Don't use tilde code blocks with braced attributes in gfm output.

A consequence of this change is that the backtick form will be
preferred in general if both are enabled.  I think that is good,
as it is much more widespread than the tilde form.

Closes #1084.

src/Text/Pandoc/Writers/Markdown.hs

@@ -381,13 +381,11 @@ blockToMarkdown opts (CodeBlock (_,classes,_) str)
     isEnabled Ext_literate_haskell opts =
   return $ prefixed "> " (text str) <> blankline
 blockToMarkdown opts (CodeBlock attribs str) = return $
-  case attribs of
-     x | x /= nullAttr && isEnabled Ext_fenced_code_blocks opts ->
-          tildes <> " " <> attrs <> cr <> text str <>
-           cr <> tildes <> blankline
-     (_,(cls:_),_) | isEnabled Ext_backtick_code_blocks opts ->
-          backticks <> " " <> text cls <> cr <> text str <>
-           cr <> backticks <> blankline
+  case attribs == nullAttr of
+     False | isEnabled Ext_backtick_code_blocks opts ->
+          backticks <> attrs <> cr <> text str <> cr <> backticks <> blankline
+           | isEnabled Ext_fenced_code_blocks opts ->
+          tildes <> attrs <> cr <> text str <> cr <> tildes <> blankline
      _ -> nest (writerTabStop opts) (text str) <> blankline
    where tildes    = text $ case [ln | ln <- lines str, all (=='~') ln] of
                                [] -> "~~~~"
@@ -396,8 +394,10 @@ blockToMarkdown opts (CodeBlock attribs str) = return $
                                             | otherwise -> replicate (n+1) '~'
          backticks = text "```"
          attrs  = if isEnabled Ext_fenced_code_attributes opts
-                     then nowrap $ attrsToMarkdown attribs
-                     else empty
+                     then nowrap $ " " <> attrsToMarkdown attribs
+                     else case attribs of
+                                (_,[cls],_) -> " " <> text cls
+                                _           -> empty
 blockToMarkdown opts (BlockQuote blocks) = do
   st <- get
   -- if we're writing literate haskell, put a space before the bird tracks

tests/lhs-test.markdown

@@ -4,11 +4,11 @@ lhs test
 `unsplit` is an arrow that takes a pair of values and combines them to return
 a single value:
 
-~~~~ {.sourceCode .literate .haskell}
+``` {.sourceCode .literate .haskell}
 unsplit :: (Arrow a) => (b -> c -> d) -> a (b, c) d
 unsplit = arr . uncurry
           -- arr (\op (x,y) -> x `op` y)
-~~~~
+```
 
 `(***)` combines two arrows into a new arrow by running the two arrows on a
 pair of values (one arrow on the first item of the pair and one arrow on the