Lexer.x

-----------------------------------------------------------------------------
-- (c) The University of Glasgow, 2006
--
-- GHC's lexer for Haskell 2010 [1].
--
-- This is a combination of an Alex-generated lexer [2] from a regex
-- definition, with some hand-coded bits. [3]
--
-- Completely accurate information about token-spans within the source
-- file is maintained.  Every token has a start and end RealSrcLoc
-- attached to it.
--
-- References:
-- [1] https://www.haskell.org/onlinereport/haskell2010/haskellch2.html
-- [2] http://www.haskell.org/alex/
-- [3] https://ghc.haskell.org/trac/ghc/wiki/Commentary/Compiler/Parser
--
-----------------------------------------------------------------------------

--   ToDo / known bugs:
--    - parsing integers is a bit slow
--    - readRational is a bit slow
--
--   Known bugs, that were also in the previous version:
--    - M... should be 3 tokens, not 1.
--    - pragma-end should be only valid in a pragma

--   qualified operator NOTES.
--
--   - If M.(+) is a single lexeme, then..
--     - Probably (+) should be a single lexeme too, for consistency.
--       Otherwise ( + ) would be a prefix operator, but M.( + ) would not be.
--     - But we have to rule out reserved operators, otherwise (..) becomes
--       a different lexeme.
--     - Should we therefore also rule out reserved operators in the qualified
--       form?  This is quite difficult to achieve.  We don't do it for
--       qualified varids.


-- -----------------------------------------------------------------------------
-- Alex "Haskell code fragment top"

{
-- XXX The above flags turn off warnings in the generated code:
{-# LANGUAGE BangPatterns #-}
{-# LANGUAGE DeriveDataTypeable #-}
{-# OPTIONS_GHC -fno-warn-unused-matches #-}
{-# OPTIONS_GHC -fno-warn-unused-binds #-}
{-# OPTIONS_GHC -fno-warn-unused-imports #-}
{-# OPTIONS_GHC -fno-warn-missing-signatures #-}
-- But alex still generates some code that causes the "lazy unlifted bindings"
-- warning, and old compilers don't know about it so we can't easily turn
-- it off, so for now we use the sledge hammer:
{-# OPTIONS_GHC -w #-}

{-# OPTIONS_GHC -funbox-strict-fields #-}

module Lexer (
   Token(..), lexer, pragState, mkPState, PState(..),
   P(..), ParseResult(..), getSrcLoc,
   getPState, getDynFlags, withThisPackage,
   failLocMsgP, failSpanMsgP, srcParseFail,
   getMessages,
   popContext, pushCurrentContext, setLastToken, setSrcLoc,
   activeContext, nextIsEOF,
   getLexState, popLexState, pushLexState,
   extension, bangPatEnabled, datatypeContextsEnabled,
   traditionalRecordSyntaxEnabled,
   explicitForallEnabled,
   inRulePrag,
   explicitNamespacesEnabled,
   patternSynonymsEnabled,
   sccProfilingOn, hpcEnabled,
   addWarning,
   lexTokenStream,
   addAnnotation
  ) where

-- base
import Control.Applicative
import Control.Monad
import Data.Bits
import Data.Char
import Data.List
import Data.Maybe
import Data.Ratio
import Data.Word

-- bytestring
import Data.ByteString (ByteString)

-- containers
import Data.Map (Map)
import qualified Data.Map as Map

-- data/typeable
import Data.Data
import Data.Typeable

-- compiler/utils
import Bag
import Outputable
import StringBuffer
import FastString
import UniqFM
import Util             ( readRational )

-- compiler/main
import ErrUtils
import DynFlags

-- compiler/basicTypes
import SrcLoc
import Module
import BasicTypes       ( InlineSpec(..), RuleMatchInfo(..), FractionalLit(..) )

-- compiler/parser
import Ctype

import ApiAnnotation
}

-- -----------------------------------------------------------------------------
-- Alex "Character set macros"

-- NB: The logic behind these definitions is also reflected in basicTypes/Lexeme.hs
-- Any changes here should likely be reflected there.
$unispace    = \x05 -- Trick Alex into handling Unicode. See alexGetByte.
$nl          = [\n\r\f]
$whitechar   = [$nl\v\ $unispace]
$white_no_nl = $whitechar # \n -- TODO #8424
$tab         = \t

$ascdigit  = 0-9
$unidigit  = \x03 -- Trick Alex into handling Unicode. See alexGetByte.
$decdigit  = $ascdigit -- for now, should really be $digit (ToDo)
$digit     = [$ascdigit $unidigit]

$special   = [\(\)\,\;\[\]\`\{\}]
$ascsymbol = [\!\#\$\%\&\*\+\.\/\<\=\>\?\@\\\^\|\-\~\:]
$unisymbol = \x04 -- Trick Alex into handling Unicode. See alexGetByte.
$symbol    = [$ascsymbol $unisymbol] # [$special \_\"\']

$unilarge  = \x01 -- Trick Alex into handling Unicode. See alexGetByte.
$asclarge  = [A-Z]
$large     = [$asclarge $unilarge]

$unismall  = \x02 -- Trick Alex into handling Unicode. See alexGetByte.
$ascsmall  = [a-z]
$small     = [$ascsmall $unismall \_]

$unigraphic = \x06 -- Trick Alex into handling Unicode. See alexGetByte.
$graphic   = [$small $large $symbol $digit $special $unigraphic \"\']

$binit     = 0-1
$octit     = 0-7
$hexit     = [$decdigit A-F a-f]
$idchar    = [$small $large $digit \']

$pragmachar = [$small $large $digit]

$docsym    = [\| \^ \* \$]


-- -----------------------------------------------------------------------------
-- Alex "Regular expression macros"

@varid     = $small $idchar*          -- variable identifiers
@conid     = $large $idchar*          -- constructor identifiers

@varsym    = ($symbol # \:) $symbol*  -- variable (operator) symbol
@consym    = \: $symbol*              -- constructor (operator) symbol

@decimal     = $decdigit+
@binary      = $binit+
@octal       = $octit+
@hexadecimal = $hexit+
@exponent    = [eE] [\-\+]? @decimal

@qual = (@conid \.)+
@qvarid = @qual @varid
@qconid = @qual @conid
@qvarsym = @qual @varsym
@qconsym = @qual @consym

@floating_point = @decimal \. @decimal @exponent? | @decimal @exponent

-- normal signed numerical literals can only be explicitly negative,
-- not explicitly positive (contrast @exponent)
@negative = \-
@signed = @negative ?


-- -----------------------------------------------------------------------------
-- Alex "Identifier"

haskell :-


-- -----------------------------------------------------------------------------
-- Alex "Rules"

-- everywhere: skip whitespace
$white_no_nl+ ;
$tab+         { warn Opt_WarnTabs (text "Tab character") }

-- Everywhere: deal with nested comments.  We explicitly rule out
-- pragmas, "{-#", so that we don't accidentally treat them as comments.
-- (this can happen even though pragmas will normally take precedence due to
-- longest-match, because pragmas aren't valid in every state, but comments
-- are). We also rule out nested Haddock comments, if the -haddock flag is
-- set.

"{-" / { isNormalComment } { nested_comment lexToken }

-- Single-line comments are a bit tricky.  Haskell 98 says that two or
-- more dashes followed by a symbol should be parsed as a varsym, so we
-- have to exclude those.

-- Since Haddock comments aren't valid in every state, we need to rule them
-- out here.

-- The following two rules match comments that begin with two dashes, but
-- continue with a different character. The rules test that this character
-- is not a symbol (in which case we'd have a varsym), and that it's not a
-- space followed by a Haddock comment symbol (docsym) (in which case we'd
-- have a Haddock comment). The rules then munch the rest of the line.

"-- " ~[$docsym \#] .* { lineCommentToken }
"--" [^$symbol \ ] .* { lineCommentToken }

-- Next, match Haddock comments if no -haddock flag

"-- " [$docsym \#] .* / { ifExtension (not . haddockEnabled) } { lineCommentToken }

-- Now, when we've matched comments that begin with 2 dashes and continue
-- with a different character, we need to match comments that begin with three
-- or more dashes (which clearly can't be Haddock comments). We only need to
-- make sure that the first non-dash character isn't a symbol, and munch the
-- rest of the line.

"---"\-* ~$symbol .* { lineCommentToken }

-- Since the previous rules all match dashes followed by at least one
-- character, we also need to match a whole line filled with just dashes.

"--"\-* / { atEOL } { lineCommentToken }

-- We need this rule since none of the other single line comment rules
-- actually match this case.

"-- " / { atEOL } { lineCommentToken }

-- 'bol' state: beginning of a line.  Slurp up all the whitespace (including
-- blank lines) until we find a non-whitespace character, then do layout
-- processing.
--
-- One slight wibble here: what if the line begins with {-#? In
-- theory, we have to lex the pragma to see if it's one we recognise,
-- and if it is, then we backtrack and do_bol, otherwise we treat it
-- as a nested comment.  We don't bother with this: if the line begins
-- with {-#, then we'll assume it's a pragma we know about and go for do_bol.
<bol> {
  \n                                    ;
  ^\# (line)?                           { begin line_prag1 }
  ^\# pragma .* \n                      ; -- GCC 3.3 CPP generated, apparently
  ^\# \! .* \n                          ; -- #!, for scripts
  ()                                    { do_bol }
}

-- after a layout keyword (let, where, do, of), we begin a new layout
-- context if the curly brace is missing.
-- Careful! This stuff is quite delicate.
<layout, layout_do, layout_if> {
  \{ / { notFollowedBy '-' }            { hopefully_open_brace }
        -- we might encounter {-# here, but {- has been handled already
  \n                                    ;
  ^\# (line)?                           { begin line_prag1 }
}

-- after an 'if', a vertical bar starts a layout context for MultiWayIf
<layout_if> {
  \| / { notFollowedBySymbol }          { new_layout_context True ITvbar }
  ()                                    { pop }
}

-- do is treated in a subtly different way, see new_layout_context
<layout>    ()                          { new_layout_context True  ITvocurly }
<layout_do> ()                          { new_layout_context False ITvocurly }

-- after a new layout context which was found to be to the left of the
-- previous context, we have generated a '{' token, and we now need to
-- generate a matching '}' token.
<layout_left>  ()                       { do_layout_left }

<0,option_prags> \n                     { begin bol }

"{-#" $whitechar* $pragmachar+ / { known_pragma linePrags }
                                { dispatch_pragmas linePrags }

-- single-line line pragmas, of the form
--    # <line> "<file>" <extra-stuff> \n
<line_prag1> @decimal                   { setLine line_prag1a }
<line_prag1a> \" [$graphic \ ]* \"      { setFile line_prag1b }
<line_prag1b> .*                        { pop }

-- Haskell-style line pragmas, of the form
--    {-# LINE <line> "<file>" #-}
<line_prag2> @decimal                   { setLine line_prag2a }
<line_prag2a> \" [$graphic \ ]* \"      { setFile line_prag2b }
<line_prag2b> "#-}"|"-}"                { pop }
   -- NOTE: accept -} at the end of a LINE pragma, for compatibility
   -- with older versions of GHC which generated these.

<0,option_prags> {
  "{-#" $whitechar* $pragmachar+
        $whitechar+ $pragmachar+ / { known_pragma twoWordPrags }
                                 { dispatch_pragmas twoWordPrags }

  "{-#" $whitechar* $pragmachar+ / { known_pragma oneWordPrags }
                                 { dispatch_pragmas oneWordPrags }

  -- We ignore all these pragmas, but don't generate a warning for them
  "{-#" $whitechar* $pragmachar+ / { known_pragma ignoredPrags }
                                 { dispatch_pragmas ignoredPrags }

  -- ToDo: should only be valid inside a pragma:
  "#-}"                          { endPrag }
}

<option_prags> {
  "{-#"  $whitechar* $pragmachar+ / { known_pragma fileHeaderPrags }
                                   { dispatch_pragmas fileHeaderPrags }

  "-- #"                           { multiline_doc_comment }
}

<0> {
  -- In the "0" mode we ignore these pragmas
  "{-#"  $whitechar* $pragmachar+ / { known_pragma fileHeaderPrags }
                     { nested_comment lexToken }
}

<0> {
  "-- #" .* { lineCommentToken }
}

<0,option_prags> {
  "{-#"  { warnThen Opt_WarnUnrecognisedPragmas (text "Unrecognised pragma")
                    (nested_comment lexToken) }
}

-- '0' state: ordinary lexemes

-- Haddock comments

<0,option_prags> {
  "-- " $docsym      / { ifExtension haddockEnabled } { multiline_doc_comment }
  "{-" \ ? $docsym   / { ifExtension haddockEnabled } { nested_doc_comment }
}

-- "special" symbols

<0> {
  "[:" / { ifExtension parrEnabled }    { token ITopabrack }
  ":]" / { ifExtension parrEnabled }    { token ITcpabrack }
}

<0> {
  "[|"        / { ifExtension thEnabled } { token ITopenExpQuote }
  "[||"       / { ifExtension thEnabled } { token ITopenTExpQuote }
  "[e|"       / { ifExtension thEnabled } { token ITopenExpQuote }
  "[e||"      / { ifExtension thEnabled } { token ITopenTExpQuote }
  "[p|"       / { ifExtension thEnabled } { token ITopenPatQuote }
  "[d|"       / { ifExtension thEnabled } { layout_token ITopenDecQuote }
  "[t|"       / { ifExtension thEnabled } { token ITopenTypQuote }
  "|]"        / { ifExtension thEnabled } { token ITcloseQuote }
  "||]"       / { ifExtension thEnabled } { token ITcloseTExpQuote }
  \$ @varid   / { ifExtension thEnabled } { skip_one_varid ITidEscape }
  "$$" @varid / { ifExtension thEnabled } { skip_two_varid ITidTyEscape }
  "$("        / { ifExtension thEnabled } { token ITparenEscape }
  "$$("       / { ifExtension thEnabled } { token ITparenTyEscape }

-- For backward compatibility, accept the old dollar syntax
  "[$" @varid "|"  / { ifExtension qqEnabled }
                     { lex_quasiquote_tok }

  "[" @varid "|"  / { ifExtension qqEnabled }
                     { lex_quasiquote_tok }

  -- qualified quasi-quote (#5555)
  "[" @qvarid "|"  / { ifExtension qqEnabled }
                     { lex_qquasiquote_tok }
}

<0> {
  "(|" / { ifExtension arrowsEnabled `alexAndPred` notFollowedBySymbol }
                                        { special IToparenbar }
  "|)" / { ifExtension arrowsEnabled }  { special ITcparenbar }
}

<0> {
  \? @varid / { ifExtension ipEnabled } { skip_one_varid ITdupipvarid }
}

<0> {
  "(#" / { ifExtension unboxedTuplesEnabled }
         { token IToubxparen }
  "#)" / { ifExtension unboxedTuplesEnabled }
         { token ITcubxparen }
}

<0,option_prags> {
  \(                                    { special IToparen }
  \)                                    { special ITcparen }
  \[                                    { special ITobrack }
  \]                                    { special ITcbrack }
  \,                                    { special ITcomma }
  \;                                    { special ITsemi }
  \`                                    { special ITbackquote }

  \{                                    { open_brace }
  \}                                    { close_brace }
}

<0,option_prags> {
  @qvarid                       { idtoken qvarid }
  @qconid                       { idtoken qconid }
  @varid                        { varid }
  @conid                        { idtoken conid }
}

<0> {
  @qvarid "#"+      / { ifExtension magicHashEnabled } { idtoken qvarid }
  @qconid "#"+      / { ifExtension magicHashEnabled } { idtoken qconid }
  @varid "#"+       / { ifExtension magicHashEnabled } { varid }
  @conid "#"+       / { ifExtension magicHashEnabled } { idtoken conid }
}

-- ToDo: - move `var` and (sym) into lexical syntax?
--       - remove backquote from $special?
<0> {
  @qvarsym                                         { idtoken qvarsym }
  @qconsym                                         { idtoken qconsym }
  @varsym                                          { varsym }
  @consym                                          { consym }
}

-- For the normal boxed literals we need to be careful
-- when trying to be close to Haskell98
<0> {
  -- Normal integral literals (:: Num a => a, from Integer)
  @decimal                                                               { tok_num positive 0 0 decimal }
  0[bB] @binary                / { ifExtension binaryLiteralsEnabled }   { tok_num positive 2 2 binary }
  0[oO] @octal                                                           { tok_num positive 2 2 octal }
  0[xX] @hexadecimal                                                     { tok_num positive 2 2 hexadecimal }
  @negative @decimal           / { ifExtension negativeLiteralsEnabled } { tok_num negative 1 1 decimal }
  @negative 0[bB] @binary      / { ifExtension negativeLiteralsEnabled `alexAndPred`
                                   ifExtension binaryLiteralsEnabled }   { tok_num negative 3 3 binary }
  @negative 0[oO] @octal       / { ifExtension negativeLiteralsEnabled } { tok_num negative 3 3 octal }
  @negative 0[xX] @hexadecimal / { ifExtension negativeLiteralsEnabled } { tok_num negative 3 3 hexadecimal }

  -- Normal rational literals (:: Fractional a => a, from Rational)
  @floating_point                                                        { strtoken tok_float }
  @negative @floating_point    / { ifExtension negativeLiteralsEnabled } { strtoken tok_float }
}

<0> {
  -- Unboxed ints (:: Int#) and words (:: Word#)
  -- It's simpler (and faster?) to give separate cases to the negatives,
  -- especially considering octal/hexadecimal prefixes.
  @decimal                     \# / { ifExtension magicHashEnabled } { tok_primint positive 0 1 decimal }
  0[bB] @binary                \# / { ifExtension magicHashEnabled `alexAndPred`
                                      ifExtension binaryLiteralsEnabled } { tok_primint positive 2 3 binary }
  0[oO] @octal                 \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 octal }
  0[xX] @hexadecimal           \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 hexadecimal }
  @negative @decimal           \# / { ifExtension magicHashEnabled } { tok_primint negative 1 2 decimal }
  @negative 0[bB] @binary      \# / { ifExtension magicHashEnabled `alexAndPred`
                                      ifExtension binaryLiteralsEnabled } { tok_primint negative 3 4 binary }
  @negative 0[oO] @octal       \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 octal }
  @negative 0[xX] @hexadecimal \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 hexadecimal }

  @decimal                     \# \# / { ifExtension magicHashEnabled } { tok_primword 0 2 decimal }
  0[bB] @binary                \# \# / { ifExtension magicHashEnabled `alexAndPred`
                                         ifExtension binaryLiteralsEnabled } { tok_primword 2 4 binary }
  0[oO] @octal                 \# \# / { ifExtension magicHashEnabled } { tok_primword 2 4 octal }
  0[xX] @hexadecimal           \# \# / { ifExtension magicHashEnabled } { tok_primword 2 4 hexadecimal }

  -- Unboxed floats and doubles (:: Float#, :: Double#)
  -- prim_{float,double} work with signed literals
  @signed @floating_point \# / { ifExtension magicHashEnabled } { init_strtoken 1 tok_primfloat }
  @signed @floating_point \# \# / { ifExtension magicHashEnabled } { init_strtoken 2 tok_primdouble }
}

-- Strings and chars are lexed by hand-written code.  The reason is
-- that even if we recognise the string or char here in the regex
-- lexer, we would still have to parse the string afterward in order
-- to convert it to a String.
<0> {
  \'                            { lex_char_tok }
  \"                            { lex_string_tok }
}


-- -----------------------------------------------------------------------------
-- Alex "Haskell code fragment bottom"

{
-- -----------------------------------------------------------------------------
-- The token type

data Token
  = ITas                        -- Haskell keywords
  | ITcase
  | ITclass
  | ITdata
  | ITdefault
  | ITderiving
  | ITdo
  | ITelse
  | IThiding
  | ITforeign
  | ITif
  | ITimport
  | ITin
  | ITinfix
  | ITinfixl
  | ITinfixr
  | ITinstance
  | ITlet
  | ITmodule
  | ITnewtype
  | ITof
  | ITqualified
  | ITthen
  | ITtype
  | ITwhere

  | ITforall                    -- GHC extension keywords
  | ITexport
  | ITlabel
  | ITdynamic
  | ITsafe
  | ITinterruptible
  | ITunsafe
  | ITstdcallconv
  | ITccallconv
  | ITcapiconv
  | ITprimcallconv
  | ITjavascriptcallconv
  | ITmdo
  | ITfamily
  | ITrole
  | ITgroup
  | ITby
  | ITusing
  | ITpattern

  -- Pragmas
  | ITinline_prag InlineSpec RuleMatchInfo
  | ITspec_prag                 -- SPECIALISE
  | ITspec_inline_prag Bool     -- SPECIALISE INLINE (or NOINLINE)
  | ITsource_prag
  | ITrules_prag
  | ITwarning_prag
  | ITdeprecated_prag
  | ITline_prag
  | ITscc_prag
  | ITgenerated_prag
  | ITcore_prag                 -- hdaume: core annotations
  | ITunpack_prag
  | ITnounpack_prag
  | ITann_prag
  | ITclose_prag
  | IToptions_prag String
  | ITinclude_prag String
  | ITlanguage_prag
  | ITvect_prag
  | ITvect_scalar_prag
  | ITnovect_prag
  | ITminimal_prag
  | IToverlappable_prag         -- instance overlap mode
  | IToverlapping_prag          -- instance overlap mode
  | IToverlaps_prag             -- instance overlap mode
  | ITincoherent_prag           -- instance overlap mode
  | ITctype

  | ITdotdot                    -- reserved symbols
  | ITcolon
  | ITdcolon
  | ITequal
  | ITlam
  | ITlcase
  | ITvbar
  | ITlarrow
  | ITrarrow
  | ITat
  | ITtilde
  | ITtildehsh
  | ITdarrow
  | ITminus
  | ITbang
  | ITstar
  | ITdot

  | ITbiglam                    -- GHC-extension symbols

  | ITocurly                    -- special symbols
  | ITccurly
  | ITvocurly
  | ITvccurly
  | ITobrack
  | ITopabrack                  -- [:, for parallel arrays with -XParallelArrays
  | ITcpabrack                  -- :], for parallel arrays with -XParallelArrays
  | ITcbrack
  | IToparen
  | ITcparen
  | IToubxparen
  | ITcubxparen
  | ITsemi
  | ITcomma
  | ITunderscore
  | ITbackquote
  | ITsimpleQuote               --  '

  | ITvarid   FastString        -- identifiers
  | ITconid   FastString
  | ITvarsym  FastString
  | ITconsym  FastString
  | ITqvarid  (FastString,FastString)
  | ITqconid  (FastString,FastString)
  | ITqvarsym (FastString,FastString)
  | ITqconsym (FastString,FastString)
  | ITprefixqvarsym (FastString,FastString)
  | ITprefixqconsym (FastString,FastString)

  | ITdupipvarid   FastString   -- GHC extension: implicit param: ?x

  | ITchar       Char
  | ITstring     FastString
  | ITinteger    Integer
  | ITrational   FractionalLit

  | ITprimchar   Char
  | ITprimstring ByteString
  | ITprimint    Integer
  | ITprimword   Integer
  | ITprimfloat  FractionalLit
  | ITprimdouble FractionalLit

  -- Template Haskell extension tokens
  | ITopenExpQuote              --  [| or [e|
  | ITopenPatQuote              --  [p|
  | ITopenDecQuote              --  [d|
  | ITopenTypQuote              --  [t|
  | ITcloseQuote                --  |]
  | ITopenTExpQuote             --  [||
  | ITcloseTExpQuote            --  ||]
  | ITidEscape   FastString     --  $x
  | ITparenEscape               --  $(
  | ITidTyEscape   FastString   --  $$x
  | ITparenTyEscape             --  $$(
  | ITtyQuote                   --  ''
  | ITquasiQuote (FastString,FastString,RealSrcSpan)
    -- ITquasiQuote(quoter, quote, loc)
    -- represents a quasi-quote of the form
    -- [quoter| quote |]
  | ITqQuasiQuote (FastString,FastString,FastString,RealSrcSpan)
    -- ITqQuasiQuote(Qual, quoter, quote, loc)
    -- represents a qualified quasi-quote of the form
    -- [Qual.quoter| quote |]

  -- Arrow notation extension
  | ITproc
  | ITrec
  | IToparenbar                 --  (|
  | ITcparenbar                 --  |)
  | ITlarrowtail                --  -<
  | ITrarrowtail                --  >-
  | ITLarrowtail                --  -<<
  | ITRarrowtail                --  >>-

  | ITunknown String            -- Used when the lexer can't make sense of it
  | ITeof                       -- end of file token

  -- Documentation annotations
  | ITdocCommentNext  String     -- something beginning '-- |'
  | ITdocCommentPrev  String     -- something beginning '-- ^'
  | ITdocCommentNamed String     -- something beginning '-- $'
  | ITdocSection      Int String -- a section heading
  | ITdocOptions      String     -- doc options (prune, ignore-exports, etc)
  | ITdocOptionsOld   String     -- doc options declared "-- # ..."-style
  | ITlineComment     String     -- comment starting by "--"
  | ITblockComment    String     -- comment in {- -}

  deriving Show

instance Outputable Token where
  ppr x = text (show x)

-- the bitmap provided as the third component indicates whether the
-- corresponding extension keyword is valid under the extension options
-- provided to the compiler; if the extension corresponding to *any* of the
-- bits set in the bitmap is enabled, the keyword is valid (this setup
-- facilitates using a keyword in two different extensions that can be
-- activated independently)
--
reservedWordsFM :: UniqFM (Token, ExtsBitmap)
reservedWordsFM = listToUFM $
    map (\(x, y, z) -> (mkFastString x, (y, z)))
        [( "_",              ITunderscore,    0 ),
         ( "as",             ITas,            0 ),
         ( "case",           ITcase,          0 ),
         ( "class",          ITclass,         0 ),
         ( "data",           ITdata,          0 ),
         ( "default",        ITdefault,       0 ),
         ( "deriving",       ITderiving,      0 ),
         ( "do",             ITdo,            0 ),
         ( "else",           ITelse,          0 ),
         ( "hiding",         IThiding,        0 ),
         ( "if",             ITif,            0 ),
         ( "import",         ITimport,        0 ),
         ( "in",             ITin,            0 ),
         ( "infix",          ITinfix,         0 ),
         ( "infixl",         ITinfixl,        0 ),
         ( "infixr",         ITinfixr,        0 ),
         ( "instance",       ITinstance,      0 ),
         ( "let",            ITlet,           0 ),
         ( "module",         ITmodule,        0 ),
         ( "newtype",        ITnewtype,       0 ),
         ( "of",             ITof,            0 ),
         ( "qualified",      ITqualified,     0 ),
         ( "then",           ITthen,          0 ),
         ( "type",           ITtype,          0 ),
         ( "where",          ITwhere,         0 ),

         ( "forall",         ITforall,        xbit ExplicitForallBit .|.
                                              xbit InRulePragBit),
         ( "mdo",            ITmdo,           xbit RecursiveDoBit),
             -- See Note [Lexing type pseudo-keywords]
         ( "family",         ITfamily,        0 ),
         ( "role",           ITrole,          0 ),
         ( "pattern",        ITpattern,       xbit PatternSynonymsBit),
         ( "group",          ITgroup,         xbit TransformComprehensionsBit),
         ( "by",             ITby,            xbit TransformComprehensionsBit),
         ( "using",          ITusing,         xbit TransformComprehensionsBit),

         ( "foreign",        ITforeign,       xbit FfiBit),
         ( "export",         ITexport,        xbit FfiBit),
         ( "label",          ITlabel,         xbit FfiBit),
         ( "dynamic",        ITdynamic,       xbit FfiBit),
         ( "safe",           ITsafe,          xbit FfiBit .|.
                                              xbit SafeHaskellBit),
         ( "interruptible",  ITinterruptible, xbit InterruptibleFfiBit),
         ( "unsafe",         ITunsafe,        xbit FfiBit),
         ( "stdcall",        ITstdcallconv,   xbit FfiBit),
         ( "ccall",          ITccallconv,     xbit FfiBit),
         ( "capi",           ITcapiconv,      xbit CApiFfiBit),
         ( "prim",           ITprimcallconv,  xbit FfiBit),
         ( "javascript",     ITjavascriptcallconv, xbit FfiBit),

         ( "rec",            ITrec,           xbit ArrowsBit .|.
                                              xbit RecursiveDoBit),
         ( "proc",           ITproc,          xbit ArrowsBit)
     ]

{-----------------------------------
Note [Lexing type pseudo-keywords]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

One might think that we wish to treat 'family' and 'role' as regular old
varids whenever -XTypeFamilies and -XRoleAnnotations are off, respectively.
But, there is no need to do so. These pseudo-keywords are not stolen syntax:
they are only used after the keyword 'type' at the top-level, where varids are
not allowed. Furthermore, checks further downstream (TcTyClsDecls) ensure that
type families and role annotations are never declared without their extensions
on. In fact, by unconditionally lexing these pseudo-keywords as special, we
can get better error messages.

Also, note that these are included in the `varid` production in the parser --
a key detail to make all this work.
-------------------------------------}

reservedSymsFM :: UniqFM (Token, ExtsBitmap -> Bool)
reservedSymsFM = listToUFM $
    map (\ (x,y,z) -> (mkFastString x,(y,z)))
      [ ("..",  ITdotdot,   always)
        -- (:) is a reserved op, meaning only list cons
       ,(":",   ITcolon,    always)
       ,("::",  ITdcolon,   always)
       ,("=",   ITequal,    always)
       ,("\\",  ITlam,      always)
       ,("|",   ITvbar,     always)
       ,("<-",  ITlarrow,   always)
       ,("->",  ITrarrow,   always)
       ,("@",   ITat,       always)
       ,("~",   ITtilde,    always)
       ,("~#",  ITtildehsh, magicHashEnabled)
       ,("=>",  ITdarrow,   always)
       ,("-",   ITminus,    always)
       ,("!",   ITbang,     always)

        -- For data T (a::*) = MkT
       ,("*", ITstar, always) -- \i -> kindSigsEnabled i || tyFamEnabled i)
        -- For 'forall a . t'
       ,(".", ITdot,  always) -- \i -> explicitForallEnabled i || inRulePrag i)

       ,("-<",  ITlarrowtail, arrowsEnabled)
       ,(">-",  ITrarrowtail, arrowsEnabled)
       ,("-<<", ITLarrowtail, arrowsEnabled)
       ,(">>-", ITRarrowtail, arrowsEnabled)

       ,("∷",   ITdcolon, unicodeSyntaxEnabled)
       ,("⇒",   ITdarrow, unicodeSyntaxEnabled)
       ,("∀",   ITforall, unicodeSyntaxEnabled)
       ,("→",   ITrarrow, unicodeSyntaxEnabled)
       ,("←",   ITlarrow, unicodeSyntaxEnabled)

       ,("⤙",   ITlarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
       ,("⤚",   ITrarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
       ,("⤛",   ITLarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
       ,("⤜",   ITRarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)

       ,("★", ITstar, unicodeSyntaxEnabled)

        -- ToDo: ideally, → and ∷ should be "specials", so that they cannot
        -- form part of a large operator.  This would let us have a better
        -- syntax for kinds: ɑ∷*→* would be a legal kind signature. (maybe).
       ]

-- -----------------------------------------------------------------------------
-- Lexer actions

type Action = RealSrcSpan -> StringBuffer -> Int -> P (RealLocated Token)

special :: Token -> Action
special tok span _buf _len = return (L span tok)

token, layout_token :: Token -> Action
token t span _buf _len = return (L span t)
layout_token t span _buf _len = pushLexState layout >> return (L span t)

idtoken :: (StringBuffer -> Int -> Token) -> Action
idtoken f span buf len = return (L span $! (f buf len))

skip_one_varid :: (FastString -> Token) -> Action
skip_one_varid f span buf len
  = return (L span $! f (lexemeToFastString (stepOn buf) (len-1)))

skip_two_varid :: (FastString -> Token) -> Action
skip_two_varid f span buf len
  = return (L span $! f (lexemeToFastString (stepOn (stepOn buf)) (len-2)))

strtoken :: (String -> Token) -> Action
strtoken f span buf len =
  return (L span $! (f $! lexemeToString buf len))

init_strtoken :: Int -> (String -> Token) -> Action
-- like strtoken, but drops the last N character(s)
init_strtoken drop f span buf len =
  return (L span $! (f $! lexemeToString buf (len-drop)))

begin :: Int -> Action
begin code _span _str _len = do pushLexState code; lexToken

pop :: Action
pop _span _buf _len = do _ <- popLexState
                         lexToken

hopefully_open_brace :: Action
hopefully_open_brace span buf len
 = do relaxed <- extension relaxedLayout
      ctx <- getContext
      (AI l _) <- getInput
      let offset = srcLocCol l
          isOK = relaxed ||
                 case ctx of
                 Layout prev_off : _ -> prev_off < offset
                 _                   -> True
      if isOK then pop_and open_brace span buf len
              else failSpanMsgP (RealSrcSpan span) (text "Missing block")

pop_and :: Action -> Action
pop_and act span buf len = do _ <- popLexState
                              act span buf len

{-# INLINE nextCharIs #-}
nextCharIs :: StringBuffer -> (Char -> Bool) -> Bool
nextCharIs buf p = not (atEnd buf) && p (currentChar buf)

{-# INLINE nextCharIsNot #-}
nextCharIsNot :: StringBuffer -> (Char -> Bool) -> Bool
nextCharIsNot buf p = not (nextCharIs buf p)

notFollowedBy :: Char -> AlexAccPred ExtsBitmap
notFollowedBy char _ _ _ (AI _ buf)
  = nextCharIsNot buf (== char)

notFollowedBySymbol :: AlexAccPred ExtsBitmap
notFollowedBySymbol _ _ _ (AI _ buf)
  = nextCharIsNot buf (`elem` "!#$%&*+./<=>?@\\^|-~")

-- We must reject doc comments as being ordinary comments everywhere.
-- In some cases the doc comment will be selected as the lexeme due to
-- maximal munch, but not always, because the nested comment rule is
-- valid in all states, but the doc-comment rules are only valid in
-- the non-layout states.
isNormalComment :: AlexAccPred ExtsBitmap
isNormalComment bits _ _ (AI _ buf)
  | haddockEnabled bits = notFollowedByDocOrPragma
  | otherwise           = nextCharIsNot buf (== '#')
  where
    notFollowedByDocOrPragma
       = afterOptionalSpace buf (\b -> nextCharIsNot b (`elem` "|^*$#"))

afterOptionalSpace :: StringBuffer -> (StringBuffer -> Bool) -> Bool
afterOptionalSpace buf p
    = if nextCharIs buf (== ' ')
      then p (snd (nextChar buf))
      else p buf

atEOL :: AlexAccPred ExtsBitmap
atEOL _ _ _ (AI _ buf) = atEnd buf || currentChar buf == '\n'

ifExtension :: (ExtsBitmap -> Bool) -> AlexAccPred ExtsBitmap
ifExtension pred bits _ _ _ = pred bits

multiline_doc_comment :: Action
multiline_doc_comment span buf _len = withLexedDocType (worker "")
  where
    worker commentAcc input docType oneLine = case alexGetChar' input of
      Just ('\n', input')
        | oneLine -> docCommentEnd input commentAcc docType buf span
        | otherwise -> case checkIfCommentLine input' of
          Just input -> worker ('\n':commentAcc) input docType False
          Nothing -> docCommentEnd input commentAcc docType buf span
      Just (c, input) -> worker (c:commentAcc) input docType oneLine
      Nothing -> docCommentEnd input commentAcc docType buf span

    checkIfCommentLine input = check (dropNonNewlineSpace input)
      where
        check input = case alexGetChar' input of
          Just ('-', input) -> case alexGetChar' input of
            Just ('-', input) -> case alexGetChar' input of
              Just (c, _) | c /= '-' -> Just input
              _ -> Nothing
            _ -> Nothing
          _ -> Nothing

        dropNonNewlineSpace input = case alexGetChar' input of
          Just (c, input')
            | isSpace c && c /= '\n' -> dropNonNewlineSpace input'
            | otherwise -> input
          Nothing -> input

lineCommentToken :: Action
lineCommentToken span buf len = do
  b <- extension rawTokenStreamEnabled
  if b then strtoken ITlineComment span buf len else lexToken

{-
  nested comments require traversing by hand, they can't be parsed
  using regular expressions.
-}
nested_comment :: P (RealLocated Token) -> Action
nested_comment cont span buf len = do
  input <- getInput
  go (reverse $ drop 2 $ lexemeToString buf len) (1::Int) input
  where
    go commentAcc 0 input = do
      setInput input
      b <- extension rawTokenStreamEnabled
      if b
        then docCommentEnd input commentAcc ITblockComment buf span
        else cont
    go commentAcc n input = case alexGetChar' input of
      Nothing -> errBrace input span
      Just ('-',input) -> case alexGetChar' input of
        Nothing  -> errBrace input span
        Just ('\125',input) -> go commentAcc (n-1) input
        Just (_,_)          -> go ('-':commentAcc) n input
      Just ('\123',input) -> case alexGetChar' input of
        Nothing  -> errBrace input span
        Just ('-',input) -> go ('-':'\123':commentAcc) (n+1) input
        Just (_,_)       -> go ('\123':commentAcc) n input
      Just (c,input) -> go (c:commentAcc) n input

nested_doc_comment :: Action
nested_doc_comment span buf _len = withLexedDocType (go "")
  where
    go commentAcc input docType _ = case alexGetChar' input of
      Nothing -> errBrace input span
      Just ('-',input) -> case alexGetChar' input of
        Nothing -> errBrace input span
        Just ('\125',input) ->
          docCommentEnd input commentAcc docType buf span
        Just (_,_) -> go ('-':commentAcc) input docType False
      Just ('\123', input) -> case alexGetChar' input of
        Nothing  -> errBrace input span
        Just ('-',input) -> do
          setInput input
          let cont = do input <- getInput; go commentAcc input docType False
          nested_comment cont span buf _len
        Just (_,_) -> go ('\123':commentAcc) input docType False
      Just (c,input) -> go (c:commentAcc) input docType False

withLexedDocType :: (AlexInput -> (String -> Token) -> Bool -> P (RealLocated Token))
                 -> P (RealLocated Token)
withLexedDocType lexDocComment = do
  input@(AI _ buf) <- getInput
  case prevChar buf ' ' of
    '|' -> lexDocComment input ITdocCommentNext False
    '^' -> lexDocComment input ITdocCommentPrev False
    '$' -> lexDocComment input ITdocCommentNamed False
    '*' -> lexDocSection 1 input
    '#' -> lexDocComment input ITdocOptionsOld False
    _ -> panic "withLexedDocType: Bad doc type"
 where
    lexDocSection n input = case alexGetChar' input of
      Just ('*', input) -> lexDocSection (n+1) input
      Just (_,   _)     -> lexDocComment input (ITdocSection n) True
      Nothing -> do setInput input; lexToken -- eof reached, lex it normally

-- RULES pragmas turn on the forall and '.' keywords, and we turn them
-- off again at the end of the pragma.
rulePrag :: Action