Parse.hs

{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE ViewPatterns #-}

-- |
-- SPDX-License-Identifier: BSD-3-Clause
--
-- Parser for the Swarm language.  Note, you probably don't want to
-- use this directly, unless there is a good reason to parse a term
-- without also type checking it; use
-- 'Swarm.Language.Pipeline.processTerm' instead, which parses,
-- typechecks, elaborates, and capability checks a term all at once.
module Swarm.Language.Parse (
  -- * Reserved words
  reservedWords,

  -- * Parsers
  Parser,
  parsePolytype,
  parseType,
  parseTerm,
  binOps,
  unOps,

  -- * Utility functions
  runParser,
  runParserTH,
  readTerm,
  readTerm',
  showShortError,
  showErrorPos,
  getLocRange,
  unTuple,
) where

import Control.Lens (view, (^.))
import Control.Monad (guard, join)
import Control.Monad.Combinators.Expr
import Control.Monad.Reader (
  MonadReader (ask),
  ReaderT (runReaderT),
 )
import Data.Bifunctor
import Data.Foldable (asum)
import Data.List (foldl', nub)
import Data.List.NonEmpty qualified (head)
import Data.Map.Strict (Map)
import Data.Map.Strict qualified as Map
import Data.Maybe (fromMaybe, mapMaybe)
import Data.Set qualified as S
import Data.Set.Lens (setOf)
import Data.Text (Text, index, toLower)
import Data.Text qualified as T
import Data.Void
import Swarm.Language.Syntax
import Swarm.Language.Types
import Swarm.Util (failT, findDup, squote)
import Swarm.Util.Parse (fully, fullyMaybe)
import Text.Megaparsec hiding (runParser)
import Text.Megaparsec.Char
import Text.Megaparsec.Char.Lexer qualified as L
import Text.Megaparsec.Pos qualified as Pos
import Witch

-- Imports for doctests (cabal-docspec needs this)

-- $setup
-- >>> import qualified Data.Map.Strict as Map

-- | When parsing a term using a quasiquoter (i.e. something in the
--   Swarm source code that will be parsed at compile time), we want
--   to allow antiquoting, i.e. writing something like $x to refer to
--   an existing Haskell variable.  But when parsing a term entered by
--   the user at the REPL, we do not want to allow this syntax.
data Antiquoting = AllowAntiquoting | DisallowAntiquoting
  deriving (Eq, Ord, Show)

type Parser = ReaderT Antiquoting (Parsec Void Text)

type ParserError = ParseErrorBundle Text Void

--------------------------------------------------
-- Lexer

-- | List of reserved words that cannot be used as variable names.
reservedWords :: [Text]
reservedWords =
  map (syntax . constInfo) (filter isUserFunc allConst)
    ++ map directionSyntax allDirs
    ++ [ "void"
       , "unit"
       , "int"
       , "text"
       , "dir"
       , "bool"
       , "actor"
       , "key"
       , "cmd"
       , "delay"
       , "let"
       , "def"
       , "end"
       , "in"
       , "true"
       , "false"
       , "forall"
       , "require"
       , "requirements"
       ]

-- | Skip spaces and comments.
sc :: Parser ()
sc =
  L.space
    space1
    (L.skipLineComment "//")
    (L.skipBlockComment "/*" "*/")

-- | In general, we follow the convention that every token parser
--   assumes no leading whitespace and consumes all trailing
--   whitespace.  Concretely, we achieve this by wrapping every token
--   parser using 'lexeme'.
lexeme :: Parser a -> Parser a
lexeme = L.lexeme sc

-- | A lexeme consisting of a literal string.
symbol :: Text -> Parser Text
symbol = L.symbol sc

-- | Parse a case-insensitive reserved word, making sure it is not a
--   prefix of a longer variable name, and allowing the parser to
--   backtrack if it fails.
reserved :: Text -> Parser ()
reserved w = (lexeme . try) $ string' w *> notFollowedBy (alphaNumChar <|> char '_')

-- | Parse an identifier, i.e. any non-reserved string containing
--   alphanumeric characters and underscores and not starting with a
--   number.
identifier :: Parser Var
identifier = lvVar <$> locIdentifier

-- | Parse an identifier together with its source location info.
locIdentifier :: Parser LocVar
locIdentifier = uncurry LV <$> parseLocG ((lexeme . try) (p >>= check) <?> "variable name")
 where
  p = (:) <$> (letterChar <|> char '_') <*> many (alphaNumChar <|> char '_' <|> char '\'')
  check (into @Text -> t)
    | toLower t `elem` reservedWords =
        failT ["reserved word", squote t, "cannot be used as variable name"]
    | otherwise = return t

-- | Parse a text literal (including escape sequences) in double quotes.
textLiteral :: Parser Text
textLiteral = into <$> lexeme (char '"' >> manyTill L.charLiteral (char '"'))

-- | Parse a positive integer literal token, in decimal, binary,
--   octal, or hexadecimal notation.  Note that negation is handled as
--   a separate operator.
integer :: Parser Integer
integer =
  label "integer literal" $
    lexeme $ do
      n <-
        string "0b" *> L.binary
          <|> string "0o" *> L.octal
          <|> string "0x" *> L.hexadecimal
          <|> L.decimal
      notFollowedBy alphaNumChar
      return n

braces :: Parser a -> Parser a
braces = between (symbol "{") (symbol "}")

parens :: Parser a -> Parser a
parens = between (symbol "(") (symbol ")")

brackets :: Parser a -> Parser a
brackets = between (symbol "[") (symbol "]")

--------------------------------------------------
-- Parser

-- | Parse a Swarm language polytype, which starts with an optional
--   quanitifation (@forall@ followed by one or more variables and a
--   period) followed by a type.  Note that anything accepted by
--   'parseType' is also accepted by 'parsePolytype'.
parsePolytype :: Parser Polytype
parsePolytype =
  join $
    quantify
      <$> (fromMaybe [] <$> optional (reserved "forall" *> some identifier <* symbol "."))
      <*> parseType
 where
  quantify :: [Var] -> Type -> Parser Polytype
  quantify xs ty
    -- Iplicitly quantify over free type variables if the user didn't write a forall
    | null xs = return $ Forall (S.toList free) ty
    -- Otherwise, require all variables to be explicitly quantified
    | S.null free = return $ Forall xs ty
    | otherwise =
        fail $
          unlines
            [ "  Type contains free variable(s): " ++ unwords (map from (S.toList free))
            , "  Try adding them to the 'forall'."
            ]
   where
    free = tyVars ty `S.difference` S.fromList xs

-- | Parse a Swarm language (mono)type.
parseType :: Parser Type
parseType = makeExprParser parseTypeAtom table
 where
  table =
    [ [InfixR ((:*:) <$ symbol "*")]
    , [InfixR ((:+:) <$ symbol "+")]
    , [InfixR ((:->:) <$ symbol "->")]
    ]

parseTypeAtom :: Parser Type
parseTypeAtom =
  TyVoid <$ reserved "void"
    <|> TyUnit <$ reserved "unit"
    <|> TyVar <$> identifier
    <|> TyInt <$ reserved "int"
    <|> TyText <$ reserved "text"
    <|> TyDir <$ reserved "dir"
    <|> TyBool <$ reserved "bool"
    <|> TyActor <$ reserved "actor"
    <|> TyKey <$ reserved "key"
    <|> TyCmd <$> (reserved "cmd" *> parseTypeAtom)
    <|> TyDelay <$> braces parseType
    <|> TyRcd <$> brackets (parseRecord (symbol ":" *> parseType))
    <|> parens parseType

parseRecord :: Parser a -> Parser (Map Var a)
parseRecord p = (parseBinding `sepBy` symbol ",") >>= fromListUnique
 where
  parseBinding = (,) <$> identifier <*> p
  fromListUnique kvs = case findDup (map fst kvs) of
    Nothing -> return $ Map.fromList kvs
    Just x -> failT ["duplicate field name", squote x, "in record literal"]

parseDirection :: Parser Direction
parseDirection = asum (map alternative allDirs) <?> "direction constant"
 where
  alternative d = d <$ (reserved . directionSyntax) d

-- | Parse Const as reserved words (e.g. @Fail <$ reserved "fail"@)
parseConst :: Parser Const
parseConst = asum (map alternative consts) <?> "built-in user function"
 where
  consts = filter isUserFunc allConst
  alternative c = c <$ reserved (syntax $ constInfo c)

-- | Add 'SrcLoc' to a parser
parseLocG :: Parser a -> Parser (SrcLoc, a)
parseLocG pa = do
  start <- getOffset
  a <- pa
  end <- getOffset
  pure (SrcLoc start end, a)

-- | Add 'SrcLoc' to a 'Term' parser
parseLoc :: Parser Term -> Parser Syntax
parseLoc pterm = uncurry Syntax <$> parseLocG pterm

-- | Parse an atomic term, optionally trailed by record projections like @t.x.y.z@.
--   Record projection binds more tightly than function application.
parseTermAtom :: Parser Syntax
parseTermAtom = do
  s1 <- parseTermAtom2
  ps <- many (symbol "." *> parseLocG identifier)
  return $ foldl' (\(Syntax l1 t) (l2, x) -> Syntax (l1 <> l2) (TProj t x)) s1 ps

-- | Parse an atomic term.
parseTermAtom2 :: Parser Syntax
parseTermAtom2 =
  parseLoc
    ( TUnit <$ symbol "()"
        <|> TConst <$> parseConst
        <|> TVar <$> identifier
        <|> TDir <$> parseDirection
        <|> TInt <$> integer
        <|> TText <$> textLiteral
        <|> TBool <$> ((True <$ reserved "true") <|> (False <$ reserved "false"))
        <|> reserved "require"
          *> ( ( TRequireDevice
                  <$> (textLiteral <?> "device name in double quotes")
               )
                <|> ( TRequire
                        <$> (fromIntegral <$> integer)
                        <*> (textLiteral <?> "entity name in double quotes")
                    )
             )
        <|> uncurry SRequirements <$> (reserved "requirements" *> match parseTerm)
        <|> SLam
          <$> (symbol "\\" *> locIdentifier)
          <*> optional (symbol ":" *> parseType)
          <*> (symbol "." *> parseTerm)
        <|> sLet
          <$> (reserved "let" *> locIdentifier)
          <*> optional (symbol ":" *> parsePolytype)
          <*> (symbol "=" *> parseTerm)
          <*> (reserved "in" *> parseTerm)
        <|> sDef
          <$> (reserved "def" *> locIdentifier)
          <*> optional (symbol ":" *> parsePolytype)
          <*> (symbol "=" *> parseTerm <* reserved "end")
        <|> SRcd <$> brackets (parseRecord (optional (symbol "=" *> parseTerm)))
        <|> parens (view sTerm . mkTuple <$> (parseTerm `sepBy` symbol ","))
    )
    -- Potential syntax for explicitly requesting memoized delay.
    -- Perhaps we will not need this in the end; see the discussion at
    -- https://github.com/swarm-game/swarm/issues/150 .
    -- <|> parseLoc (TDelay SimpleDelay (TConst Noop) <$ try (symbol "{{" *> symbol "}}"))
    -- <|> parseLoc (SDelay MemoizedDelay <$> dbraces parseTerm)

    <|> parseLoc (TDelay SimpleDelay (TConst Noop) <$ try (symbol "{" *> symbol "}"))
    <|> parseLoc (SDelay SimpleDelay <$> braces parseTerm)
    <|> parseLoc (ask >>= (guard . (== AllowAntiquoting)) >> parseAntiquotation)

mkTuple :: [Syntax] -> Syntax
mkTuple [] = Syntax NoLoc TUnit -- should never happen
mkTuple [x] = x
mkTuple (x : xs) = let r = mkTuple xs in loc x r $ SPair x r
 where
  loc a b = Syntax $ (a ^. sLoc) <> (b ^. sLoc)

unTuple :: Syntax' ty -> [Syntax' ty]
unTuple = \case
  Syntax' _ (SPair s1 s2) _ -> s1 : unTuple s2
  s -> [s]

-- | Construct an 'SLet', automatically filling in the Boolean field
--   indicating whether it is recursive.
sLet :: LocVar -> Maybe Polytype -> Syntax -> Syntax -> Term
sLet x ty t1 = SLet (lvVar x `S.member` setOf freeVarsV t1) x ty t1

-- | Construct an 'SDef', automatically filling in the Boolean field
--   indicating whether it is recursive.
sDef :: LocVar -> Maybe Polytype -> Syntax -> Term
sDef x ty t = SDef (lvVar x `S.member` setOf freeVarsV t) x ty t

parseAntiquotation :: Parser Term
parseAntiquotation =
  TAntiText <$> (lexeme . try) (symbol "$str:" *> identifier)
    <|> TAntiInt <$> (lexeme . try) (symbol "$int:" *> identifier)

-- | Parse a Swarm language term.
parseTerm :: Parser Syntax
parseTerm = sepEndBy1 parseStmt (symbol ";") >>= mkBindChain

mkBindChain :: [Stmt] -> Parser Syntax
mkBindChain stmts = case last stmts of
  Binder x _ -> return $ foldr mkBind (STerm (TApp (TConst Return) (TVar (lvVar x)))) stmts
  BareTerm t -> return $ foldr mkBind t (init stmts)
 where
  mkBind (BareTerm t1) t2 = loc Nothing t1 t2 $ SBind Nothing t1 t2
  mkBind (Binder x t1) t2 = loc (Just x) t1 t2 $ SBind (Just x) t1 t2
  loc mx a b = Syntax $ maybe NoLoc lvSrcLoc mx <> (a ^. sLoc) <> (b ^. sLoc)

data Stmt
  = BareTerm Syntax
  | Binder LocVar Syntax
  deriving (Show)

parseStmt :: Parser Stmt
parseStmt =
  mkStmt <$> optional (try (locIdentifier <* symbol "<-")) <*> parseExpr

mkStmt :: Maybe LocVar -> Syntax -> Stmt
mkStmt Nothing = BareTerm
mkStmt (Just x) = Binder x

-- | When semicolons are missing between definitions, for example:
--     def a = 1 end def b = 2 end def c = 3 end
--   The makeExprParser produces:
--     App (App (TDef a) (TDef b)) (TDef x)
--   This function fix that by converting the Apps into Binds, so that it results in:
--     Bind a (Bind b (Bind c))
fixDefMissingSemis :: Syntax -> Syntax
fixDefMissingSemis term =
  case nestedDefs term [] of
    [] -> term
    defs -> foldr1 mkBind defs
 where
  mkBind t1 t2 = Syntax ((t1 ^. sLoc) <> (t2 ^. sLoc)) $ SBind Nothing t1 t2
  nestedDefs term' acc = case term' of
    def@(Syntax _ SDef {}) -> def : acc
    (Syntax _ (SApp nestedTerm def@(Syntax _ SDef {}))) -> nestedDefs nestedTerm (def : acc)
    -- Otherwise returns an empty list to keep the term unchanged
    _ -> []

parseExpr :: Parser Syntax
parseExpr =
  parseLoc $ ascribe <$> parseExpr' <*> optional (symbol ":" *> parsePolytype)
 where
  ascribe :: Syntax -> Maybe Polytype -> Term
  ascribe s Nothing = s ^. sTerm
  ascribe s (Just ty) = SAnnotate s ty

parseExpr' :: Parser Syntax
parseExpr' = fixDefMissingSemis <$> makeExprParser parseTermAtom table
 where
  table = snd <$> Map.toDescList tableMap
  tableMap =
    Map.unionsWith
      (++)
      [ Map.singleton 9 [InfixL (exprLoc2 $ SApp <$ string "")]
      , binOps
      , unOps
      ]

  -- add location for ExprParser by combining all
  exprLoc2 :: Parser (Syntax -> Syntax -> Term) -> Parser (Syntax -> Syntax -> Syntax)
  exprLoc2 p = do
    (l, f) <- parseLocG p
    pure $ \s1 s2 -> Syntax (l <> (s1 ^. sLoc) <> (s2 ^. sLoc)) $ f s1 s2

-- | Precedences and parsers of binary operators.
--
-- >>> Map.map length binOps
-- fromList [(0,1),(2,1),(3,1),(4,6),(6,3),(7,2),(8,1)]
binOps :: Map.Map Int [Operator Parser Syntax]
binOps = Map.unionsWith (++) $ mapMaybe binOpToTuple allConst
 where
  binOpToTuple c = do
    let ci = constInfo c
    ConstMBinOp assoc <- pure (constMeta ci)
    let assI = case assoc of
          L -> InfixL
          N -> InfixN
          R -> InfixR
    pure $
      Map.singleton
        (fixity ci)
        [assI (mkOp c <$ operatorString (syntax ci))]

-- | Precedences and parsers of unary operators (currently only 'Neg').
--
-- >>> Map.map length unOps
-- fromList [(7,1)]
unOps :: Map.Map Int [Operator Parser Syntax]
unOps = Map.unionsWith (++) $ mapMaybe unOpToTuple allConst
 where
  unOpToTuple c = do
    let ci = constInfo c
    ConstMUnOp assoc <- pure (constMeta ci)
    let assI = case assoc of
          P -> Prefix
          S -> Postfix
    pure $
      Map.singleton
        (fixity ci)
        [assI (exprLoc1 $ SApp (noLoc $ TConst c) <$ operatorString (syntax ci))]

  -- combine location for ExprParser
  exprLoc1 :: Parser (Syntax -> Term) -> Parser (Syntax -> Syntax)
  exprLoc1 p = do
    (l, f) <- parseLocG p
    pure $ \s -> Syntax (l <> s ^. sLoc) $ f s

operatorString :: Text -> Parser Text
operatorString n = (lexeme . try) (string n <* notFollowedBy operatorSymbol)

operatorSymbol :: Parser Text
operatorSymbol = T.singleton <$> oneOf opChars
 where
  isOp = \case { ConstMFunc {} -> False; _ -> True } . constMeta
  opChars = nub . concatMap (from . syntax) . filter isOp $ map constInfo allConst

--------------------------------------------------
-- Utilities

-- | Run a parser on some input text, returning either the result or a
--   pretty-printed parse error message.
runParser :: Parser a -> Text -> Either Text a
runParser p t = first (from . errorBundlePretty) (parse (runReaderT p DisallowAntiquoting) "" t)

-- | A utility for running a parser in an arbitrary 'MonadFail' (which
--   is going to be the TemplateHaskell 'Q' monad --- see
--   "Swarm.Language.Parse.QQ"), with a specified source position.
runParserTH :: (Monad m, MonadFail m) => (String, Int, Int) -> Parser a -> String -> m a
runParserTH (file, line, col) p s =
  case snd (runParser' (runReaderT (fully sc p) AllowAntiquoting) initState) of
    Left err -> fail $ errorBundlePretty err
    Right e -> return e
 where
  -- This is annoying --- megaparsec does not export its function to
  -- construct an initial parser state, so we can't just use that
  -- and then change the one field we need to be different (the
  -- pstateSourcePos). We have to copy-paste the whole thing.
  initState :: State Text Void
  initState =
    State
      { stateInput = from s
      , stateOffset = 0
      , statePosState =
          PosState
            { pstateInput = from s
            , pstateOffset = 0
            , pstateSourcePos = SourcePos file (mkPos line) (mkPos col)
            , pstateTabWidth = defaultTabWidth
            , pstateLinePrefix = ""
            }
      , stateParseErrors = []
      }

-- | Parse some input 'Text' completely as a 'Term', consuming leading
--   whitespace and ensuring the parsing extends all the way to the
--   end of the input 'Text'.  Returns either the resulting 'Term' (or
--   @Nothing@ if the input was only whitespace) or a pretty-printed
--   parse error message.
readTerm :: Text -> Either Text (Maybe Syntax)
readTerm = runParser (fullyMaybe sc parseTerm)

-- | A lower-level `readTerm` which returns the megaparsec bundle error
--   for precise error reporting.
readTerm' :: Text -> Either ParserError (Maybe Syntax)
readTerm' = parse (runReaderT (fullyMaybe sc parseTerm) DisallowAntiquoting) ""

-- | A utility for converting a ParserError into a one line message:
--   <line-nr>: <error-msg>
showShortError :: ParserError -> String
showShortError pe = show (line + 1) <> ": " <> from msg
 where
  ((line, _), _, msg) = showErrorPos pe

-- | A utility for converting a ParseError into a range and error message.
showErrorPos :: ParserError -> ((Int, Int), (Int, Int), Text)
showErrorPos (ParseErrorBundle errs sourcePS) = (minusOne start, minusOne end, from msg)
 where
  -- convert megaparsec source pos to starts at 0
  minusOne (x, y) = (x - 1, y - 1)

  -- get the first error position (ps) and line content (str)
  err = Data.List.NonEmpty.head errs
  offset = case err of
    TrivialError x _ _ -> x
    FancyError x _ -> x
  (str, ps) = reachOffset offset sourcePS
  msg = parseErrorTextPretty err

  -- extract the error starting position
  start@(line, col) = getLineCol ps

  -- compute the ending position based on the word at starting position
  wordlength = case break (== ' ') . drop col <$> str of
    Just (word, _) -> length word + 1
    _ -> 0
  end = (line, col + wordlength)

getLineCol :: PosState a -> (Int, Int)
getLineCol ps = (line, col)
 where
  line = unPos $ sourceLine $ pstateSourcePos ps
  col = unPos $ sourceColumn $ pstateSourcePos ps

-- | A utility for converting a SrcLoc into a range
getLocRange :: Text -> (Int, Int) -> ((Int, Int), (Int, Int))
getLocRange code (locStart, locEnd) = (start, end)
 where
  start = getLocPos locStart
  end = getLocPos (dropWhiteSpace locEnd)

  -- remove trailing whitespace that got included by the lexer
  dropWhiteSpace offset
    | isWhiteSpace offset = dropWhiteSpace (offset - 1)
    | otherwise = offset
  isWhiteSpace offset =
    -- Megaparsec offset needs to be (-1) to start at 0
    Data.Text.index code (offset - 1) `elem` [' ', '\n', '\r', '\t']

  -- using megaparsec offset facility, compute the line/col
  getLocPos offset =
    let sourcePS =
          PosState
            { pstateInput = code
            , pstateOffset = 0
            , pstateSourcePos = Pos.initialPos ""
            , pstateTabWidth = Pos.defaultTabWidth
            , pstateLinePrefix = ""
            }
        (_, ps) = reachOffset offset sourcePS
     in getLineCol ps