Please test if you're able to, and give us feedback.
It looks like GADTs (or something else new) conflict with normal Haskell'98 type inference. The following example used to compile just fine with all previous versions of ghc and nhc98. $ ghc-6.4.20050210 -package lang -c -o Parse.o Parse.hs Parse.hs:209:4: Couldn't match the rigid variable `a' against the rigid variable `a1' `a' is bound by the type signature for `parseValdef' `a1' is bound by the type signature for `parseWhere' Expected type: Parser (Decls TokenId) [PosToken] a1 Inferred type: Parser (Decls TokenId) [(Pos, Lex, Lexical.LexState, [LexPre.PosTokenPre])] a In the expression: ((((lit L_where) `revChk` lcurl) `revChk` parseDecls) `chk` rcurl) `orelse` (parse (DeclsParse [])) In the definition of `parseWhere': parseWhere = ((((lit L_where) `revChk` lcurl) `revChk` parseDecls) `chk` rcurl) `orelse` (parse (DeclsParse [])) The quoted expressions look a little bit hairy, but if you examine the explicit type signatures in question, it is very clear that there should be no error here. To reproduce the bug, just build the nhc98 compiler proper. $ wget ftp://ftp.cs.york.ac.uk/pub/haskell/nhc98/nhc98src-1.17.tar.gz $ tar zxf nhc98src-1.17.tar.gz $ ./configure --buildwith=ghc-6.4.20050210 $ make basic Regards, Malcolm