Generalize filterM to Applicative

David Feuer

27 Dec 2014 27 Dec '14

11:59 p.m.

Currently, filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] filterM p = foldr go (return []) where go x r = do flg <- p x ys <- r return (if flg then x:ys else ys) We could change this to filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a] filterM p = foldr go (pure []) where go x r = f <$> p x <*> r where f flg ys = if flg then x:ys else ys

Show replies by date

Yitzchak Gale

28 Dec 28 Dec

1:14 p.m.

David Feuer wrote:

...

filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] We could change this to filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a]

Following the usual conventions, that should be filterA. And I'm not sure where it should go. We could also generalize the list parameter to a Foldable and put it in Data.Foldable, but I don't think that would be very useful, since in the end it anyway returns a list. We could add a new "Filterable" class, but that's also probably not too useful for anything not list-like. For example, Data.Tree has two useful filters, filterGraft and filterPrune. So maybe it should go in Control.Applicative? Not very satisfying. Any ideas? As for the fate of the name filterM: Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA. Thanks, Yitz

David Feuer

1:40 p.m.

In a departure from the filterM tradition, I think if we add a filterA for lists then it should go in Data.List. That way, we can have a different filterA in Data.Sequence, etc. The A makes it Applicative already; the rest seems tied to the container. On Sun, Dec 28, 2014 at 1:14 PM, Yitzchak Gale wrote:

...

David Feuer wrote:

...
filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] We could change this to filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a]

Following the usual conventions, that should be filterA.

And I'm not sure where it should go. We could also generalize the list parameter to a Foldable and put it in Data.Foldable, but I don't think that would be very useful, since in the end it anyway returns a list.

We could add a new "Filterable" class, but that's also probably not too useful for anything not list-like. For example, Data.Tree has two useful filters, filterGraft and filterPrune.

So maybe it should go in Control.Applicative? Not very satisfying. Any ideas?

As for the fate of the name filterM:

Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA.

Thanks, Yitz

Edward Kmett

6:15 p.m.

The way we've been doing things so far is to just generalize the existing signature and keep the existing name. This avoids needless name proliferation and keeping needlessly over-constrained versions of code lying around. FWIW- I have no objection to a more general filterM. -Edward On Sun, Dec 28, 2014 at 1:14 PM, Yitzchak Gale wrote:

...

David Feuer wrote:

...
filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] We could change this to filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a]

Following the usual conventions, that should be filterA.

And I'm not sure where it should go. We could also generalize the list parameter to a Foldable and put it in Data.Foldable, but I don't think that would be very useful, since in the end it anyway returns a list.

We could add a new "Filterable" class, but that's also probably not too useful for anything not list-like. For example, Data.Tree has two useful filters, filterGraft and filterPrune.

So maybe it should go in Control.Applicative? Not very satisfying. Any ideas?

As for the fate of the name filterM:

Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA.

Thanks, Yitz _______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

Andreas Abel

29 Dec 29 Dec

5:52 a.m.

+1 generalize filterM to Applicative. I am not fond of ...A functions if the ...M functions can be generalized to Applicative, so -1 on filterA. On 29.12.2014 00:15, Edward Kmett wrote:

...

The way we've been doing things so far is to just generalize the existing signature and keep the existing name.

This avoids needless name proliferation and keeping needlessly over-constrained versions of code lying around.

FWIW- I have no objection to a more general filterM.

-Edward

On Sun, Dec 28, 2014 at 1:14 PM, Yitzchak Gale mailto:gale@sefer.org> wrote:

David Feuer wrote: > filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] > We could change this to > filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a]

Following the usual conventions, that should be filterA.

And I'm not sure where it should go. We could also generalize the list parameter to a Foldable and put it in Data.Foldable, but I don't think that would be very useful, since in the end it anyway returns a list.

We could add a new "Filterable" class, but that's also probably not too useful for anything not list-like. For example, Data.Tree has two useful filters, filterGraft and filterPrune.

So maybe it should go in Control.Applicative? Not very satisfying. Any ideas?

As for the fate of the name filterM:

Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA.

Thanks, Yitz _______________________________________________ Libraries mailing list Libraries@haskell.org mailto:Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Andreas Abel <>< Du bist der geliebte Mensch. Department of Computer Science and Engineering Chalmers and Gothenburg University, Sweden andreas.abel@gu.se http://www2.tcs.ifi.lmu.de/~abel/

Edward Kmett

7:58 a.m.

...

Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA.

What would be gained for having a crippled version of filterM that only worked for monads, but didn't use any of the additional power gained by that constraint? To borrow from John Maynard Keynes: "When the facts change, I change my mind. What do you do, sir?" =) Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive? -Edward On Sun, Dec 28, 2014 at 1:14 PM, Yitzchak Gale wrote:

...

David Feuer wrote:

...
filterM :: (Monad m) => (a -> m Bool) -> [a] -> m [a] We could change this to filterM :: (Applicative f) => (a -> f Bool) -> [a] -> f [a]

Following the usual conventions, that should be filterA.

And I'm not sure where it should go. We could also generalize the list parameter to a Foldable and put it in Data.Foldable, but I don't think that would be very useful, since in the end it anyway returns a list.

We could add a new "Filterable" class, but that's also probably not too useful for anything not list-like. For example, Data.Tree has two useful filters, filterGraft and filterPrune.

So maybe it should go in Control.Applicative? Not very satisfying. Any ideas?

As for the fate of the name filterM:

Personally, I would want filterM to remain a convenient type-specialized version for monads. But recent madness would turn it into yet another unsightly overly polymorphic namespace wasting synonym for filterA.

Thanks, Yitz _______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

Kim-Ee Yeoh

10:50 a.m.

On Mon, Dec 29, 2014 at 7:58 PM, Edward Kmett wrote:

...

Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive?

Madness is such strong language for this august list. May I speak on behalf of haskell newcomers for a time? Haskell places such an emphasis on uniformity and regularity. Functions with names that end with M once meant they were monadic variants of those that don't. That's no longer uniformly the case, because of the FAM restructuring. The names of functions matter. Anachronistic labels confuse. Leaving filterM with a type signature of Applicative cannot be the long-term solution. -- Kim-Ee

Andreas Abel

1:31 p.m.

Or you can widen the interpretation of suffix ...M as "effectful", which could be a monadic or applicative effect. On 29.12.2014 16:50, Kim-Ee Yeoh wrote:

...

On Mon, Dec 29, 2014 at 7:58 PM, Edward Kmett mailto:ekmett@gmail.com> wrote:

Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive?

Madness is such strong language for this august list.

May I speak on behalf of haskell newcomers for a time?

Haskell places such an emphasis on uniformity and regularity. Functions with names that end with M once meant they were monadic variants of those that don't. That's no longer uniformly the case, because of the FAM restructuring.

The names of functions matter.

Anachronistic labels confuse.

Leaving filterM with a type signature of Applicative cannot be the long-term solution.

-- Kim-Ee

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Andreas Abel <>< Du bist der geliebte Mensch. Department of Computer Science and Engineering Chalmers and Gothenburg University, Sweden andreas.abel@gu.se http://www2.tcs.ifi.lmu.de/~abel/

Edward Kmett

3:17 p.m.

I think this is the interpretation I think we're probably best left with. I agree that names matter and that anachronistic labels confuse, but I think it is the lesser of evils to widen the definition of the 'M' suffix than it is to double up on almost all the names taken in the environment and force the entire community to go through a positively gigantic deprecation cycle. -Edward On Mon, Dec 29, 2014 at 1:31 PM, Andreas Abel wrote:

...

Or you can widen the interpretation of suffix ...M as "effectful", which could be a monadic or applicative effect.

On 29.12.2014 16:50, Kim-Ee Yeoh wrote:

...
On Mon, Dec 29, 2014 at 7:58 PM, Edward Kmett mailto:ekmett@gmail.com> wrote:

Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive?

Madness is such strong language for this august list.

May I speak on behalf of haskell newcomers for a time?

Haskell places such an emphasis on uniformity and regularity. Functions with names that end with M once meant they were monadic variants of those that don't. That's no longer uniformly the case, because of the FAM restructuring.

The names of functions matter.

Anachronistic labels confuse.

Leaving filterM with a type signature of Applicative cannot be the long-term solution.

-- Kim-Ee

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Andreas Abel <>< Du bist der geliebte Mensch.

Department of Computer Science and Engineering Chalmers and Gothenburg University, Sweden

andreas.abel@gu.se http://www2.tcs.ifi.lmu.de/~abel/

David Feuer

30 Dec 30 Dec

12:34 p.m.

I realized just now that we can actually make the type a little bit more general still, interpreting `filterM` for lists as being an applicative filter *producing* lists: filterM :: (Applicative f, Foldable t) => (a -> f Bool) -> t a -> f [a] On Mon, Dec 29, 2014 at 3:17 PM, Edward Kmett wrote:

...

I think this is the interpretation I think we're probably best left with.

I agree that names matter and that anachronistic labels confuse, but I think it is the lesser of evils to widen the definition of the 'M' suffix than it is to double up on almost all the names taken in the environment and force the entire community to go through a positively gigantic deprecation cycle.

-Edward

On Mon, Dec 29, 2014 at 1:31 PM, Andreas Abel wrote:

...
Or you can widen the interpretation of suffix ...M as "effectful", which could be a monadic or applicative effect.

On 29.12.2014 16:50, Kim-Ee Yeoh wrote:

...
On Mon, Dec 29, 2014 at 7:58 PM, Edward Kmett mailto:ekmett@gmail.com> wrote:

Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive?

Madness is such strong language for this august list.

May I speak on behalf of haskell newcomers for a time?

Haskell places such an emphasis on uniformity and regularity. Functions with names that end with M once meant they were monadic variants of those that don't. That's no longer uniformly the case, because of the FAM restructuring.

The names of functions matter.

Anachronistic labels confuse.

Leaving filterM with a type signature of Applicative cannot be the long-term solution.

-- Kim-Ee

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Andreas Abel <>< Du bist der geliebte Mensch.

Department of Computer Science and Engineering Chalmers and Gothenburg University, Sweden

andreas.abel@gu.se http://www2.tcs.ifi.lmu.de/~abel/

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

Mario Blažević

5 Jan 5 Jan

11:53 a.m.

On 14-12-30 12:34 PM, David Feuer wrote:

...

I realized just now that we can actually make the type a little bit more general still, interpreting `filterM` for lists as being an applicative filter *producing* lists:

filterM :: (Applicative f, Foldable t) => (a -> f Bool) -> t a -> f [a]

I'm +1 on the more limited generalization proposal, but not on this one. I've been feeling for a while that we need a subclass of Traversable to generalize functions like filter, mapMaybe, and partition (potentially span, splitAt and others as well). If you call this missing class Filterable, the properly-generalized signature would be

...

filterM :: (Applicative f, Filterable t) => (a -> f Bool) -> t a -> f (t a)

...

On Mon, Dec 29, 2014 at 3:17 PM, Edward Kmett wrote:

...
I think this is the interpretation I think we're probably best left with.

I agree that names matter and that anachronistic labels confuse, but I think it is the lesser of evils to widen the definition of the 'M' suffix than it is to double up on almost all the names taken in the environment and force the entire community to go through a positively gigantic deprecation cycle.

-Edward

On Mon, Dec 29, 2014 at 1:31 PM, Andreas Abel wrote:

...
Or you can widen the interpretation of suffix ...M as "effectful", which could be a monadic or applicative effect.

On 29.12.2014 16:50, Kim-Ee Yeoh wrote:

...
On Mon, Dec 29, 2014 at 7:58 PM, Edward Kmett mailto:ekmett@gmail.com> wrote:

Is it "madness" to want to avoid namespace proliferation and maximize the usefulness of an existing combinator now that the constraints that forged it have changed to allow it to be slightly more permissive?

Madness is such strong language for this august list.

May I speak on behalf of haskell newcomers for a time?

Haskell places such an emphasis on uniformity and regularity. Functions with names that end with M once meant they were monadic variants of those that don't. That's no longer uniformly the case, because of the FAM restructuring.

The names of functions matter.

Anachronistic labels confuse.

Leaving filterM with a type signature of Applicative cannot be the long-term solution.

-- Kim-Ee

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Andreas Abel <>< Du bist der geliebte Mensch.

Department of Computer Science and Engineering Chalmers and Gothenburg University, Sweden

andreas.abel@gu.se http://www2.tcs.ifi.lmu.de/~abel/

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

_______________________________________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/mailman/listinfo/libraries

-- Mario Blazevic mblazevic@stilo.com Stilo International This message, including any attachments, is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure, copying, or distribution is strictly prohibited. If you are not the intended recipient(s) please contact the sender by reply email and destroy all copies of the original message and any attachments.

M Farkas-Dyck

25 Mar 25 Mar

4:06 p.m.

On 05/01/2015 at 11:53:11 -0500, Mario Blažević wrote:

...

On 14-12-30 12:34 PM, David Feuer wrote:

...
I realized just now that we can actually make the type a little bit more general still, interpreting `filterM` for lists as being an applicative filter *producing* lists:

filterM :: (Applicative f, Foldable t) => (a -> f Bool) -> t a -> f [a]

I'm +1 on the more limited generalization proposal, but not on this one. I've been feeling for a while that we need a subclass of Traversable to generalize functions like filter, mapMaybe, and partition (potentially span, splitAt and others as well). If you call this missing class Filterable, the properly-generalized signature would be

...
filterM :: (Applicative f, Filterable t) => (a -> f Bool) -> t a -> f (t a)

https://hackage.haskell.org/package/witherable Interesting name choice...

M Farkas-Dyck

4:20 p.m.

On 25/03/2015 at 15:06:47 -0500, M Farkas-Dyck wrote:

...

https://hackage.haskell.org/package/witherable

Interesting name choice...

This has superclass Traversable tho, which is not needed as wither f xs = catMaybes <$> traverse f xs filterA f = mapMaybeA (\ x -> (x <$) . guard <$> f x)

3709

Age (days ago)

3796

Last active (days ago)

List overview

Download

12 comments

7 participants

participants (7)

Andreas Abel
David Feuer
Edward Kmett
Kim-Ee Yeoh
M Farkas-Dyck
Mario Blažević
Yitzchak Gale