Proposal: Add Data.Semigroup to base, as a superclass of Monoid

John Wiegley

11 Jun 2013 11 Jun '13

2:46 p.m.

1. I propose that we add the following package to base: http://hackage.haskell.org/packages/archive/semigroups/0.9.2/doc/html/Data-S... This is somewhat in the spirit of the AMP proposal: further improving the correctness of our algebraic abstractions. 2. That we make Semigroup a superclass of Monoid, so that (minimally): class Semigroup a where (<>) :: a -> a -> a class Semigroup a => Monoid a where mempty :: a mconcat :: [a] -> a mconcat = foldr (<>) mempty mappend :: Semigroup a => a -> a -> a mappend = (<>) 3. (Optional, recommended) There are other useful functions that can be added to Semigroup, such as sconcat and times1p, but I will let Edward speak to whether those should be proposed at this time. 4. (Optional, recommended) That we fix the Monoid instance for Maybe to be: instance Semigroup a => Semigroup (Maybe a) where Just x <> Just y = Just (x <> y) _ <> _ = Nothing instance Semigroup a => Monoid (Maybe a) where mempty = Nothing For some clarification on what semigroups are and why we'd want to change Monoid, I excerpt here a selection from Brent Yorgey's "Monoids and Variations" paper: Semigroups A semigroup is like a monoid without the requirement of an identity element: it consists simply of a set with an associative binary operation.... Of course, any monoid is automatically a semigroup (by forgetting about its identity element). In the other direction, to turn a semigroup into a monoid, simply add a new distinguished element to serve as the identity, and extend the definition of the binary operation appropriately. This creates an identity element by definition, and it is not hard to see that it preserves associativity.... Adding a new distinguished element to a type is typically accomplished by wrapping it in Maybe. One might therefore expect to turn an instance of Semigroup into an instance of Monoid by wrapping it in Maybe. Sadly, Data.Monoid does not define semigroups, and has a Monoid instance for Maybe which requires a Monoid constraint on its argument type... This is somewhat odd: in essence, it ignores the identity element of [the type] and replaces it with a different one. -- John Wiegley FP Complete Haskell tools, training and consulting http://fpcomplete.com johnw on #haskell/irc.freenode.net

Show replies by date

John Wiegley

11 Jun 11 Jun

2:51 p.m.

...

...
...
...
...
John Wiegley writes:

...

4. (Optional, recommended) That we fix the Monoid instance for Maybe to be:

...

instance Semigroup a => Semigroup (Maybe a) where Just x <> Just y = Just (x <> y) _ <> _ = Nothing

Slight correction (thanks to Brent Yorgey for catching this): instance Semigroup a => Semigroup (Maybe a) where Just x <> Just y = Just (x <> y) Nothing <> x = x x <> Nothing = x -- John Wiegley FP Complete Haskell tools, training and consulting http://fpcomplete.com johnw on #haskell/irc.freenode.net

Mario Blažević

5:05 p.m.

On 06/11/13 14:46, John Wiegley wrote:

...

1. I propose that we add the following package to base:

http://hackage.haskell.org/packages/archive/semigroups/0.9.2/doc/html/Data-S...

This is somewhat in the spirit of the AMP proposal: further improving the correctness of our algebraic abstractions.

I was wondering how much longer until this proposal came up. +1 from me in general, but I have some quibbles with details.

...

2. That we make Semigroup a superclass of Monoid, so that (minimally):

class Semigroup a where (<>) :: a -> a -> a

class Semigroup a => Monoid a where mempty :: a mconcat :: [a] -> a mconcat = foldr (<>) mempty

mappend :: Semigroup a => a -> a -> a mappend = (<>)

+1, though I'd prefer to leave mappend restricted to the Monoid class. In the long term, I'd rather have it killed off than kept as a synonym for (<>). Besides, it's m(onoid)append.

...

3. (Optional, recommended) There are other useful functions that can be added to Semigroup, such as sconcat and times1p, but I will let Edward speak to whether those should be proposed at this time.

They should be added in now or never, there's no reason to break compatibility twice. I don't think times1p can be accepted in its current form as it depends on a different library. Edward, can you make a concrete proposal for these?

...

4. (Optional, recommended) That we fix the Monoid instance for Maybe to be:

instance Semigroup a => Monoid (Maybe a) where mempty = Nothing

...

instance Semigroup a => Semigroup (Maybe a) where Just x <> Just y = Just (x <> y) Nothing <> x = x x <> Nothing = x

Edward Kmett

5:50 p.m.

I find myself somewhat on the fence about this proposal. Ultimately, I'd like to see semigroups in base. That said doing so with the current changes in flight is somewhat awkward, and introducing it carries a lot of distributed work. We have two outstanding proposals that have received overwhelming support: namely the Applicative => Monad (AMP) proposal and the generalization of the types in the Prelude for mapM, etc. to use Foldable and Traversable (FT). Between these two proposals, we wind up with the need to bring in Foldable, Traversable, Applicative and Monoid into the Prelude. Applicative as a superclass of Monad and for traverse and Monoid because Foldable brings in foldMap. Consequently fixing the situation to make Semigroup a superclass of Monoid isn't something that can be done purely by adding a module to base, but it would infect Prelude. * * This makes it have to clear a much higher bar! A real proposal that involved adding Semigroup as a superclass of Monoid could possibly be concocted like David's AMP patch, whereupon Monoid would be given a default mappend = (<>) That could enable users to just have to write an extra instance line like they would for the AMP, but it has some immediate headaches. In particular (<>) has been in use in pretty printing libraries since time immemorial. Suddenly exporting a version from the Prelude is likely to be a fairly breaking change to those libraries as the associativity of the (<>) provided by Data.Monoid (and the one provided by Data.Semigroup) both disagree with the (<>) in Text.PrettyPrint.HughesPJ, which is in the platform already. This could lead to some rather annoying breakages -- worse, silent breakages. I seem to recall that there was some concern that this would change the output of some pretty printing in GHC when the associativity issue was last raised on this list over the introduction of (<>). I think to have a serious proposal that could actually be accepted, a lot of work would have to be put into a patch and then it would need to be explored how much the associativity issue bites users in practice along with gauging the amount of effort that would be involved in retrofitting semigroup instances into virtually everyone's libraries. My knee jerk reaction is to try to put this off until the dust has settled from the AMP and FT changes and push back with a request for the necessary analysis. -Edward On Tue, Jun 11, 2013 at 5:05 PM, Mario Blažević wrote:

...

On 06/11/13 14:46, John Wiegley wrote:

...
1. I propose that we add the following package to base:

http://hackage.haskell.org/**packages/archive/semigroups/0.** 9.2/doc/html/Data-Semigroup.**htmlhttp://hackage.haskell.org/packages/archive/semigroups/0.9.2/doc/html/Data-S...

This is somewhat in the spirit of the AMP proposal: further improving the correctness of our algebraic abstractions.

I was wondering how much longer until this proposal came up. +1 from me in general, but I have some quibbles with details.

2. That we make Semigroup a superclass of Monoid, so that (minimally):

...
class Semigroup a where (<>) :: a -> a -> a

class Semigroup a => Monoid a where mempty :: a mconcat :: [a] -> a mconcat = foldr (<>) mempty

mappend :: Semigroup a => a -> a -> a mappend = (<>)

+1, though I'd prefer to leave mappend restricted to the Monoid class. In the long term, I'd rather have it killed off than kept as a synonym for (<>). Besides, it's m(onoid)append.

3. (Optional, recommended) There are other useful functions that can be

...
added to Semigroup, such as sconcat and times1p, but I will let Edward speak to whether those should be proposed at this time.

They should be added in now or never, there's no reason to break compatibility twice. I don't think times1p can be accepted in its current form as it depends on a different library. Edward, can you make a concrete proposal for these?

...
4. (Optional, recommended) That we fix the Monoid instance for Maybe to be:

instance Semigroup a => Monoid (Maybe a) where mempty = Nothing

instance Semigroup a => Semigroup (Maybe a) where

...
Just x <> Just y = Just (x <> y) Nothing <> x = x x <> Nothing = x

+1

______________________________**_________________ Libraries mailing list Libraries@haskell.org http://www.haskell.org/**mailman/listinfo/libraries http://www.haskell.org/mailman/listinfo/libraries

Johan Tibell

6:40 p.m.

New subject: Fwd: Proposal: Add Data.Semigroup to base, as a superclass of Monoid

-1 This has the same problem as making Functor a superclass of Monad, all current instances will break.

...

This is somewhat in the spirit of the AMP proposal: further improving the correctness of our algebraic abstractions.

I don't think building a tower of all possible algebraic abstractions is a useful goal. We should add those that are actually useful (which functors, applicative functions, monads, and monoids have proved to be). I don't want to see us break all current code every time someone decides that we should add another layer (pointed, say) between e.g. functor and monad. On Tue, Jun 11, 2013 at 11:46 AM, John Wiegley wrote:

...

1. I propose that we add the following package to base:

http://hackage.haskell.org/packages/archive/semigroups/0.9.2/doc/html/Data-S...

This is somewhat in the spirit of the AMP proposal: further improving the correctness of our algebraic abstractions.

2. That we make Semigroup a superclass of Monoid, so that (minimally):

class Semigroup a where (<>) :: a -> a -> a

class Semigroup a => Monoid a where mempty :: a mconcat :: [a] -> a mconcat = foldr (<>) mempty

mappend :: Semigroup a => a -> a -> a mappend = (<>)

3. (Optional, recommended) There are other useful functions that can be added to Semigroup, such as sconcat and times1p, but I will let Edward speak to whether those should be proposed at this time.

4. (Optional, recommended) That we fix the Monoid instance for Maybe to be:

instance Semigroup a => Semigroup (Maybe a) where Just x <> Just y = Just (x <> y) _ <> _ = Nothing

instance Semigroup a => Monoid (Maybe a) where mempty = Nothing

For some clarification on what semigroups are and why we'd want to change Monoid, I excerpt here a selection from Brent Yorgey's "Monoids and Variations" paper:

Semigroups

A semigroup is like a monoid without the requirement of an identity element: it consists simply of a set with an associative binary operation....

Of course, any monoid is automatically a semigroup (by forgetting about its identity element). In the other direction, to turn a semigroup into a monoid, simply add a new distinguished element to serve as the identity, and extend the definition of the binary operation appropriately. This creates an identity element by definition, and it is not hard to see that it preserves associativity....

Adding a new distinguished element to a type is typically accomplished by wrapping it in Maybe. One might therefore expect to turn an instance of Semigroup into an instance of Monoid by wrapping it in Maybe. Sadly, Data.Monoid does not define semigroups, and has a Monoid instance for Maybe which requires a Monoid constraint on its argument type...

This is somewhat odd: in essence, it ignores the identity element of [the type] and replaces it with a different one.

-- John Wiegley FP Complete Haskell tools, training and consulting http://fpcomplete.com johnw on #haskell/irc.freenode.net

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

Nikita Volkov

12 Jun 12 Jun

5:03 a.m.

On Jun 12, 2013, at 2:40 AM, Johan Tibell wrote:

...

I don't want to see us break all current code every time someone decides that we should add another layer (pointed, say) between e.g. functor and monad.

This can also be considered a good argument to make as many of the breaking changes as possible at once. This is why I believe we should initiate polls on all imaginable initiatives, including the Pointed. My votes for Semigroup and Pointed are neutral, i.e. 0 and 0. Concerning the discussion on Monoid's <>. Since we've stepped on a path of Prelude generalization, what about reimplementing the ++ in terms of Monoid? It's proven to be quite useful in the Snoyman's (of Yesod) classy-prelude experiment. Just think about it: we don't lose anything, and we get the expected String-like API for working with Text, ByteString and etc, and we also free up the <> operator.

Andreas Abel

4:13 p.m.

On 12.06.13 11:03 AM, Nikita Volkov wrote:

...

Concerning the discussion on Monoid's <>. Since we've stepped on a path of Prelude generalization, what about reimplementing the ++ in terms of Monoid? It's proven to be quite useful in the Snoyman's (of Yesod) classy-prelude http://hackage.haskell.org/packages/archive/classy-prelude/0.5.8/doc/html/Cl... experiment. Just think about it: we don't lose anything, and we get the expected String-like API for working with Text, ByteString and etc, and we also free up the <> operator.

Makes sense. +1 for ++ instead of <>. Both are right-associative, so there are no problems with that. -- Andreas Abel <>< Du bist der geliebte Mensch. Theoretical Computer Science, University of Munich Oettingenstr. 67, D-80538 Munich, GERMANY andreas.abel@ifi.lmu.de http://www2.tcs.ifi.lmu.de/~abel/

Stephen Tetley

13 Jun 13 Jun

1:05 p.m.

-1 I'm against adding Semigroup to Base as it only has one useful operation in the class and no useful generic functions (without adding more overhead to Base - non-empty lists, naturals...).

Petr Pudlák

2:20 p.m.

Dne 06/13/2013 07:05 PM, Stephen Tetley napsal(a):

...

-1

I'm against adding Semigroup to Base as it only has one useful operation in the class and no useful generic functions (without adding more overhead to Base - non-empty lists, naturals...). The most useful function from Semigroups for me is `times1p` (or its monoidal variant `timesN`). What about adding a function similar to `timesN` (perhaps with another name) to Prelude (and implement `(^)` with it)?

Petr

...

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

harry

11 Aug 11 Aug

9:37 a.m.

There are two significant changes proposed here: Add Data.Semigroup to base Make Monoid a subclass of Semigroup Perhaps a smoother path would be to first add Semigroup to base (minimal impact on existing code), then make Monoid a subclass at a later date? The advantages of this are: People will stop reinventing the NonEmpty wheel, because they don't know it exists or isn't worth creating a new dependency for. Adding a package to base is much less intrusive than changing the hierarchy of an existing class.. -- View this message in context: http://haskell.1045720.n5.nabble.com/Proposal-Add-Data-Semigroup-to-base-as-... Sent from the Haskell - Libraries mailing list archive at Nabble.com.

3935

Age (days ago)

4361

Last active (days ago)

List overview

Download

9 comments

9 participants

participants (9)

Andreas Abel
Edward Kmett
harry
Johan Tibell
John Wiegley
Mario Blažević
Nikita Volkov
Petr Pudlák
Stephen Tetley