why has it the type a -> b -> b ? - Haskell-Cafe

'par' - why has it the type a -> b -> b ?

Henning Thielemann

29 Sep 2008 29 Sep '08

4:55 a.m.

What is the reason for implementing parallelism with 'par :: a -> b -> b'? Analogy to 'seq'? I thought parallelism would be introduced most naturally by a function which does two computations in parallel and puts together their results after completion. Say

...

par2 :: (a -> b -> c) -> (a -> b -> c)

to be used like

...

par2 (+) expensiveComputationA expensiveComputationB

I assume that par2 can be implemented this way:

...

par2 f x y = f x (par x y)

Show replies by date

Achim Schneider

29 Sep 29 Sep

10:53 a.m.

Henning Thielemann wrote: par2 :: (a -> b -> c) -> a -> b -> c

...

...
par2 f x y = f x (par x y)

($!) :: (a -> b) -> a -> b f $! x = x `seq` f x It's terseness vs. maximum composability. I don't even want to think about implementing seq in terms of $!, makes my brain twist. -- (c) this sig last receiving data processing entity. Inspect headers for copyright history. All rights reserved. Copying, hiring, renting, performance and/or broadcasting of this signature prohibited.

Malcolm Wallace

11:01 a.m.

...

par2 :: (a -> b -> c) -> a -> b -> c

...
...
par2 f x y = f x (par x y)

Here is the dual: 'par' implemented in terms of parallel application: a `par` b = par2 (\x y-> y) a b

...

($!) :: (a -> b) -> a -> b f $! x = x `seq` f x

It's terseness vs. maximum composability. I don't even want to think about implementing seq in terms of $!, makes my brain twist.

It's not so difficult. (And you may see some similarity with the above defn of par.) a `seq` b = (\x y-> y) $! a b Regards, Malcolm

Achim Schneider

11:59 a.m.

Malcolm Wallace wrote:

...

(\x y -> y)

*shudder* I just can't stand such things. -- (c) this sig last receiving data processing entity. Inspect headers for copyright history. All rights reserved. Copying, hiring, renting, performance and/or quoting of this signature prohibited.

Sean Leather

1:12 p.m.

...

...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"? Sean

Achim Schneider

1:50 p.m.

"Sean Leather" wrote:

...

...
...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"?

It's the missing "x" on the right side. Makes my internal C compiler ache. -- (c) this sig last receiving data processing entity. Inspect headers for copyright history. All rights reserved. Copying, hiring, renting, performance and/or quoting of this signature prohibited.

Jonathan Cast

1:53 p.m.

On Mon, 2008-09-29 at 19:50 +0200, Achim Schneider wrote:

...

"Sean Leather" wrote:

...
...
...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"?

It's the missing "x" on the right side. Makes my internal C compiler ache.

(\ _ y -> y) Or fix your internal C compiler. jcc

Achim Schneider

2:34 p.m.

Jonathan Cast wrote:

...

On Mon, 2008-09-29 at 19:50 +0200, Achim Schneider wrote:

...
"Sean Leather" wrote:

...
...
...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"?

It's the missing "x" on the right side. Makes my internal C compiler ache.

(\ _ y -> y)

Still begs for a rewrite at the call-site... oh wait, that won't work. PS: I should tell claws that my inbox "Haskell Cafe" has no associated email account. Would stop me from replying off-list. -- (c) this sig last receiving data processing entity. Inspect headers for copyright history. All rights reserved. Copying, hiring, renting, performance and/or quoting of this signature prohibited.

Jonathan Cast

2:32 p.m.

On Mon, 2008-09-29 at 20:34 +0200, Achim Schneider wrote:

...

Jonathan Cast wrote:

...
On Mon, 2008-09-29 at 19:50 +0200, Achim Schneider wrote:

...
"Sean Leather" wrote:

...
...
...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"?

It's the missing "x" on the right side. Makes my internal C compiler ache.

(\ _ y -> y)

Still begs for a rewrite at the call-site... oh wait, that won't work.

And I still think you should fix your internal C compiler. This sort of thing can be taken to excess (see Fudgets :), but in moderation it's a perfectly fine *Haskell* idiom. jcc

Henning Thielemann

2:47 p.m.

On Mon, 29 Sep 2008, Sean Leather wrote:

...

...
...
(\x y -> y)

*shudder*

I just can't stand such things.

What is it that you can't stand? Would you prefer "flip const"?

No, certainly "const id". :-)

Jonathan Cast

1:29 p.m.

On Mon, 2008-09-29 at 17:59 +0200, Achim Schneider wrote:

...

Malcolm Wallace wrote:

...
(\x y -> y)

*shudder*

I just can't stand such things.

Then call it `flip const'. jcc

Bertram Felgenhauer

6:50 p.m.

Henning Thielemann wrote:

...

What is the reason for implementing parallelism with 'par :: a -> b -> b'? Analogy to 'seq'?

I'd think it's actually easier to implement than par2 below; evaluating par x y "sparks" a thread evaluating x, and then returns y. The analogy to 'seq' is there, of course.

...

I thought parallelism would be introduced most naturally by a function which does two computations in parallel and puts together their results after completion. Say

...
par2 :: (a -> b -> c) -> (a -> b -> c)

to be used like

...
par2 (+) expensiveComputationA expensiveComputationB

I assume that par2 can be implemented this way:

...
par2 f x y = f x (par x y)

For this to work, f has to evaluate its second argument before the first one, or the par will be useless. Try this:

...

par2 f x y = x `par` y `par` f x y

(In complete analogy to using `seq` for enforcing strictness.) HTH, Bertram

6081

Age (days ago)

6081

Last active (days ago)

List overview

Download

11 comments

6 participants

participants (6)

Achim Schneider
Bertram Felgenhauer
Henning Thielemann
Jonathan Cast
Malcolm Wallace
Sean Leather