Code Generation for JSFFI/JSADDLE

[epsilonhalbe]

Hi I am studying the code generation - in order to make function calls safe - where necessary.
This ticket is a follow up - to provide a PR for ghcjs/ghcjs-dom#78
I have had a look at the idl-files in webkitgtk-2.17.2, as well as some of hdirect-0.21.

To make the unsafe foreign function calls safe - I guess one would have to modify

domconv-webkit-jsffi.hs
701:           H.HsForeignImport nullLoc (if promise then "javascript interruptible" else "javascript") H.HsUnsafe jsimpl (H.HsIdent defop) tpsig
952:           H.HsForeignImport nullLoc (if promise then "javascript interruptible" else "javascript") H.HsUnsafe jsimpl (H.HsIdent defop) tpsig

(maybe a few more) and change the H.HsUnsafe - dependently if there are *MayThrowException annotations in the idl-files, which are used to genereate the Haskell bindings.

Status of DOMCONV

If I understood correctly there is

domconv-webkit/domconv-webkit-jsffi.hs

889:    constructorRaises | I.ExtAttr (I.Id "ConstructorRaisesException") [] `elem` at = [I.Raises ["RaisesException"]]

domconv-webkit/domconv-webkit-jsaddle.hs

879:    constructorRaises | I.ExtAttr (I.Id "ConstructorRaisesException") [] `elem` at = [I.Raises ["RaisesException"]]

and in the syntax/parser files

domconv-webkit/hdirect-0.21/OmgParser.y
556:getter_decl :: { (Id, [Param], [Raises], Maybe Context)  }
562:setter_decl :: { (Id, [Param], [Raises], Maybe Context)  }
568:deleter_decl :: { (Id, [Param], [Raises], Maybe Context)  }
616:raises_expr :: { Raises }
617:   : RAISES '(' scoped_name_list ')' { Raises (reverse $3) }
618:   | GETTER RAISES '(' scoped_name_list ')' { GetRaises (reverse $4) }
619:   | SETTER RAISES '(' scoped_name_list ')' { SetRaises (reverse $4) }
621:raises_exprs :: { [Raises] }

domconv-webkit/hdirect-0.21/IDLSyn.lhs
30: | Attribute  [Id] Bool Type [Raises] [ExtAttribute]
31: | DictionaryAttribute  [Id] Bool Type [Raises] [ExtAttribute]
32: | Operation  Id Type {-[Param]-} [Raises] (Maybe Context) [ExtAttribute]
139:data Raises = GetRaises [Name] | SetRaises [Name] | Raises [Name] deriving ( Eq, Ord, Show )
140:getterRaises = filter isGet
142:        isGet (SetRaises _) = False
144:setterRaises = filter isSet
146:        isSet (GetRaises _) = False

domconv-webkit/hdirect-0.21/OmgParser.hs
63:        | HappyAbsSyn80 (Maybe Raises)

that relate to exception parsing.

But I do not understand which parts in the idl-files they resemble - I tried to > ack "ConstructorRaisesException" in the webkit-2.17.2 directory (ack) without any results.

Status of IDL-Files

On the other hand searching for MayThrowException which can be found in for example in Websocket.idl.

    [SetterMayThrowException] attribute DOMString binaryType;

    [MayThrowException] void send(ArrayBuffer data);
    [MayThrowException] void send(ArrayBufferView data);
    [MayThrowException] void send(Blob data);
    [MayThrowException] void send(USVString data);

    [MayThrowException] void close(optional [Clamp] unsigned short code, optional DOMString reason);

There is yields no result in the domconv-webkit files.

The only idl files where I found "ConstructorRaisesException - is some ancient webkit repository of adobe adobe/webkit from 2012.

Questions

• Could you give me some info how to modify the parsers to incude the MayThrowException?
• What are the RaisesException statements in the existing parsers for?

[hamishmack]

The things in square brackets like MayThrowException should be parsed as ExtAttribute values. Search the code for ExtAttr for examples of how to check them.

[epsilonhalbe]

ah good - I'll have a go at that thx

[epsilonhalbe]

Hi - I have made a few modifications and want to test them - could you tell me how I would do that?

When I call stack exec -- domconv-webkit-jsffi it tells me to call it with webgtk.idl -Iwebkit-2.17.5/Source/WebCore, but I don't have a webkit.idl - and if I call it with webkitgtk-2.17.2/Source/WebCore/Modules/websockets/WebSocket.idl -I... I get an error about a non-existing ghcjs-dom-jsffi/reexported-modules.txt

[hamishmack]

Sorry, I missed this somehow.

The idea is that you run domconv-webkit-jsffi in a clone of the ghcjs-dom git repo. So it will pick up this webkit.idl file.

When using domconv-webkit-jsaddle you run it in a clone of the jsaddle-dom git repo (it also contains a webkit.idl file.

[epsilonhalbe]

should I provide a pr for ghcjs-dom-jsffi, which uses the new code generator?