All, So far as I know c2hs doesn't support hierarchical modules well. I've not tried the latest version, so apologies if this has already been addressed. Under the Hierarchical Module Namespace Extension http://www.haskell.org/hierarchical-modules/ module names can have dots '.' in them and when mapping these to filenames they are converted to directory separators. Foo.Bar.Baz -> Foo/Bar/Bas(.hs|.chi|etc) I've knocked up a patch that I think solves the problem (works for me), but an expert eye would be appreciated. I've made a change in two places: * CHSLexer.hs: identOrKW function, added '.' as an allowable character in an identifier within binding hooks. * CHS.hs: parseImport function, inserted a call to a new function moduleNameToFileName before calling loadCHI. The new function does the module name to file path conversion. Here's a patch (it's against the gtk2hs fork of c2hs but it's quite small so the idea should be clear) diff -U2 -r1.2 CHSLexer.hs --- CHSLexer.hs 1 Oct 2002 15:17:07 -0000 1.2 +++ CHSLexer.hs 16 Jul 2004 18:52:26 -0000 @@ -557,5 +557,5 @@ -- identOrKW = - letter +> (letter >|< digit >|< char '\'')`star` epsilon + letter +> (letter >|< digit >|< char '\'' >|< char '.')`star` epsilon `lexactionName` \cs pos name -> (idkwtok $!pos) cs name where diff -U2 -r1.4 CHS.hs --- CHS.hs 20 May 2004 16:42:17 -0000 1.4 +++ CHS.hs 16 Jul 2004 18:52:26 -0000 @@ -747,9 +747,14 @@ CHSTokQualif _: CHSTokIdent _ ide:toks -> return (True , ide, toks) _ -> syntaxError toks - chi <- loadCHI . identToLexeme $ modid + chi <- loadCHI . moduleNameToFileName . identToLexeme $ modid toks'' <- parseEndHook toks' frags <- parseFrags toks'' return $ CHSHook (CHSImport qual modid chi pos) : frags +moduleNameToFileName :: String -> FilePath +moduleNameToFileName = map dotToSlash + where dotToSlash '.' = '/' + dotToSlash c = c + parseContext :: Position -> [CHSToken] -> CST s [CHSFrag] parseContext pos toks = do Does this seem ok? It makes me a bit nervous changing the lexical definition of an identifier (we really only want to change the module identifier definition, not all identifiers). Duncan
On Fri, 2004-07-16 at 19:57, Duncan Coutts wrote:
Does this seem ok? It makes me a bit nervous changing the lexical definition of an identifier (we really only want to change the module identifier definition, not all identifiers).
Ok, it does break things, for example this would get lexed as a single identifier: {#get GList.data#} So either the lexer needs to be cleverer and lex module identifiers differently from ordinary identifiers or the parser will need to reconstruct the module identifier from a sequence of identifier and dot tokens (preferably the former since the latter would be modulo whitespace which is not strictly correct). Duncan
On Sat, 2004-07-17 at 02:01, Duncan Coutts wrote:
Ok, it does break things, for example this would get lexed as a single identifier:
{#get GList.data#}
So either the lexer needs to be cleverer and lex module identifiers differently from ordinary identifiers or the parser will need to reconstruct the module identifier from a sequence of identifier and dot tokens (preferably the former since the latter would be modulo whitespace which is not strictly correct).
Here's a patch that implements the latter solution. It's not lovely because it has to reconstruct an identifier from several identifiers and Ident is a bit too abstract for this to be easy. I tried altering the lexer but it's not clear to me how to recognise a sequence of things easily. You can do sequences of things in a regexp, but then its hard to return multiple tokens. Lexers can only be combined by disjunction as far as I can see. Unlike the last patch this one seems to work, gtk2hs builds ok with it. Anyway, here's the patch (this time changing only the parser, not the lexer): diff -C2 -r1.4 CHS.hs *** c2hs/chs/CHS.hs 20 May 2004 16:42:17 -0000 1.4 --- c2hs/chs/CHS.hs 23 Jul 2004 15:40:21 -0000 *************** *** 744,755 **** (qual, modid, toks') <- case toks of ! CHSTokIdent _ ide :toks -> return (False, ide, toks) ! CHSTokQualif _: CHSTokIdent _ ide:toks -> return (True , ide, toks) _ -> syntaxError toks ! chi <- loadCHI . identToLexeme $ modid toks'' <- parseEndHook toks' frags <- parseFrags toks'' return $ CHSHook (CHSImport qual modid chi pos) : frags parseContext :: Position -> [CHSToken] -> CST s [CHSFrag] parseContext pos toks = do --- 744,773 ---- (qual, modid, toks') <- case toks of ! CHSTokIdent _ ide :toks -> ! let (ide', toks') = rebuildModuleId ide toks ! in return (False, ide', toks') ! CHSTokQualif _: CHSTokIdent _ ide:toks -> ! let (ide', toks') = rebuildModuleId ide toks ! in return (True , ide', toks') _ -> syntaxError toks ! chi <- loadCHI . moduleNameToFileName . identToLexeme $ modid toks'' <- parseEndHook toks' frags <- parseFrags toks'' return $ CHSHook (CHSImport qual modid chi pos) : frags + -- Qualified module names do not get lexed as a single token so we need to + -- reconstruct it from a sequence of identifer and dot tokens. + -- + rebuildModuleId ide (CHSTokDot _ : CHSTokIdent _ ide' : toks) = + let catIdent ide ide' = onlyPosIdent (posOf ide) --FIXME: unpleasent hack + (identToLexeme ide ++ '.' : identToLexeme ide') + in rebuildModuleId (catIdent ide ide') toks + rebuildModuleId ide toks = (ide, toks) + + moduleNameToFileName :: String -> FilePath + moduleNameToFileName = map dotToSlash + where dotToSlash '.' = '/' + dotToSlash c = c + parseContext :: Position -> [CHSToken] -> CST s [CHSFrag] parseContext pos toks = do Duncan
participants (1)
-
Duncan Coutts