A somewhat better solution to the words/UTF-8 problem: do encoding last, but only on what appears to be string literals.

src/GF/Compile.hs

@@ -208,7 +208,7 @@ compileSourceModule opts env@(k,gr,_) mo@(i,mi) = do
 generateModuleCode :: Options -> FilePath -> SourceModule -> IOE SourceModule
 generateModuleCode opts file minfo = do
   let minfo1 = subexpModule minfo
-      out    = prGrammar (MGrammar [codeSourceModule encodeUTF8 minfo1])
+      out    = codeStringLiterals encodeUTF8 $ prGrammar (MGrammar [minfo1])
   putPointE Normal opts ("  wrote file" +++ file) $ ioeIO $ writeFile file $ out
   return minfo1
 

src/GF/Compile/Coding.hs

@@ -48,3 +48,13 @@ codeSourceModule co (id,moi) = case moi of
       PSeq p q -> PSeq (codp p) (codp q)
       PAlt p q -> PAlt (codp p) (codp q)
       _ -> p
+
+-- | Run an encoding function on all string literals within the given string.
+codeStringLiterals :: (String -> String) -> String -> String
+codeStringLiterals _ [] = []
+codeStringLiterals co ('"':cs) = '"' : inStringLiteral cs
+  where inStringLiteral [] = error "codeStringLiterals: unterminated string literal"
+        inStringLiteral ('"':ds) = '"' : codeStringLiterals co ds
+        inStringLiteral ('\\':d:ds) = '\\' : co [d] ++ inStringLiteral ds
+        inStringLiteral (d:ds) = co [d] ++ inStringLiteral ds
+codeStringLiterals co (c:cs) = c : codeStringLiterals co cs

src/GF/Infra/CompactPrint.hs

@@ -5,8 +5,7 @@ compactPrint = compactPrintCustom keywordGF (const False)
 
 compactPrintGFCC = compactPrintCustom (const False) keywordGFCC
 
--- FIXME: using words is not safe, since this is run on UTF-8 encoded data.
+compactPrintCustom pre post = dps . concat . map (spaceIf pre post) . words 
-compactPrintCustom pre post = id -- dps . concat . map (spaceIf pre post) . words 
 
 dps = dropWhile isSpace