I fail to see how this doesn't help lens, unless we're assuming no buy-in from class declarations. Also, your approach would require c*n pragmas to be declared, whereas mine only requires c. Also your method seems to require having both the class and type in scope, in which case one could simply declare the instance in that module anyway. On Mon, Oct 20, 2014 at 9:29 AM, David Feuer wrote:

I don't think your approach is flexible enough to accomplish the purpose. For example, it does almost nothing to help lens. Even my approach should, arguably, be extended transitively, allowing the named module to delegate that authority, but such an extension could easily be put off till later. On Oct 19, 2014 7:17 PM, "John Lato" wrote:

...

Thinking about this, I came to a slightly different scheme. What if we instead add a pragma:

{-# OrphanModule ClassName ModuleName #-}

and furthermore require that, if OrphanModule is specified, all instances can *only* appear in the module where the class is defined, the involved types are defined, or the given OrphanModule? We would also need to add support for the compiler to understand that multiple modules may appear under the same name, which might be a bit tricky to implement, but I think it's feasible (perhaps in a restricted manner).

I think I'd prefer this when implementing orphan instances, and probably when writing the pragmas as well.

On Mon, Oct 20, 2014 at 1:02 AM, David Feuer wrote:

...

Orphan instances are bad. The standard approach to avoiding the orphan hazard is to always put an instance declaration in the module that declares the type or the one that declares the class. Unfortunately, this forces packages like lens to have an ungodly number of dependencies. Yesterday, I had a simple germ of an idea for solving this (fairly narrow) problem, at least in some cases: allow a programmer to declare where an instance declaration must be. I have no sense of sane syntax, but the rough idea is:

{-# InstanceIn NamedModule [Context =>] C1 T1 [T2 ...] #-}

This pragma would appear in a module declaring a class or type. The named module would not have to be available, either now or ever, but attempting to declare such an instance in any module *other* than the named one would be an error by default, with a flag -XAllowForbiddenInstancesAndInviteNasalDemons to turn it off. The optional context allows multiple such pragmas to appear in the type/class-declaring modules, to allow overlapping instances (all of them declared in advance).

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

In the past I've spent some time thinking about the orphan instances problem. I concluded that the Right Thing to do is to turn instances into first-class citizens and allow them to be explicitly imported and exported. I think devising pragmas is a workaround, not a solution. Janek Dnia poniedziałek, 20 października 2014, David Feuer napisał:

OK, so first off, I don't have anything against your pragma; I just think that something akin to mine would be good to have too. Mine was not intended to require both class and type to be in scope; if one of them is not, then it should be given its full name:

{-# InstanceIn Module Foo.Class Type #-} {-# InstanceIn Module Class Bar.Type #-}

As Edward Kmett explained to me, there are reasons for module authors not to want to include instances for lens stuff—in particular, they apparently tend to use a lot of non-portable code, but even aside from that, they may just not want to have to deal with maintaining that particular code. This leads to a slew of instances being dumped into lens modules, forcing the lens package to depend on a bunch of others. What I'm suggesting is that sticking {-# InstanceIn Data.Text.Lens Strict Data.Text.Lazy.Text Data.Text.Text #-} into Control.Lens.Iso (and so on) would allow Data.Text.Lens to be broken off into a separate package, removing the text dependency from lens.

Note also: I described a way to (try to) support overlapping instances for mine, but I think it would be valuable to offer mine even without that feature (dropping the context stuff), if it's just too complex.

On Sun, Oct 19, 2014 at 9:43 PM, John Lato wrote:

...

I fail to see how this doesn't help lens, unless we're assuming no buy-in from class declarations. Also, your approach would require c*n pragmas to be declared, whereas mine only requires c. Also your method seems to require having both the class and type in scope, in which case one could simply declare the instance in that module anyway.

On Mon, Oct 20, 2014 at 9:29 AM, David Feuer

wrote:

...

I don't think your approach is flexible enough to accomplish the purpose. For example, it does almost nothing to help lens. Even my approach should, arguably, be extended transitively, allowing the named module to delegate that authority, but such an extension could easily be put off till later.

On Oct 19, 2014 7:17 PM, "John Lato" wrote:

...

Thinking about this, I came to a slightly different scheme. What if we instead add a pragma:

{-# OrphanModule ClassName ModuleName #-}

and furthermore require that, if OrphanModule is specified, all instances can *only* appear in the module where the class is defined, the involved types are defined, or the given OrphanModule? We would also need to add support for the compiler to understand that multiple modules may appear under the same name, which might be a bit tricky to implement, but I think it's feasible (perhaps in a restricted manner).

I think I'd prefer this when implementing orphan instances, and probably when writing the pragmas as well.

On Mon, Oct 20, 2014 at 1:02 AM, David Feuer

wrote:

...

Orphan instances are bad. The standard approach to avoiding the orphan hazard is to always put an instance declaration in the module that declares the type or the one that declares the class. Unfortunately, this forces packages like lens to have an ungodly number of dependencies. Yesterday, I had a simple germ of an idea for solving this (fairly narrow) problem, at least in some cases: allow a programmer to declare where an instance declaration must be. I have no sense of sane syntax, but the rough idea is:

{-# InstanceIn NamedModule [Context =>] C1 T1 [T2 ...] #-}

This pragma would appear in a module declaring a class or type. The named module would not have to be available, either now or ever, but attempting to declare such an instance in any module *other* than the named one would be an error by default, with a flag -XAllowForbiddenInstancesAndInviteNasalDemons to turn it off. The optional context allows multiple such pragmas to appear in the type/class-declaring modules, to allow overlapping instances (all of them declared in advance).

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

As I said before, it still doesn't solve the problem I'm trying to solve. Look at a package like criterion, for example. criterion depends on aeson. Why? Because statistics depends on it. Why? Because statistics wants a couple types it defines to be instances of classes defined in aeson. John Lato's proposal would require the pragma to appear in the relevant aeson module, and would prevent *anyone* else from defining instances of those classes. With my proposal, statistics would be able to declare {-# InstanceIn Statistics.AesonInstances AesonModule.AesonClass StatisticsType #-} Then it would split the Statistics.AesonInstances module off into a statistics-aeson package and accomplish its objective without stepping on anyone else. We'd get a lot more (mostly tiny) packages, but in exchange the dependencies would get much thinner. On Oct 21, 2014 11:52 AM, "Stephen Paul Weber" wrote:

Somebody claiming to be John Lato wrote:

...

Thinking about this, I came to a slightly different scheme. What if we instead add a pragma:

{-# OrphanModule ClassName ModuleName #-}

I really like this. It solve all the real orphan instance cases I've had in my libraries.

-- Stephen Paul Weber, @singpolyma See http://singpolyma.net for how I prefer to be contacted edition right joseph

On Oct 21, 2014 1:22 PM, "John Lato" wrote:

Perhaps you misunderstood my proposal if you think it would prevent

anyone else from defining instances of those classes? Part of the proposal was also adding support to the compiler to allow for a multiple files to use a single module name. That may be a larger technical challenge, but I think it's achievable. You are right; I definitely did not realize this. What happens when files using the same module name both define instances for the same class and type(s)? I don't know nearly enough about how these things work to know if there's a nice way to catch this. Could you explain a bit more about how it would work? Also, what exactly would be in scope in each of these? Would adding a file to the module necessitate recompilation of everything depending on it?

I think one key difference is that my proposal puts the onus on class implementors, and David's puts the onus on datatype implementors, so they certainly are complementary and could co-exist.

Mine puts the onus on either, actually, to support both the pattern of a maintainer maintaining a class with instances and of one maintaining a type with instances. To a certain extent these could even be mixed. For example, a module in base could delegate a number of instances of a certain class, but we wouldn't want pragmas relating to Hackagy types in there. One nice thing about my approach is that any program that's correct *with* the pragma is also correct *without* it—it's entirely negative. In particular, if someone should come up with a broader/better/ultimate solution to the orphan instance problem, the pragma could just go away without breaking anything. Something using multiple files to define one module inherently requires more support from the future.

It seems that my previous mail went unnoticed. Perhaps because I didn't provide enough justification for my solution. I'll try to make up for that now. First of all let's remind ourselves why orphan instances are a problem. Let's say package A defines some data types and package B defines some type classes. Now, package C might make data types from A instances of type classes from B. Someone who imports C will have these instances in scope. But since C defines neither the data types nor the type classes it might be surprising for the user of C that C makes A data types instances of B type classes. So we issue a warning that this is potentially dangerous. Of course person implementing C might suppress these warnings so the user of C can end up with unexpected instances without knowing anything. I feel that devising some sort of pragmas to define which orphan instances are allowed does not address the heart of the problem. And the heart of the problem is that we can't control importing and exporting of instances. Pragmas are just a workaround, not a real solution. It would be much better if we could just write this (warning, half-baked idea ahead): module BazModule ( instance Bar Foo ) where import FooModule (Foo (...)) -- import Foo data type from FooModule import BarModule (class Bar) -- import class Bar from BazModule instance Bar Foo ... And then someone importing BazModule can decide to import the instance: module User where import FooModule (Foo(..)) import BarModule (class Bar) import BazModule (instance Bar Foo) Of course requiring that classes and instances are exported and imported just like everything else would be a backawrds incompatible change and would therefore require effort similar to AMP proposal, ie. first release GHC version that warns about upcoming change and only enforce the change some time later. Janek Dnia wtorek, 21 października 2014, RodLogic napisał:

One other benefit of multiple files to use a single module name is that it would be easy to separate testing code from real code even when testing internal/non-exported functions.

On Tue, Oct 21, 2014 at 1:22 PM, John Lato wrote:

...

Perhaps you misunderstood my proposal if you think it would prevent anyone else from defining instances of those classes? Part of the proposal was also adding support to the compiler to allow for a multiple files to use a single module name. That may be a larger technical challenge, but I think it's achievable.

I think one key difference is that my proposal puts the onus on class implementors, and David's puts the onus on datatype implementors, so they certainly are complementary and could co-exist.

On Tue, Oct 21, 2014 at 9:11 AM, David Feuer

wrote:

...

As I said before, it still doesn't solve the problem I'm trying to solve. Look at a package like criterion, for example. criterion depends on aeson. Why? Because statistics depends on it. Why? Because statistics wants a couple types it defines to be instances of classes defined in aeson. John Lato's proposal would require the pragma to appear in the relevant aeson module, and would prevent *anyone* else from defining instances of those classes. With my proposal, statistics would be able to declare

{-# InstanceIn Statistics.AesonInstances AesonModule.AesonClass StatisticsType #-}

Then it would split the Statistics.AesonInstances module off into a statistics-aeson package and accomplish its objective without stepping on anyone else. We'd get a lot more (mostly tiny) packages, but in exchange the dependencies would get much thinner. On Oct 21, 2014 11:52 AM, "Stephen Paul Weber"

wrote:

...

Somebody claiming to be John Lato wrote:

...

Thinking about this, I came to a slightly different scheme. What if we instead add a pragma:

{-# OrphanModule ClassName ModuleName #-}

I really like this. It solve all the real orphan instance cases I've had in my libraries.

-- Stephen Paul Weber, @singpolyma See http://singpolyma.net for how I prefer to be contacted edition right joseph

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

These are certainly good points and I'm far from claiming that I have solved all the potential problems that may arise (if I had I would probably be implementing this right now). But I still believe that pragmas are not a good solution, while control of imports and exports is. Unless the problems turn out to be impossible to overcome. Janek Dnia środa, 22 października 2014, David Feuer napisał:

You're not the first one to come up with this idea (and I don't know who is). Unfortunately, there are some complications. I'm pretty sure there are simpler examples than this, but this is what I could think of. Suppose we have

module PotatoModule (Root (..), T (..)) where -- Does not export instance Root T class Root t where cook :: t -> String

data T = T data Weird :: * -> * where Weird :: Root t => t -> Weird t

instance Root T where cook T = "Boil, then eat straight out of the pot."

potato :: Weird T potato = Weird T

-- --------------

module ParsnipModule where import PotatoModule

instance Root T where cook T = "Slice into wedges or rounds and put in the soup."

parsnip :: Weird T parsnip = Weird T

mash :: Weird t -> Weird t -> String mash (Weird x) (Weird y) = cook x ++ cook y

mush :: String mush = mash potato parsnip

-- --------------

OK, so what happens when we compile mash? Well, we have a bit of a problem! When we mash the potato and the parsnip, the mash function gets access to two different dictionaries for Root T, and two values of type T. There is absolutely nothing to indicate whether we should use the dictionary that's "in the air" because Root T has an instance in ParsnipModule, the dictionary that we pull out of parsnip (which is the same), or the dictionary we pull out of potato (which is different). I think inlining and specialization will make things even stranger and less predictable. In particular, the story of what goes on with inlining gets much harder to understand at the Haskell level: if mash and mush are put into a third module, and potato and parsnip are inlined there, that becomes a type error, because there's no visible Root T instance there!

On Wed, Oct 22, 2014 at 12:56 PM, Jan Stolarek

wrote:

...

It seems that my previous mail went unnoticed. Perhaps because I didn't provide enough justification for my solution. I'll try to make up for that now.

First of all let's remind ourselves why orphan instances are a problem. Let's say package A defines some data types and package B defines some type classes. Now, package C might make data types from A instances of type classes from B. Someone who imports C will have these instances in scope. But since C defines neither the data types nor the type classes it might be surprising for the user of C that C makes A data types instances of B type classes. So we issue a warning that this is potentially dangerous. Of course person implementing C might suppress these warnings so the user of C can end up with unexpected instances without knowing anything.

I feel that devising some sort of pragmas to define which orphan instances are allowed does not address the heart of the problem. And the heart of the problem is that we can't control importing and exporting of instances. Pragmas are just a workaround, not a real solution. It would be much better if we could just write this (warning, half-baked idea ahead):

module BazModule ( instance Bar Foo ) where

import FooModule (Foo (...)) -- import Foo data type from FooModule import BarModule (class Bar) -- import class Bar from BazModule

instance Bar Foo ...

And then someone importing BazModule can decide to import the instance:

module User where import FooModule (Foo(..)) import BarModule (class Bar) import BazModule (instance Bar Foo)

Of course requiring that classes and instances are exported and imported just like everything else would be a backawrds incompatible change and would therefore require effort similar to AMP proposal, ie. first release GHC version that warns about upcoming change and only enforce the change some time later.

Janek

Dnia wtorek, 21 października 2014, RodLogic napisał:

...

One other benefit of multiple files to use a single module name is that

it

...

would be easy to separate testing code from real code even when testing internal/non-exported functions.

On Tue, Oct 21, 2014 at 1:22 PM, John Lato wrote:

...

Perhaps you misunderstood my proposal if you think it would prevent anyone else from defining instances of those classes? Part of the proposal was also adding support to the compiler to allow for a

multiple

...

...

files to use a single module name. That may be a larger technical challenge, but I think it's achievable.

I think one key difference is that my proposal puts the onus on class implementors, and David's puts the onus on datatype implementors, so

they

...

...

certainly are complementary and could co-exist.

On Tue, Oct 21, 2014 at 9:11 AM, David Feuer

wrote:

...

As I said before, it still doesn't solve the problem I'm trying to solve. Look at a package like criterion, for example. criterion

depends

...

...

...

on aeson. Why? Because statistics depends on it. Why? Because

statistics

...

...

...

wants a couple types it defines to be instances of classes defined in aeson. John Lato's proposal would require the pragma to appear in the relevant aeson module, and would prevent *anyone* else from defining instances of those classes. With my proposal, statistics would be able to declare

{-# InstanceIn Statistics.AesonInstances AesonModule.AesonClass StatisticsType #-}

Then it would split the Statistics.AesonInstances module off into a statistics-aeson package and accomplish its objective without stepping on anyone else. We'd get a lot more (mostly tiny) packages, but in exchange the dependencies would get much thinner. On Oct 21, 2014 11:52 AM, "Stephen Paul Weber"

wrote:

...

Somebody claiming to be John Lato wrote: > Thinking about this, I came to a slightly different scheme. What > if we instead add a pragma: > > {-# OrphanModule ClassName ModuleName #-}

I really like this. It solve all the real orphan instance cases I've had in my libraries.

-- Stephen Paul Weber, @singpolyma See http://singpolyma.net for how I prefer to be contacted edition right joseph

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

+1. I have followed the road of trying to enable instances to be imported and exported, without success: a paper that discusses the subject and argues in favour of this support is available at: http://www.dcc.ufmg.br/~camarao/controlling-the-scope-of-instances-in-Haskel... A previous version was rejected by the 2011 Haskell Symposium program committee. Referee reports are attached, since perhaps they can be useful to the discussion. Carlos ---------- Forwarded message ---------- From: Jan Stolarek Date: Wed, Oct 22, 2014 at 2:56 PM Subject: Re: Avoiding the hazards of orphan instances without dependency problems To: ghc-devs@haskell.org Cc: RodLogic , David Feuer It seems that my previous mail went unnoticed. Perhaps because I didn't provide enough justification for my solution. I'll try to make up for that now. First of all let's remind ourselves why orphan instances are a problem. Let's say package A defines some data types and package B defines some type classes. Now, package C might make data types from A instances of type classes from B. Someone who imports C will have these instances in scope. But since C defines neither the data types nor the type classes it might be surprising for the user of C that C makes A data types instances of B type classes. So we issue a warning that this is potentially dangerous. Of course person implementing C might suppress these warnings so the user of C can end up with unexpected instances without knowing anything. I feel that devising some sort of pragmas to define which orphan instances are allowed does not address the heart of the problem. And the heart of the problem is that we can't control importing and exporting of instances. Pragmas are just a workaround, not a real solution. It would be much better if we could just write this (warning, half-baked idea ahead): module BazModule ( instance Bar Foo ) where import FooModule (Foo (...)) -- import Foo data type from FooModule import BarModule (class Bar) -- import class Bar from BazModule instance Bar Foo ... And then someone importing BazModule can decide to import the instance: module User where import FooModule (Foo(..)) import BarModule (class Bar) import BazModule (instance Bar Foo) Of course requiring that classes and instances are exported and imported just like everything else would be a backawrds incompatible change and would therefore require effort similar to AMP proposal, ie. first release GHC version that warns about upcoming change and only enforce the change some time later. Janek Dnia wtorek, 21 października 2014, RodLogic napisał:

One other benefit of multiple files to use a single module name is that it would be easy to separate testing code from real code even when testing internal/non-exported functions.

On Tue, Oct 21, 2014 at 1:22 PM, John Lato wrote:

...

Perhaps you misunderstood my proposal if you think it would prevent anyone else from defining instances of those classes? Part of the proposal was also adding support to the compiler to allow for a multiple files to use a single module name. That may be a larger technical challenge, but I think it's achievable.

I think one key difference is that my proposal puts the onus on class implementors, and David's puts the onus on datatype implementors, so they certainly are complementary and could co-exist.

On Tue, Oct 21, 2014 at 9:11 AM, David Feuer

wrote:

...

As I said before, it still doesn't solve the problem I'm trying to solve. Look at a package like criterion, for example. criterion depends on aeson. Why? Because statistics depends on it. Why? Because statistics wants a couple types it defines to be instances of classes defined in aeson. John Lato's proposal would require the pragma to appear in the relevant aeson module, and would prevent *anyone* else from defining instances of those classes. With my proposal, statistics would be able to declare

{-# InstanceIn Statistics.AesonInstances AesonModule.AesonClass StatisticsType #-}

Then it would split the Statistics.AesonInstances module off into a statistics-aeson package and accomplish its objective without stepping on anyone else. We'd get a lot more (mostly tiny) packages, but in exchange the dependencies would get much thinner. On Oct 21, 2014 11:52 AM, "Stephen Paul Weber"

wrote:

...

Somebody claiming to be John Lato wrote:

...

Thinking about this, I came to a slightly different scheme. What if we instead add a pragma:

{-# OrphanModule ClassName ModuleName #-}

I really like this. It solve all the real orphan instance cases I've had in my libraries.

-- Stephen Paul Weber, @singpolyma See http://singpolyma.net for how I prefer to be contacted edition right joseph

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs