path: root/Parser.hs

{-# LANGUAGE LambdaCase #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE TupleSections #-}
{-# LANGUAGE DataKinds #-}
{-# LANGUAGE GADTs #-}
{-# LANGUAGE StandaloneDeriving #-}
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE RankNTypes #-}
{-# LANGUAGE TypeOperators #-}
-- I don't want a warning for 'head' and 'tail' in this file. But I also don't
-- want GHCs before 9.8 to complain that they don't know the x-partial warning.
{-# OPTIONS_GHC -Wno-unrecognised-warning-flags -Wno-x-partial #-}
module Parser (
  parse,
  printErrMsg,
  -- * Re-exports
  These(..),
) where

import Control.Applicative
import Control.Monad
import Control.Monad.Chronicle
import Control.Monad.Reader
import Control.Monad.State.Lazy
import Data.Bifunctor (first)
import Data.Char
import Data.List.NonEmpty (NonEmpty(..))
import Data.These
import Data.Tuple (swap)

import Debug.Trace

import AST


-- Positions are zero-based in both dimensions.
-- See 'isInsideBlock' and 'isAtBlockLeft' for the two relevant "inside the
-- block" conditions.
data PS = PS
  { psBlkLine :: Int  -- ^ Start line of current layout block
  , psBlkCol :: Int   -- ^ Start column of current layout block
  , psLine :: Int  -- ^ Current line
  , psCol :: Int   -- ^ Current column
  , psRest :: String  -- ^ Rest of the input
  }
  deriving (Show)

data Context = Context
  { ctxFile :: FilePath
  , ctxStack :: [String]  -- ^ Stack of syntax scopes, for error reporting
  }
  deriving (Show)

-- ReaderT Context (ChronicleT [ErrMsg] (State PS) a)
-- Context -> ChronicleT [ErrMsg] (State PS) a
-- Context -> State PS (These [ErrMsg] a)
-- Context -> PS -> Identity (These [ErrMsg] a, PS)
-- Context -> PS -> (These [ErrMsg] a, PS)
-- whereas I want:
-- Context -> PS -> These [ErrMsg] (a, PS)
-- which is not any transformer stack, but a new monad.
newtype Parser a = Parser { runParser :: Context -> PS -> These [ErrMsg] (PS, a) }

instance Functor Parser where
  fmap f (Parser g) = Parser (\ctx ps -> fmap (fmap f) (g ctx ps))

instance Applicative Parser where
  pure x = Parser (\_ ps -> That (ps, x))
  (<*>) = ap

instance Monad Parser where
  Parser g >>= f = Parser $ \ctx ps ->
    case g ctx ps of
      This errs -> This errs
      That (ps', x) -> runParser (f x) ctx ps'
      These errs (ps', x) -> case runParser (f x) ctx ps' of
                               This errs' -> This (errs <> errs')
                               That res -> These errs res
                               These errs' res -> These (errs <> errs') res

instance Alternative Parser where
  empty = Parser (\_ _ -> This mempty)
  Parser f <|> Parser g = Parser $ \ctx ps ->
    case f ctx ps of
      This _ -> g ctx ps
      success -> success

instance MonadState PS Parser where
  state f = Parser $ \_ ps -> That (swap (f ps))

instance MonadReader Context Parser where
  reader f = Parser $ \ctx ps -> That (ps, f ctx)
  local f (Parser g) = Parser (g . f)

instance MonadChronicle [ErrMsg] Parser where
  dictate errs = Parser (\_ ps -> These errs (ps, ()))
  confess errs = Parser (\_ _ -> This errs)
  memento (Parser f) = Parser (\ctx ps -> case f ctx ps of
                                            This errs -> That (ps, Left errs)
                                            That res -> That (Right <$> res)
                                            These errs res -> These errs (Right <$> res))
  absolve def (Parser f) = Parser (\ctx ps -> case f ctx ps of
                                                This _ -> That (ps, def)
                                                success -> success)
  condemn (Parser f) = Parser (\ctx ps -> case f ctx ps of
                                            These errs _ -> This errs
                                            res -> res)
  retcon g (Parser f) = Parser (\ctx ps -> first g (f ctx ps))
  chronicle th = Parser (\_ ps -> (ps,) <$> th)

-- Positions are zero-based in both dimensions
data ErrMsg = ErrMsg
  { errFile :: FilePath
  , errStk :: [String]
  , errLine :: Int
  , errCol :: Int
  , errMsg :: String }
  deriving (Show)

printErrMsg :: ErrMsg -> String
printErrMsg (ErrMsg fp stk y x s) =
  unlines (map (\descr -> "In " ++ descr ++ ":") (reverse stk)) ++
  fp ++ ":" ++ show (y + 1) ++ ":" ++ show (x + 1) ++ ": " ++ s

parse :: FilePath -> String -> These [ErrMsg] (Program ())
parse fp source = fmap snd $ runParser pProgram (Context fp []) (PS 0 0 0 0 source)

pProgram :: Parser (Program ())
pProgram = do
  prog <- Program <$> many pFunDef
  skipWhiteComment
  assertEOF Error
  return prog

pFunDef :: Parser (FunDef ())
pFunDef = do
  skipWhiteComment
  pFunDef0

pFunDef0 :: Parser (FunDef ())
pFunDef0 = do
  mtypesig <- optional pStandaloneTypesig0
  let mname = fst <$> mtypesig
      mtype = snd <$> mtypesig
  (clauses, name) <- someClauses mname
  return (FunDef name mtype clauses)
  where
    someClauses :: Maybe Name -> Parser (NonEmpty (FunEq ()), Name)
    someClauses Nothing = do
        clause@(FunEq name _ _) <- pFunEq Nothing
        (,name) . (clause :|) <$> many (pFunEq (Just name))
    someClauses (Just name) = (,name) <$> someNE (pFunEq (Just name))

-- | Given the name of the type signature, if any.
pFunEq :: Maybe Name -> Parser (FunEq ())
pFunEq mCheckName = do
  skipWhiteComment
  assertAtBlockLeft Fatal "Expected function clause, found indented stuff"

  name <- pIdentifier0 AtLeft Lowercase
  case mCheckName of
    Just checkName | name /= checkName ->
        raise Fatal "Name of function clause does not correspond with type signature"
    _ -> return ()

  pats <- many (pPattern 11)
  rhs <- pRHS "="
  return (FunEq name pats rhs)

-- | Pass "=" for function definitions and "->" for case clauses.
pRHS :: String -> Parser (RHS ())
pRHS sepsym = do
  -- TODO: parse guards
  inlineWhite
  pKeySym sepsym
  Plain <$> pExpr

pPattern :: Int -> Parser (Pattern ())
pPattern d = inlineWhite >> pPattern0 d

pPattern0 :: Int -> Parser (Pattern ())
pPattern0 d = do
  asum [pPatWildcard0
       ,pPatVarOrAs0
       ,pPatCon0
       ,pPatList0
       ,pPatParens0]
  where
    pPatWildcard0 = pKeySym "_" >> return (PWildcard ())
    pPatVarOrAs0 = do
      var <- pIdentifier0 InBlock Lowercase
      asum [do inlineWhite
               pKeySym "@"
               p <- pPattern 11
               return (PAs () var p)
           ,return (PVar () var)]
    pPatCon0 = do
      con <- pIdentifier0 InBlock Uppercase
      if d > 0
        then return (PCon () con [])
        else do args <- many (pPattern 11)
                return (PCon () con args)
    pPatList0 = do
      char '['  -- special syntax, no need for pKeySym
      ps <- pPattern 0 `sepBy` (inlineWhite >> char ',')
      inlineWhite
      char ']'
      return (PList () ps)
    pPatParens0 = do
      char '('
      inlineWhite
      asum [do char ')'
               return (PTup () [])
           ,do p <- pPattern0 0
               inlineWhite
               asum [do char ')'
                        return p
                    ,do char ','
                        ps <- pPattern 0 `sepBy1` (inlineWhite >> char ',')
                        return (PTup () (p : ps))]]

pExpr :: Parser (Expr ())
pExpr = do
  inlineWhite
  -- basics: lit, list, var, con, tup
  -- expression atom: application of basics
  -- expression parser: op
  -- around: let, case, if
  asum [pELet0
       ,pECase0
       ,pEIf0
       ,pExprOpExpr0 0]

pELet0 :: Parser (Expr ())
pELet0 = do
  pKeyword "let"
  inlineWhite
  startLayoutBlock $ do
    -- The first occurrence is also going to skip whitespace in front,
    -- which is redundant -- but not harmful.
    defs <- many $ do
      skipWhiteComment
      -- Note: now not necessarily in the indented block. Which is
      -- precisely what we need here. If we see "in", let the 'many'
      -- choice fail so that the defs loop ends. But let it fail outside
      -- this asum so that it is the many that picks it up, not this
      -- asum.
      res <- asum [Nothing <$ lookAhead (pKeyword "in")
                  ,Just <$> pFunDef0]
      case res of
        Nothing -> empty
        Just def -> return def
    inlineWhite
    body <- pExpr
    return (ELet () defs body)

pECase0 :: Parser (Expr ())
pECase0 = do
  pKeyword "case"
  e <- pExpr
  inlineWhite
  pKeyword "of"
  inlineWhite
  startLayoutBlock $ do
    -- The first clause is going to skip whitespace, but that's harmless
    -- (though redundant).
    let pClause = do
          skipWhiteComment
          whenM (not <$> isInsideBlock) (() <$ empty)
          pat <- pPattern0 0
          rhs <- pRHS "->"
          return (pat, rhs)
    clauses <- many pClause
    return (ECase () e clauses)

pEIf0 :: Parser (Expr ())
pEIf0 = do
  pKeyword "if"
  e1 <- pExpr
  inlineWhite
  pKeyword "then"
  e2 <- pExpr
  inlineWhite
  pKeyword "else"
  e3 <- pExpr
  return (EIf () e1 e2 e3)

pExprOpExpr :: Int -> Parser (Expr ())
pExprOpExpr d = inlineWhite >> pExprOpExpr0 d

pExprOpExpr0 :: Int -> Parser (Expr ())
pExprOpExpr0 d = do
  e0 <- pEApp0
  climbRight e0 Nothing
  where
    climbRight :: Expr () -> Maybe ParsedOperator -> Parser (Expr ())
    climbRight lhs mlhsop = do
      asum [do paop@(PaOp op d2 a2) <- pInfixOp
               guard (d2 >= d)  -- respect global minimum precedence
               case mlhsop of  -- check operator compatibility
                 Just (PaOp _ d1 a1) ->
                     guard (d1 > d2 || (d1 == d2 && a1 == a2 && a1 /= AssocNone))
                 Nothing ->
                     return ()
               let oprhsd = case a2 of AssocRight -> d2 ; _ -> d2 + 1
               rhs <- pExprOpExpr oprhsd
               climbRight (EOp () lhs op rhs) (Just paop)
           ,return lhs]

pEApp0 :: Parser (Expr ())
pEApp0 = do
  e1 <- pEAtom0
  es <- many (inlineWhite >> pEAtom0)
  case es of
    [] -> return e1
    _ -> return (EApp () e1 es)

pEAtom0 :: Parser (Expr ())
pEAtom0 = (ELit () <$> pLiteral0) <|> pEList0 <|> pEVar0 <|> pECon0 <|> pEParens0

pLiteral0 :: Parser Literal
pLiteral0 = asum
  [do as <- some (satisfy isDigit)
      let a = read as :: Integer
      asum
        [do char '.'
            bs <- some (satisfy isDigit)
            let b = read bs :: Integer
            cs <- optional $ do
                    char 'e'
                    cs <- some (satisfy isDigit)
                    return cs
            let c = maybe 0 read cs :: Integer
            return (LFloat ((fromIntegral a + fromIntegral b / 10 ^ length bs) * 10 ^ c))
        ,return (LInt a)]
  ,do char '\''
      c <- pStringChar
      char '\''
      return (LChar c)
  ,do char '"'
      s <- many pStringChar
      char '"'
      return (LString s)]

pStringChar :: Parser Char
pStringChar = asum
  [do char '\\'
      char 'x'
      let hexdig = do
            c <- satisfy $ \c' ->
                    let c = toLower c'
                    in 'a' <= c && c <= 'f' || '0' <= c && c <= '9'
            return $ if 'a' <= c then 10 + ord c - ord 'a'
                                 else ord c - ord '0'
      digs <- some hexdig
      return (chr (sum (zipWith (*) (reverse digs) (iterate (*16) 1))))
  ,do char '\\'
      satisfy (const True) >>= \case
        'n' -> return '\n'
        'r' -> return '\r'
        't' -> return '\t'
        'a' -> return '\a'
        'b' -> return '\b'
        '\'' -> return '\''
        '\"' -> return '\"'
        '0' -> return '\0'
        c -> do raise Error $ "Invalid escape sequence: \\" ++ [c]
                return '?'
  ,do satisfy (\c -> c `notElem` "\n\r\\\'")]

pEList0 :: Parser (Expr ())
pEList0 = do
  char '['  -- special syntax, no need for pKeySym
  es <- sepBy pExpr (inlineWhite >> char ',')
  inlineWhite
  char ']'
  return (EList () es)

pEVar0 :: Parser (Expr ())
pEVar0 = EVar () <$> pIdentifier0 InBlock Lowercase

pECon0 :: Parser (Expr ())
pECon0 = ECon () <$> pIdentifier0 InBlock Uppercase

pEParens0 :: Parser (Expr ())
pEParens0 = do
  char '('
  e <- pExpr
  inlineWhite
  char ')'
  return e

data Associativity = AssocLeft | AssocRight | AssocNone
  deriving (Show, Eq)

data ParsedOperator = PaOp Operator Int Associativity
  deriving (Show)

pInfixOp :: Parser ParsedOperator
pInfixOp = do
  inlineWhite
  asum [PaOp OEqu 4 AssocNone  <$ pKeySym "=="
       ,PaOp OAdd 6 AssocLeft  <$ pKeySym "+"
       ,PaOp OSub 6 AssocLeft  <$ pKeySym "-"
       ,PaOp OMul 7 AssocLeft  <$ pKeySym "*"
       ,PaOp ODiv 7 AssocLeft  <$ pKeySym "/"
       ,PaOp OMod 7 AssocLeft  <$ pKeySym "%"
       ,PaOp OPow 8 AssocRight <$ pKeySym "^"
       ]

pStandaloneTypesig0 :: Parser (Name, Type)
pStandaloneTypesig0 = do
  assertAtBlockLeft Fatal "Expected top-level definition, found indented stuff"
  name@(Name namestr) <- pIdentifier0 AtLeft Lowercase
  inlineWhite
  pKeySym "::"
  pushContext ("type signature for '" ++ namestr ++ "'") $ do
      ty <- pType
      return (name, ty)

pType :: Parser Type
pType = do
  ty1 <- pTypeApp
  asum [do inlineWhite
           pKeySym "->"
           ty2 <- pType
           return (TFun ty1 ty2)
       ,return ty1]

pTypeApp :: Parser Type
pTypeApp = many pTypeAtom >>= \case
  [] -> raise Error "Expected type" >> return (TTup [])
  [t] -> return t
  t:ts -> return (TApp t ts)

pTypeAtom :: Parser Type
pTypeAtom = pTypeParens <|> pTypeList <|> pTypeCon <|> pTypeVar
  where
    pTypeParens = do
      inlineWhite
      char '('
      asum [do inlineWhite
               char ')'
               return (TTup [])
           ,do ty1 <- pType
               ty2s <- many $ do
                 inlineWhite
                 char ','
                 pType
               inlineWhite
               char ')'
               case ty2s of
                 [] -> return ty1
                 _ -> return (TTup (ty1 : ty2s))]

    pTypeList = do
      inlineWhite
      char '['
      ty <- pType
      char ']'
      return (TList ty)

    pTypeCon = inlineWhite >> TCon <$> pIdentifier0 InBlock Uppercase
    pTypeVar = inlineWhite >> TVar <$> pIdentifier0 InBlock Lowercase

-- | Parse the given name-like keyword, ensuring that it is the entire word.
pKeyword :: String -> Parser ()
pKeyword s = do
  string s
  notFollowedBy (() <$ satisfy isNameContChar)

-- | Parse the given symbol-like keyword, ensuring that it is the entire symbol.
pKeySym :: String -> Parser ()
pKeySym s = do
  string s
  notFollowedBy (() <$ satisfy isSymbolChar)

data Case = Uppercase | Lowercase
  deriving (Show)

-- | Consumes an identifier (word or parenthesised operator) at the current
-- position. The `var` production in Haskell2010.
-- var -> varid | "(" varsym ")"
pIdentifier0 :: BlockPos -> Case -> Parser Name
pIdentifier0 bpos cs = pAlphaName0 bpos cs <|> pParens0 (pSymbol0 bpos cs)
  where
    -- | Parser between parens, with the opening paren at the current position.
    pParens0 :: Parser a -> Parser a
    pParens0 p = do
      char '('
      inlineWhite
      res <- p
      inlineWhite
      char ')'
      return res

-- | Consumes a word-like name at the current position with the given case. The
-- `varid` production in Haskell2010 for 'Lowercase', `conid' for 'Uppercase'.
--
-- > varid -> (small {small | large | digit | "'"}) without reservedid
pAlphaName0 :: BlockPos -> Case -> Parser Name
pAlphaName0 bpos cs = do
  (_, s) <- readToken
      bpos
      (\atfst mc -> case (atfst, mc) of
          (True , Just c) | isNameHeadChar c -> Just (Right False)
          (True , _     )                    -> Nothing
          (False, Just c) | isNameContChar c -> Just (Right False)
          (False, _     )                    -> Just (Left ()))
      True
  name <- case cs of
    Uppercase | isLower (head s) -> do
        raise Error "Unexpected uppercase word at this position, assuming typo"
        return (toUpper (head s) : tail s)
    Lowercase | isUpper (head s) -> do
        raise Error "Unexpected lowercase word at this position, assuming typo"
        return (toLower (head s) : tail s)
    _ -> return s
  guard (name `notElem` ["case", "class", "data", "default", "deriving", "do", "else"
                        ,"foreign", "if", "import", "in", "infix", "infixl"
                        ,"infixr", "instance", "let", "module", "newtype", "of"
                        ,"then", "type", "where", "_"])
  return (Name name)

isNameHeadChar :: Char -> Bool
isNameHeadChar c = isLetter c || c == '_'

isNameContChar :: Char -> Bool
isNameContChar c = isNameHeadChar c || isDigit c || c == '\''

-- | Consumes a symbol at the current position. The `varsym` production in
-- Haskell2010 for 'Lowercase', `consym` otherwise.
--
-- > varsym -> ((symbol without ":") {symbol}) without (reservedop | dashes)
-- > consym -> (":" {symbol}) without reservedop
-- > symbol -> ascSymbol | uniSymbol without (special | "_" | "\"" | "'")
-- > ascSymbol -> ```!#$%&⋆+./<=>?@^|-~:```
-- > uniSymbol -> unicode symbol or punctuation
-- > dashes -> "--" {"-"}
-- > special -> ```(),;[]`{}```
-- > reservedop -> ".." | ":" | "::" | "=" | "\" | "|" | "<-" | "->" | "@" | "~" | "=>"
pSymbol0 :: BlockPos -> Case -> Parser Name
pSymbol0 bpos cs = do
  case bpos of
    AtLeft -> assertAtBlockLeft Fatal "Expected symbol, but found indented expression"
    InBlock -> assertInsideBlock Fatal "Expected symbol, but found end of indented expression"
  name <- (:) <$> (case cs of Lowercase -> satisfy (\c -> isSymbolChar c && c /= ':')
                              Uppercase -> satisfy (== ':'))
              <*> many (satisfy isSymbolChar)
  guard (name `notElem` ["..", ":", "::", "=", "\\", "|", "<-", "->", "@", "~", "=>"])
  guard (take 2 name /= "--")
  return (Name name)

isSymbolChar :: Char -> Bool
isSymbolChar c = (isAscSymbol || isUniSymbol) && not isSpecialExt
  where
    isSpecialExt = c `elem` "(),;[]`{}_\"'"
    isAscSymbol = c `elem` "!#$%&⋆+./<=>?@^|-~:"
    isUniSymbol = isSymbol c || isPunctuation c


sepBy1 :: Parser a -> Parser sep -> Parser [a]
sepBy1 p psep = do
  x1 <- p
  (psep >> (x1 :) <$> sepBy1 p psep) <|> return [x1]

sepBy :: Parser a -> Parser sep -> Parser [a]
sepBy p psep = sepBy1 p psep <|> return []

-- | Start a new layout block at the current position. The old layout block is
-- restored after completion of this subparser.
startLayoutBlock :: Parser a -> Parser a
startLayoutBlock p = do
  ps0 <- get
  put (ps0 { psBlkLine = psLine ps0
           , psBlkCol = psCol ps0 })
  res <- p
  modify (\ps -> ps { psBlkLine = psBlkLine ps0
                    , psBlkCol = psBlkCol ps0 })
  return res

data Fatality fatal where
  Error :: Fatality 'False
  Fatal :: Fatality 'True
deriving instance Show (Fatality fatal)

type family FatalCtx fatal a where
  FatalCtx 'False a = a ~ ()
  FatalCtx 'True a = ()

-- | Raise an error with the given fatality and description.
raise :: FatalCtx fatal a => Fatality fatal -> String -> Parser a
raise fat msg = do
  Context { ctxFile = fp , ctxStack = stk } <- ask
  PS { psLine = line, psCol = col } <- get
  let fun = case fat of
              Error -> dictate . pure
              Fatal -> confess . pure
  fun (ErrMsg fp stk line col msg)

raise' :: Fatality fatal -> String -> Parser ()
raise' Error = raise Error
raise' Fatal = raise Fatal

-- | Registers a scope description on the stack for error reporting.
pushContext :: String -> Parser a -> Parser a
pushContext descr = local (\c -> c { ctxStack = descr : ctxStack c })

data BlockPos = AtLeft | InBlock
  deriving (Show)

-- | Consumes a token at the current position, asserting that we are
-- in the position indicated by the 'BlockPos' argument. The token is defined
-- by a pure stateful parser. If encountering a newline or EOF, the parser is
-- run on this character ('Nothing' for EOF); if this produces a result, the
-- result is returned; otherwise, the parser fails. The newline is not consumed.
readToken :: BlockPos -> (s -> Maybe Char -> Maybe (Either r s)) -> s -> Parser (r, String)
readToken bpos f s0 = do
  case bpos of
    AtLeft -> assertAtBlockLeft Fatal "Expected token, but found indented expression"
    InBlock -> assertInsideBlock Fatal "Expected token, but found end of indented expression"
  let loop :: (s -> Maybe Char -> Maybe (Either r s)) -> s -> Parser (r, String)
      loop f' st = do
        ps <- get
        case psRest ps of
          []       | Just (Left res) <- f' st Nothing     -> return (res, "")
                   | otherwise -> empty
          '\n' : _ | Just (Left res) <- f' st (Just '\n') -> return (res, "")
          c : cs -> case f' st (Just c) of
                      Nothing -> empty
                      Just (Left res) -> return (res, "")
                      Just (Right st') -> do
                          put (ps { psCol = psCol ps + 1, psRest = cs })
                          fmap (c :) <$> loop f' st'
  loop f s0

-- | Consumes all whitespace and comments (including newlines), but only if
-- this then leaves the parser inside the current block. If not, this fails.
inlineWhite :: Parser ()
inlineWhite = do
  skipWhiteComment
  whenM (not <$> isInsideBlock) empty

-- | Consumes all whitespace and comments (including newlines). Note: this may
-- end outside the current block.
skipWhiteComment :: Parser ()
skipWhiteComment = do
  inlineSpaces
  _ <- many (blockComment >> inlineSpaces)
  optional_ lineComment
  optional_ (consumeNewline >> skipWhiteComment)

-- | Consumes some inline whitespace. Stops before newlines.
inlineSpaces :: Parser ()
inlineSpaces = readWhileInline isSpace

-- | Consumes an delimited comment including both end markers. Note: this may
-- end outside the current block.
blockComment :: Parser ()
blockComment = do
  string "{-"  -- no need for pKeySym here
  let loop = do
        readWhileInline (`notElem` "{-")  -- "-}" also starts with '-'
        asum [string "-}"
             ,eof >> raise Error "Unfinished {- -} comment at end of file"
             ,blockComment >> loop
             ,consumeNewline >> loop]
  loop

-- | Consumes a line comment marker and the rest of the line, excluding
-- newline.
lineComment :: Parser ()
lineComment = do
  -- '--!' is an operator, so we need to parse a whole symbol here.
  pKeySym "--"
  readWhileInline (const True)

-- | Raises an error if we're not currently at the given column.
assertAtBlockLeft :: Fatality fatal -> String -> Parser ()
assertAtBlockLeft fat msg = whenM (not <$> isAtBlockLeft) $ raise' fat msg

-- | Raises an error if psCol is not greater than psRefCol.
assertInsideBlock :: Fatality fatal -> String -> Parser ()
assertInsideBlock fat msg = whenM (not <$> isInsideBlock) $ raise' fat msg

-- | Raises an error if we're not currently at EOF.
assertEOF :: Fatality fatal -> Parser ()
assertEOF fat = gets psRest >>= \case
  [] -> return ()
  _ -> raise' fat "Unexpected stuff"

-- | Returns whether the current position is _within_ the current block, for
-- soft-wrapping content. This means that col > blkCol.
isInsideBlock :: Parser Bool
isInsideBlock = do
  ps <- get
  return $ psLine ps >= psBlkLine ps && psCol ps > psBlkCol ps

-- | Returns whether the current position is at the left border of the block;
-- this is for list content such as function definitions or let bindings. This
-- means that col == blkCol.
isAtBlockLeft :: Parser Bool
isAtBlockLeft = do
  ps <- get
  return $ psLine ps >= psBlkLine ps && psCol ps == psBlkCol ps

-- | Consumes characters while the predicate holds or until (and excluding)
-- a newline, whichever comes first.
readWhileInline :: (Char -> Bool) -> Parser ()
readWhileInline p = do
  (taken, rest) <- span (\c -> p c && c /= '\n') <$> gets psRest
  modify (\ps -> ps { psCol = psCol ps + length taken
                    , psRest = rest })

-- | Consumes exactly one newline at the current position.
consumeNewline :: Parser ()
consumeNewline = gets psRest >>= \case
  '\n' : rest -> modify (\ps -> ps { psLine = psLine ps + 1
                                   , psCol = 0
                                   , psRest = rest })
  _ -> empty

-- | Consumes exactly one character, unequal to newline, at the current position.
satisfy :: (Char -> Bool) -> Parser Char
satisfy p = do
  traceM "entering satisfy"
  r <- gets psRest
  traceM "got rest"
  r `seq` return ()
  traceM "seqd rest"
  traceM ("rest is " ++ show r)
  case r of
    c : rest | c /= '\n', p c -> do
        modify (\ps -> ps { psCol = psCol ps + 1
                          , psRest = rest })
        return c
    _ -> empty

-- | Consumes exactly this character at the current position. Must not be a
-- newline.
char :: Char -> Parser ()
char c = string [c]

-- | Consumes exactly this string at the current position. The string must not
-- contain a newline.
string :: String -> Parser ()
string s | any (== '\n') s = error "Newline in 'string' argument"
string s = do
  ps <- get
  if take (length s) (psRest ps) == s
    then put (ps { psCol = psCol ps + length s
                 , psRest = drop (length s) (psRest ps) })
    else empty

lookAhead :: Parser () -> Parser ()
lookAhead p = do
  ps <- get
  success <- absolve False (True <$ p)
  put ps  -- restore state, as if nothing happened
  when (not success) empty

notFollowedBy :: Parser () -> Parser ()
notFollowedBy p = do
  ps <- get
  success <- absolve True (False <$ p)
  put ps  -- restore state, as if nothing happened
  when (not success) empty

-- | Only succeeds at EOF.
eof :: Parser ()
eof = gets psRest >>= \case [] -> return ()
                            _ -> empty

whenM :: (Monad m, Monoid a) => m Bool -> m a -> m a
whenM mb mx = mb >>= \b -> if b then mx else return mempty

optional_ :: Alternative f => f a -> f ()
optional_ a = (() <$ a) <|> pure ()

someNE :: Alternative f => f a -> f (NonEmpty a)
someNE a = (:|) <$> a <*> many a