Fix #335: error out on '<$>' in semantic action code

.gitignore

@@ -8,3 +8,5 @@ cabal.sandbox.config
 .*.swo
 /.vscode/
 tests/*.hs
+*~
+TAGS

ChangeLog.md

@@ -1,5 +1,13 @@
 # Revision history for Happy
 
+## 2.2
+
+* Error out when `<$>` appears in semantic action code
+  ([#335](https://github.com/haskell/happy/issues/335)).
+
+* Tested with GHC 8.0 - 9.14.1.
+  The Haskell code generated by Happy is for GHC 8.0 and up.
+
 ## 2.1.7
 
 * Add support for `{-# OPTIONS_HAPPY ... #-}` pragmas

doc/syntax.rst

@@ -445,6 +445,9 @@ It is an error to refer to ``$i`` when ``i`` is larger than the number of symbol
 The symbol ``$`` may be inserted literally in the Haskell expression using the sequence ``\$`` (this isn't necessary inside a string or character literal).
 
 Additionally, the sequence ``$>`` can be used to represent the value of the rightmost symbol.
+Since version 2.2, Happy throws an error when ``<$>`` appears in the Haskell expression;
+Happy interprets this as ``< $>`` but users often mean the operator version of ``fmap``.
+(Observe that the latter is available as ``<\$>``.)
 
 A semantic value of the form ``{% ... }`` is a *monadic action*, and is only valid when the grammar file contains a ``%monad`` directive (:ref:`Monad Directive <sec-monad-decl>`).
 Monadic actions are discussed in :ref:`Monadic Parsers <sec-monads>`.

happy.cabal

@@ -1,6 +1,6 @@
 cabal-version: 1.18
 name: happy
-version: 2.1.7
+version: 2.2
 license: BSD2
 license-file: LICENSE
 copyright: (c) Andy Gill, Simon Marlow
@@ -114,6 +114,9 @@ extra-source-files:
         tests/error001.stderr
         tests/error001.stdout
         tests/error001.y
+        tests/issue335.stderr
+        tests/issue335.stdout
+        tests/issue335.y
         tests/monad001.y
         tests/monad002.ly
         tests/monaderror.y
@@ -141,7 +144,7 @@ executable happy
   main-is: Main.lhs
 
   build-depends: base >= 4.9 && < 5,
-                 happy-lib == 2.1.7
+                 happy-lib == 2.2
 
   default-language: Haskell98
   default-extensions: CPP, MagicHash, FlexibleContexts, NamedFieldPuns

lib/frontend/src/Happy/Frontend/Mangler.lhs

@@ -87,7 +87,7 @@ go do special processing.  If not, pass on to the regular processing routine
 >                       ns -> ns
 >       error_resumptive | ResumptiveErrorHandler{} <- getError dirs = True
 >                        | otherwise                                 = False
->
+
 >       start_strs  = [ startName++'_':p  | (TokenName p _ _) <- starts' ]
 
 Build up a mapping from name values to strings.
@@ -132,7 +132,7 @@ Start symbols...
 Deal with priorities...
 
 >       priodir = zip [1..] (getPrios dirs)
->
+
 >       mkPrio :: Int -> Directive a -> Priority
 >       mkPrio i (TokenNonassoc _) = Prio None i
 >       mkPrio i (TokenRight _) = Prio RightAssoc i
@@ -155,10 +155,10 @@ Translate the rules from string to name-based.
 >       convNT (Rule1 nt prods ty)
 >         = do nt' <- mapToName nt
 >              return (nt', prods, ty)
->
+
 >       transRule (nt, prods, _ty)
 >         = mapM (finishRule nt) prods
->
+
 >       finishRule :: Name -> Prod1 e -> Writer [ErrMsg] (Production e)
 >       finishRule nt (Prod1 lhs code line prec)
 >         = mapWriter (\(a,e) -> (a, map (addLine line) e)) $ do
@@ -168,7 +168,7 @@ Translate the rules from string to name-based.
 >               Left s  -> do addErr ("Undeclared precedence token: " ++ s)
 >                             return (Production nt lhs' code' No)
 >               Right p -> return (Production nt lhs' code' p)
->
+
 >       mkPrec :: [Name] -> Prec -> Either String Priority
 >       mkPrec lhs PrecNone =
 >         case filter (flip elem terminal_names) lhs of
@@ -180,9 +180,9 @@ Translate the rules from string to name-based.
 >         case lookup s prioByString of
 >                           Nothing -> Left s
 >                           Just p -> Right p
->
+
 >       mkPrec _ PrecShift = Right PrioLowest
->
+
 >   -- in
 
 >   rules1 <- mapM convNT rules
@@ -191,7 +191,7 @@ Translate the rules from string to name-based.
 >   let
 >       type_env = [(nt, t) | Rule1 nt _ (Just (t,[])) <- rules] ++
 >                  [(nt, getTokenType dirs) | nt <- terminal_strs] -- XXX: Doesn't handle $$ type!
->
+
 >       fixType (ty,s) = go "" ty
 >         where go acc [] = return (reverse acc)
 >               go acc (c:r) | isLower c = -- look for a run of alphanumerics starting with a lower case letter
@@ -205,14 +205,14 @@ Translate the rules from string to name-based.
 >                                            go1 (c:cs)
 >                                          Just t -> go1 $ "(" ++ t ++ ")"
 >                            | otherwise = go (c:acc) r
->
+
 >       convType (nm, t)
 >         = do t' <- fixType t
 >              return (nm, t')
->
+
 >   -- in
 >   tys <- mapM convType [ (nm, t) | (nm, _, Just t) <- rules1 ]
->
+
 
 >   let
 >       type_array :: Array Name (Maybe String)
@@ -238,7 +238,7 @@ Get the token specs in terms of Names.
 >      lookup_prods :: Name -> [Int]
 >      lookup_prods x | x >= firstStartTok && x < first_t = arr ! x
 >      lookup_prods _ = error "lookup_prods"
->
+
 >      productions' = start_prods ++ concat rules2
 >      prod_array  = listArray (0,length productions' - 1) productions'
 
@@ -278,7 +278,7 @@ Gofer-like stuff:
 >       combine [] = []
 >       combine ((a,b):(c,d):r) | a == c = combine ((a,b++d) : r)
 >       combine (a:r) = a : combine r
->
+
 
 For combining actions with possible error messages.
 
@@ -307,12 +307,15 @@ So is this.
 -- At the same time, we collect a list of the variables actually used in this
 -- code, which is used by the backend.
 
-> doCheckCode :: Int -> String -> M (String, [Int])
+> doCheckCode ::
+>      Int                -- ^ Arity of the rule, i.e., maximal number of a variable.
+>   -> String             -- ^ Haskell code for the semantic action of the rule.
+>   -> M (String, [Int])  -- ^ The translated code and the variable occurrences (in reverse order).
 > doCheckCode arity code0 = go code0 "" []
 >   where go code acc used =
 >           case code of
 >               [] -> return (reverse acc, used)
->
+
 >               '"'  :r    -> case reads code :: [(String,String)] of
 >                                []       -> go r ('"':acc) used
 >                                (s,r'):_ -> go r' (reverse (show s) ++ acc) used
@@ -321,13 +324,24 @@ So is this.
 >                                []       -> go r  ('\'':acc) used
 >                                (c,r'):_ -> go r' (reverse (show c) ++ acc) used
 >               '\\':'$':r -> go r ('$':acc) used
->
+
+
+Issue #335 (Andreas Abel, 2026-01-14):
+Users often write '<$>' to mean '`fmap`', but Happy interprets it as '< $>' where '$>'
+denotes the maximal variable (rightmost item).
+Thus, reject '<$>' in semantic action code and suggest to either escape the '$'
+or put a space after '<'.
+
+>               c@'<':r@('$':'>':_) -> do
+>                 addErr "Found '<$>' in semantic action code; either write '<\\$>' (for what usually is fmap) or '< $>' (for \"less than rightmost item\")"
+>                 go r (c:acc) used
+
 >               '$':'>':r -- the "rightmost token"
 >                       | arity == 0 -> do addErr "$> in empty rule"
 >                                          go r acc used
 >                       | otherwise  -> go r (reverse (mkHappyVar arity) ++ acc)
 >                                        (arity : used)
->
+
 >               '$':r@(i:_) | isDigit i ->
 >                       case reads r :: [(Int,String)] of
 >                         (j,r'):_ ->

lib/happy-lib.cabal

@@ -1,6 +1,6 @@
 cabal-version: 3.0
 name: happy-lib
-version: 2.1.7
+version: 2.2
 license: BSD-2-Clause
 copyright: (c) Andy Gill, Simon Marlow
 author: Andy Gill and Simon Marlow

tests/.gitignore

@@ -5,3 +5,5 @@
 *.n.hs
 *.stderr
 *.stdout
+!issue335.stderr
+!issue335.stdout

tests/Makefile

@@ -41,7 +41,7 @@ TESTS = Test.ly TestMulti.ly TestPrecedence.ly bug001.ly \
 	AttrGrammar001.y AttrGrammar002.y \
 	Pragma.y
 
-ERROR_TESTS = error001.y
+ERROR_TESTS = error001.y issue335.y
 
 # NOTE: `cabal` will set the `happy_datadir` env-var accordingly before invoking the test-suite
 #TEST_HAPPY_OPTS = --strict --template=..

tests/issue335.stderr

@@ -0,0 +1,2 @@
+issue335.y: 7: Found '<$>' in semantic action code; either write '<\$>' (for what usually is fmap) or '< $>' (for "less than rightmost item")
+

tests/issue335.y

@@ -0,0 +1,6 @@
+%name foo
+%tokentype { String }
+
+%%
+
+foo : foo { void <$> putStr "The '<$>' shall mean 'fmap' but happy will read it as '< $>' which is not what we want." }