- template-haskell - Haskell.org

RE: [Template-haskell] Instance declarations
by Simon Peyton-Jones 19 Feb '03

19 Feb '03

Can you give the actual source code for the offending program, please? Simon | -----Original Message----- | From: Derek Elkins [mailto:ddarius86@hotmail.com] | Sent: 13 February 2003 20:29 | To: template-haskell(a)haskell.org | Subject: [Template-haskell] Instance declarations | | My first email seems to have gotten lost, if it meanders it's way here, then | sorry for the repeat. | | Anyways, generating a set of pattern matching functions in an instance | declaration (maybe in general, though I haven't tested it) causes a | "conflicting definitions error". Rewriting as a case expression is one | work-around. | | E.g. | | instance (NFData a) => NFData (Tree a) where | rnf (Leaf a) = rnf a | rnf (Branch a b) = rnf a `seq` rnf b | | doesn't work, but rewriting as | | instance (NFData a) => NFData (Tree a) where | rnf x = case x of | (Leaf a) -> rnf a | (Branch a b) -> rnf a `seq` rnf b | | does. | | I suspect that it may have to do with the generated code being considered at | a lower level (i.e. the transformation of the above into the below is | already considered done). It may just be I'm creating the functions | incorrectly, so if you can't get the error to happen I can provide the code | I'm using. | | Separately: I've written a variety of fold algebras over data declarations. | I'm not sure how useful they'll be, but if you're interested email me, and | I'll email the source. I haven't tested them, but it's pretty mechanical | code so I don't suspect any major problems and any should be relatively | straightforward to fix. See `Dealing with Large Bananas` for the concept | (and for documentation, as it has no documentation yet). | | _________________________________________________________________ | STOP MORE SPAM with the new MSN 8 and get 2 months FREE* | http://join.msn.com/?page=features/junkmail | | _______________________________________________ | template-haskell mailing list | template-haskell(a)haskell.org | http://www.haskell.org/mailman/listinfo/template-haskell

1 0

Re: [Template-haskell] Record constructors
by Derek Elkins 18 Feb '03

18 Feb '03

>From: Ian Lynagh <igloo(a)earth.li> >To: Simon Peyton-Jones <simonpj(a)microsoft.com> >CC: template-haskell(a)haskell.org >Subject: Re: [Template-haskell] Record constructors >Date: Mon, 17 Feb 2003 15:32:46 +0000 > >On Mon, Feb 17, 2003 at 03:15:27PM -0000, Simon Peyton-Jones wrote: > > > > Now I'm confused. > > > > Ian: you are the one who's going to have a go at this, correct? > >Sorry, misread Derek's reply. Will do. > > >Thanks >Ian I've made the changes for the contexts on datatypes and records constructors. They seem to work all right. I just need to make a patch and send it somewhere. I could add the strictness stuff if you want. I know pretty much what would need to change (or at least part of it). Another bug I might go after is deriving seems to generate multiple instances. E.g. $([d| data X = X String deriving Show |]) causes a duplicate instance declaration error and panics when compiling. _________________________________________________________________ The new MSN 8: smart spam protection and 2 months FREE* http://join.msn.com/?page=features/junkmail

1 0

RE: [Template-haskell] Record constructors
by Simon Peyton-Jones 17 Feb '03

17 Feb '03

Now I'm confused. Ian: you are the one who's going to have a go at this, correct? Re you qn about strictness annotations, what you suggest looks fine. Simon | -----Original Message----- | From: Ian Lynagh [mailto:igloo@earth.li] | Sent: 17 February 2003 14:38 | To: template-haskell(a)haskell.org | Subject: Re: [Template-haskell] Record constructors | | On Fri, Feb 14, 2003 at 02:35:28PM -0500, Derek Elkins wrote: | > >From: "Simon Peyton-Jones" <simonpj(a)microsoft.com> | > >To: "Derek Elkins" <ddarius86(a)hotmail.com>,<template-haskell(a)haskell.org> | > >Subject: RE: [Template-haskell] Record constructors | > >Date: Fri, 14 Feb 2003 10:18:31 -0000 | > > | > >Good point. Indeed, THSyntax has no provision for record types and | > >record expressions. | > >Haskell is a pretty large language and I didn't cover the whole of it in | > >the first pass, I'm afraid. | > > | > > | > > IAN L, would you care to put in the THSyntax for | > > records | > > contexts in data type declarations | > > and add suitable stuff to DsMeta and Convert to get them to and | > >fro? | > > | > >(Or someone else!) I'm too much of a bottleneck for stuff like this. | > >I'm inclined to focus my time on things where it really has to be me. | > > | > >Simon | > | > Okay. I'll take a swing at it. | | OK - if you have problems I will try to help. | | Is there a reason why strictness annotations are ignored, or just the | first pass issue again? | Presumably any StrictnessMark except NotMarkedStrict should be reported | as strict? I think | data Con = Constr String [(Strictness, Typ)] | data Strictness = Strict | NonStrict | is probably as good a representation as any other. | | Any objections? | | | Thanks | Ian | | _______________________________________________ | template-haskell mailing list | template-haskell(a)haskell.org | http://www.haskell.org/mailman/listinfo/template-haskell

2 1

RE: [Template-haskell] Record constructors
by Derek Elkins 17 Feb '03

17 Feb '03

>From: "Simon Peyton-Jones" <simonpj(a)microsoft.com> >To: "Derek Elkins" <ddarius86(a)hotmail.com>,<template-haskell(a)haskell.org> >Subject: RE: [Template-haskell] Record constructors >Date: Fri, 14 Feb 2003 10:18:31 -0000 > >Good point. Indeed, THSyntax has no provision for record types and >record expressions. >Haskell is a pretty large language and I didn't cover the whole of it in >the first pass, I'm afraid. > > > IAN L, would you care to put in the THSyntax for > records > contexts in data type declarations > and add suitable stuff to DsMeta and Convert to get them to and >fro? > >(Or someone else!) I'm too much of a bottleneck for stuff like this. >I'm inclined to focus my time on things where it really has to be me. > >Simon Okay. I'll take a swing at it. I've made all the changes I think are necessary for contexts on datatypes, but I won't know until I recompile. For record constructors, I'm thinking of adding `RecConstr String [(String, Typ)]` to Con in THSyntax. This is slightly modified from the HsSyn version `HsRecDecl SrcLoc HsName [([HsName],HsBangType)]` which I suppose is setup to handle `foo,bar :: a -> b` I don't see this as very convenient for generation or reification. So unless I'm missing something or another layout is preferable, that's the setup I'm going with. _________________________________________________________________ Tired of spam? Get advanced junk mail protection with MSN 8. http://join.msn.com/?page=features/junkmail

2 1

RE: Declaration reordering
by Simon Peyton-Jones 17 Feb '03

17 Feb '03

I'm not sure I do agree that it's worthwhile. At least, of course it would be nice but the question is whether it's worth the bother. Why are you keen on it? You sent your original message to ghc-users; perhaps you'd like to summarise for the TH list. Your first option looks best. Indeed, each decl has its SrcLoc which includes a line number, so perhaps you could unscramble them that way. You are welcome to have a go; if you need help figuring out where, I'll help. Simon | -----Original Message----- | From: Ian Lynagh [mailto:igloo@earth.li] | Sent: 14 February 2003 03:11 | To: Simon Peyton-Jones | Subject: Re: Declaration reordering | | On Thu, Feb 13, 2003 at 09:16:46AM -0000, Simon Peyton-Jones wrote: | > | > The renamer performs a strongly-connected component analysis of | > declarations. Reason: it's the standard way to get maximal | > polymorphism. It's even part of the language specification. | > | > It'd be quite tricky to un-do that stuff before doing the quotation | > part. So I'm really not sure how to tackle your problem. | | Do you agree that it is worthwhile? | | The possibilities that spring to mind are: | | * tagging each declaration with its position in the list and sorting | them before returning them. The only arguments against this are it | adds a bit of code and will have a time/space impact (but I'd have | thought negligible, especially if the tagging was done strictly). | | * Keeping a copy of the original input around to pass back. Doubles | space used by declaration lists and I suspect will have problems with | right infix operators. | | * Doing the SCC analysis and type checking on a copy and throwing it | away afterwards. Would also have the problems of the above and is | probably more complicated to implement. | | | Thanks | Ian

2 1

RE: [Template-haskell] Record constructors
by Simon Peyton-Jones 14 Feb '03

14 Feb '03

Good point. Indeed, THSyntax has no provision for record types and record expressions. Haskell is a pretty large language and I didn't cover the whole of it in the first pass, I'm afraid. IAN L, would you care to put in the THSyntax for records contexts in data type declarations and add suitable stuff to DsMeta and Convert to get them to and fro? (Or someone else!) I'm too much of a bottleneck for stuff like this. I'm inclined to focus my time on things where it really has to be me. Simon | -----Original Message----- | From: Derek Elkins [mailto:ddarius86@hotmail.com] | Sent: 14 February 2003 03:09 | To: template-haskell(a)haskell.org | Subject: [Template-haskell] Record constructors | | Quasiquoted expressions involving record construction syntax dies. | Furthermore, how do you even generate a record constructor declaration? | There's no place to put the record name. | | module Main where | import Language.Haskell.THSyntax | import Text.PrettyPrint.HughesPJ (render) | | data Lang a b = Lang { | tvarAlg :: String -> a, | arrowAlg :: a -> a -> a, | varAlg :: String -> b, | applyAlg :: b -> b -> b, | lambdaAlg :: (String, a) -> b -> b | } | | e = [| let t = Lang { tvarAlg = \s -> (), | arrowAlg = \x y -> (), | varAlg = \s -> (), | applyAlg = \x y -> (), | lambdaAlg = \x y -> () | } | in tvarAlg t |] | | main = do output <- runQ e | putStrLn $ render $ pprExp output | putStrLn "\n" | putStrLn $ show output | | _________________________________________________________________ | Add photos to your messages with MSN 8. Get 2 months FREE*. | http://join.msn.com/?page=features/featuredemail | | _______________________________________________ | template-haskell mailing list | template-haskell(a)haskell.org | http://www.haskell.org/mailman/listinfo/template-haskell

1 0

Record constructors
by Derek Elkins 14 Feb '03

14 Feb '03

Quasiquoted expressions involving record construction syntax dies. Furthermore, how do you even generate a record constructor declaration? There's no place to put the record name. module Main where import Language.Haskell.THSyntax import Text.PrettyPrint.HughesPJ (render) data Lang a b = Lang { tvarAlg :: String -> a, arrowAlg :: a -> a -> a, varAlg :: String -> b, applyAlg :: b -> b -> b, lambdaAlg :: (String, a) -> b -> b } e = [| let t = Lang { tvarAlg = \s -> (), arrowAlg = \x y -> (), varAlg = \s -> (), applyAlg = \x y -> (), lambdaAlg = \x y -> () } in tvarAlg t |] main = do output <- runQ e putStrLn $ render $ pprExp output putStrLn "\n" putStrLn $ show output _________________________________________________________________ Add photos to your messages with MSN 8. Get 2 months FREE*. http://join.msn.com/?page=features/featuredemail

1 0

Instance declarations
by Derek Elkins 13 Feb '03

13 Feb '03

My first email seems to have gotten lost, if it meanders it's way here, then sorry for the repeat. Anyways, generating a set of pattern matching functions in an instance declaration (maybe in general, though I haven't tested it) causes a "conflicting definitions error". Rewriting as a case expression is one work-around. E.g. instance (NFData a) => NFData (Tree a) where rnf (Leaf a) = rnf a rnf (Branch a b) = rnf a `seq` rnf b doesn't work, but rewriting as instance (NFData a) => NFData (Tree a) where rnf x = case x of (Leaf a) -> rnf a (Branch a b) -> rnf a `seq` rnf b does. I suspect that it may have to do with the generated code being considered at a lower level (i.e. the transformation of the above into the below is already considered done). It may just be I'm creating the functions incorrectly, so if you can't get the error to happen I can provide the code I'm using. Separately: I've written a variety of fold algebras over data declarations. I'm not sure how useful they'll be, but if you're interested email me, and I'll email the source. I haven't tested them, but it's pretty mechanical code so I don't suspect any major problems and any should be relatively straightforward to fix. See `Dealing with Large Bananas` for the concept (and for documentation, as it has no documentation yet). _________________________________________________________________ STOP MORE SPAM with the new MSN 8 and get 2 months FREE* http://join.msn.com/?page=features/junkmail

1 0

instance declaration issue
by Derek Elkins 12 Feb '03

12 Feb '03

Generating a set of pattern matched function declarations in an instance declaration (see below) causes a "Conflicting definition error" upon linking. instance (G a) => F (Tree a) where f (Leaf x) = g x f (Branch x y) = g x `h` g y rewritten as instance (G a) => F (Tree a) where f x = case x of Leaf x -> g x Branch x y -> g x `h` g y works. I suspect that it has to do with the "level" the source code is considered. I.e. the transformation from the above into the below is already considered done by the compiler. The problem (it it's a problem) may be in the way I'm generating the code. If you can't get it to happen, I'll email my code. _________________________________________________________________ Help STOP SPAM with the new MSN 8 and get 2 months FREE* http://join.msn.com/?page=features/junkmail

1 0

RE: [Template-haskell] Contexts on Datatype - question/bug?
by Derek Elkins 12 Feb '03

12 Feb '03

>| On a side note: why not toss up a TH page on the Haskell wiki, >especially if >| the answer to my second question is no? > >by all means go ahead. Consider it done: http://www.haskell.org/wiki/wiki?TemplateHaskell It's pretty sparse now. >| data (Eq a) => Tree a = Leaf a | Branch (Tree a) (Tree a) >| >| main = do output <- runQ $ reifyDecl Tree >| putStr $ show output >| >| will cause the impossible to happen. >| I went in and changed DsMeta.hs to pass the cxt for reifyDecl >(repTyClD for >| case TyData), which kept the impossible from happening, but then >realized >| (at least partially) why it wasn't there before. THSyntax.Data >doesn't >| support contexts. >| >| Except that a diagnostic rather than death would be preferable, is >this an >| omission, on a todo list, or purposely left out? > >I think Contexts in data declarations are a mis-feature of Haskell (i.e. >they shouldn't be there in the first place). But I suppose that since >they are there, TH had better support them. To do! > >Simon I suspected something like that (hence suggesting purposely left out). This isn't anything that was causing me problems. If no one else wants it, I'd be more than content with just an explanatory warning message: "warning: contexts on datatypes not supported just because" ;). Or just silently accept them. _________________________________________________________________ The new MSN 8: smart spam protection and 2 months FREE* http://join.msn.com/?page=features/junkmail

1 0