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"

{
{-# LANGUAGE BangPatterns #-}

-- See Note [Warnings in code generated by Alex] in compiler/parser/Lexer.x
{-# OPTIONS_GHC -fno-warn-unused-matches #-}
{-# OPTIONS_GHC -fno-warn-unused-binds #-}
{-# OPTIONS_GHC -fno-warn-unused-imports #-}
{-# OPTIONS_GHC -fno-warn-tabs #-}
{-# OPTIONS_GHC -fno-warn-missing-signatures #-}

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

module Lexer (
   Token(..), lexer, pragState, mkPState, mkPStatePure, PState(..),
   P(..), ParseResult(..), mkParserFlags, ParserFlags(..), getSrcLoc,
   getPState, extopt, withThisPackage,
   failLocMsgP, failSpanMsgP, srcParseFail,
   getMessages,
   popContext, pushModuleContext, setLastToken, setSrcLoc,
   activeContext, nextIsEOF,
   getLexState, popLexState, pushLexState,
   extension, bangPatEnabled, datatypeContextsEnabled,
   traditionalRecordSyntaxEnabled,
   explicitForallEnabled,
   inRulePrag,
   explicitNamespacesEnabled,
   patternSynonymsEnabled,
   sccProfilingOn, hpcEnabled,
   starIsTypeEnabled,
   addWarning,
   lexTokenStream,
   addAnnotation,AddAnn,addAnnsAt,mkParensApiAnn,
   commentToAnnotation,
   moveAnnotations
  ) where

import GhcPrelude

-- base
import Control.Monad
import Control.Monad.Fail as MonadFail
import Data.Bits
import Data.Char
import Data.List
import Data.Maybe
import Data.Word

import EnumSet (EnumSet)
import qualified EnumSet

-- ghc-boot
import qualified GHC.LanguageExtensions as LangExt

-- bytestring
import Data.ByteString (ByteString)

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

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

-- compiler/main
import ErrUtils
import DynFlags

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

-- 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 [Unicode in Alex].
$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 [Unicode in Alex].
$decdigit  = $ascdigit -- for now, should really be $digit (ToDo)
$digit     = [$ascdigit $unidigit]

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

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

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

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

$binit     = 0-1
$octit     = 0-7
$hexit     = [$decdigit A-F a-f]

$uniidchar = \x07 -- Trick Alex into handling Unicode. See [Unicode in Alex].
$idchar    = [$small $large $digit $uniidchar \']

$pragmachar = [$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

-- See Note [Lexing NumericUnderscores extension] and #14473
@numspc       = _*                   -- numeric spacer (#14473)
@decimal      = $decdigit(@numspc $decdigit)*
@binary       = $binit(@numspc $binit)*
@octal        = $octit(@numspc $octit)*
@hexadecimal  = $hexit(@numspc $hexit)*
@exponent     = @numspc [eE] [\-\+]? @decimal
@bin_exponent = @numspc [pP] [\-\+]? @decimal

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

@floating_point = @numspc @decimal \. @decimal @exponent? | @numspc @decimal @exponent
@hex_floating_point = @numspc @hexadecimal \. @hexadecimal @bin_exponent? | @numspc @hexadecimal @bin_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          { warnTab }

-- 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 }
  ^\# / { followedByDigit }             { 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 dontGenerateSemic ITvbar }
  ()                                    { pop }
}

-- do is treated in a subtly different way, see new_layout_context
<layout>    ()                          { new_layout_context True  generateSemic ITvocurly }
<layout_do> ()                          { new_layout_context False generateSemic 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 $white_no_nl+ \" [$graphic \ ]* \"  { setLineAndFile line_prag1a }
  ()                                           { failLinePrag1 }
}
<line_prag1a> .*                               { popLinePrag1 }

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

-- Haskell-style column pragmas, of the form
--    {-# COLUMN <column> #-}
<column_prag> @decimal $whitechar* "#-}" { setColumn }

<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 }
}

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

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

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

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

  $unigraphic -- ⟦
    / { ifCurrentChar '⟦' `alexAndPred`
        ifExtension (\i -> unicodeSyntaxEnabled i && thQuotesEnabled i) }
    { token (ITopenExpQuote NoE UnicodeSyntax) }
  $unigraphic -- ⟧
    / { ifCurrentChar '⟧' `alexAndPred`
        ifExtension (\i -> unicodeSyntaxEnabled i && thQuotesEnabled i) }
    { token (ITcloseQuote UnicodeSyntax) }
}

  -- See Note [Lexing type applications]
<0> {
    [^ $idchar \) ] ^
  "@"
    / { ifExtension typeApplicationEnabled `alexAndPred` notFollowedBySymbol }
    { token ITtypeApp }
}

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

  $unigraphic -- ⦇
    / { ifCurrentChar '⦇' `alexAndPred`
        ifExtension (\i -> unicodeSyntaxEnabled i && arrowsEnabled i) }
    { special (IToparenbar UnicodeSyntax) }
  $unigraphic -- ⦈
    / { ifCurrentChar '⦈' `alexAndPred`
        ifExtension (\i -> unicodeSyntaxEnabled i && arrowsEnabled i) }
    { special (ITcparenbar UnicodeSyntax) }
}

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

<0> {
  "#" @varid / { ifExtension overloadedLabelsEnabled }
               { skip_one_varid ITlabelvarid }
}

<0> {
  "(#" / { orExtensions unboxedTuplesEnabled unboxedSumsEnabled }
         { token IToubxparen }
  "#)" / { orExtensions unboxedTuplesEnabled unboxedSumsEnabled }
         { 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

-- Note [Lexing NumericUnderscores extension] (#14473)
--
-- NumericUnderscores extension allows underscores in numeric literals.
-- Multiple underscores are represented with @numspc macro.
-- To be simpler, we have only the definitions with underscores.
-- And then we have a separate function (tok_integral and tok_frac)
-- that validates the literals.
-- If extensions are not enabled, check that there are no underscores.
--
<0> {
  -- Normal integral literals (:: Num a => a, from Integer)
  @decimal                                                               { tok_num positive 0 0 decimal }
  0[bB] @numspc @binary        / { ifExtension binaryLiteralsEnabled }   { tok_num positive 2 2 binary }
  0[oO] @numspc @octal                                                   { tok_num positive 2 2 octal }
  0[xX] @numspc @hexadecimal                                             { tok_num positive 2 2 hexadecimal }
  @negative @decimal           / { ifExtension negativeLiteralsEnabled } { tok_num negative 1 1 decimal }
  @negative 0[bB] @numspc @binary  / { ifExtension negativeLiteralsEnabled `alexAndPred`
                                       ifExtension binaryLiteralsEnabled }   { tok_num negative 3 3 binary }
  @negative 0[oO] @numspc @octal   / { ifExtension negativeLiteralsEnabled } { tok_num negative 3 3 octal }
  @negative 0[xX] @numspc @hexadecimal / { ifExtension negativeLiteralsEnabled } { tok_num negative 3 3 hexadecimal }

  -- Normal rational literals (:: Fractional a => a, from Rational)
  @floating_point                                                        { tok_frac 0 tok_float }
  @negative @floating_point    / { ifExtension negativeLiteralsEnabled } { tok_frac 0 tok_float }
  0[xX] @numspc @hex_floating_point     / { ifExtension hexFloatLiteralsEnabled } { tok_frac 0 tok_hex_float }
  @negative 0[xX] @numspc @hex_floating_point / { ifExtension hexFloatLiteralsEnabled `alexAndPred`
                                                  ifExtension negativeLiteralsEnabled } { tok_frac 0 tok_hex_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] @numspc @binary        \# / { ifExtension magicHashEnabled `alexAndPred`
                                      ifExtension binaryLiteralsEnabled } { tok_primint positive 2 3 binary }
  0[oO] @numspc @octal         \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 octal }
  0[xX] @numspc @hexadecimal   \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 hexadecimal }
  @negative @decimal           \# / { ifExtension magicHashEnabled } { tok_primint negative 1 2 decimal }
  @negative 0[bB] @numspc @binary  \# / { ifExtension magicHashEnabled `alexAndPred`
                                          ifExtension binaryLiteralsEnabled } { tok_primint negative 3 4 binary }
  @negative 0[oO] @numspc @octal   \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 octal }
  @negative 0[xX] @numspc @hexadecimal \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 hexadecimal }

  @decimal                     \# \# / { ifExtension magicHashEnabled } { tok_primword 0 2 decimal }
  0[bB] @numspc @binary        \# \# / { ifExtension magicHashEnabled `alexAndPred`
                                         ifExtension binaryLiteralsEnabled } { tok_primword 2 4 binary }
  0[oO] @numspc @octal         \# \# / { ifExtension magicHashEnabled } { tok_primword 2 4 octal }
  0[xX] @numspc @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 } { tok_frac 1 tok_primfloat }
  @signed @floating_point \# \# / { ifExtension magicHashEnabled } { tok_frac 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 }
}

-- Note [Lexing type applications]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-- The desired syntax for type applications is to prefix the type application
-- with '@', like this:
--
--   foo @Int @Bool baz bum
--
-- This, of course, conflicts with as-patterns. The conflict arises because
-- expressions and patterns use the same parser, and also because we want
-- to allow type patterns within expression patterns.
--
-- Disambiguation is accomplished by requiring *something* to appear between
-- type application and the preceding token. This something must end with
-- a character that cannot be the end of the variable bound in an as-pattern.
-- Currently (June 2015), this means that the something cannot end with a
-- $idchar or a close-paren. (The close-paren is necessary if the as-bound
-- identifier is symbolic.)
--
-- Note that looking for whitespace before the '@' is insufficient, because
-- of this pathological case:
--
--   foo {- hi -}@Int
--
-- This design is predicated on the fact that as-patterns are generally
-- whitespace-free, and also that this whole thing is opt-in, with the
-- TypeApplications extension.

-- -----------------------------------------------------------------------------
-- 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            IsUnicodeSyntax -- GHC extension keywords
  | ITexport
  | ITlabel
  | ITdynamic
  | ITsafe
  | ITinterruptible
  | ITunsafe
  | ITstdcallconv
  | ITccallconv
  | ITcapiconv
  | ITprimcallconv
  | ITjavascriptcallconv
  | ITmdo
  | ITfamily
  | ITrole
  | ITgroup
  | ITby
  | ITusing
  | ITpattern
  | ITstatic
  | ITstock
  | ITanyclass
  | ITvia

  -- Backpack tokens
  | ITunit
  | ITsignature
  | ITdependency
  | ITrequires

  -- Pragmas, see  note [Pragma source text] in BasicTypes
  | ITinline_prag       SourceText InlineSpec RuleMatchInfo
  | ITspec_prag         SourceText                -- SPECIALISE
  | ITspec_inline_prag  SourceText Bool    -- SPECIALISE INLINE (or NOINLINE)
  | ITsource_prag       SourceText
  | ITrules_prag        SourceText
  | ITwarning_prag      SourceText
  | ITdeprecated_prag   SourceText
  | ITline_prag         SourceText  -- not usually produced, see 'use_pos_prags'
  | ITcolumn_prag       SourceText  -- not usually produced, see 'use_pos_prags'
  | ITscc_prag          SourceText
  | ITgenerated_prag    SourceText
  | ITcore_prag         SourceText         -- hdaume: core annotations
  | ITunpack_prag       SourceText
  | ITnounpack_prag     SourceText
  | ITann_prag          SourceText
  | ITcomplete_prag     SourceText
  | ITclose_prag
  | IToptions_prag String
  | ITinclude_prag String
  | ITlanguage_prag
  | ITminimal_prag      SourceText
  | IToverlappable_prag SourceText  -- instance overlap mode
  | IToverlapping_prag  SourceText  -- instance overlap mode
  | IToverlaps_prag     SourceText  -- instance overlap mode
  | ITincoherent_prag   SourceText  -- instance overlap mode
  | ITctype             SourceText
  | ITcomment_line_prag         -- See Note [Nested comment line pragmas]

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

  | ITdupipvarid   FastString   -- GHC extension: implicit param: ?x
  | ITlabelvarid   FastString   -- Overloaded label: #x

  | ITchar     SourceText Char       -- Note [Literal source text] in BasicTypes
  | ITstring   SourceText FastString -- Note [Literal source text] in BasicTypes
  | ITinteger  IntegralLit           -- Note [Literal source text] in BasicTypes
  | ITrational FractionalLit

  | ITprimchar   SourceText Char     -- Note [Literal source text] in BasicTypes
  | ITprimstring SourceText ByteString -- Note [Literal source text] @BasicTypes
  | ITprimint    SourceText Integer  -- Note [Literal source text] in BasicTypes
  | ITprimword   SourceText Integer  -- Note [Literal source text] in BasicTypes
  | ITprimfloat  FractionalLit
  | ITprimdouble FractionalLit

  -- Template Haskell extension tokens
  | ITopenExpQuote HasE IsUnicodeSyntax --  [| or [e|
  | ITopenPatQuote                      --  [p|
  | ITopenDecQuote                      --  [d|
  | ITopenTypQuote                      --  [t|
  | ITcloseQuote IsUnicodeSyntax        --  |]
  | ITopenTExpQuote HasE                --  [|| or [e||
  | 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  IsUnicodeSyntax --  (|
  | ITcparenbar  IsUnicodeSyntax --  |)
  | ITlarrowtail IsUnicodeSyntax --  -<
  | ITrarrowtail IsUnicodeSyntax --  >-
  | ITLarrowtail IsUnicodeSyntax --  -<<
  | ITRarrowtail IsUnicodeSyntax --  >>-

  -- type application '@' (lexed differently than as-pattern '@',
  -- due to checking for preceding whitespace)
  | ITtypeApp


  | 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)
  | 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 NormalSyntax,
                                              xbit ExplicitForallBit .|.
                                              xbit InRulePragBit),
         ( "mdo",            ITmdo,           xbit RecursiveDoBit),
             -- See Note [Lexing type pseudo-keywords]
         ( "family",         ITfamily,        0 ),
         ( "role",           ITrole,          0 ),
         ( "pattern",        ITpattern,       xbit PatternSynonymsBit),
         ( "static",         ITstatic,        0 ),
         ( "stock",          ITstock,         0 ),
         ( "anyclass",       ITanyclass,      0 ),
         ( "via",            ITvia,           0 ),
         ( "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),

         ( "unit",           ITunit,          0 ),
         ( "dependency",     ITdependency,       0 ),
         ( "signature",      ITsignature,     0 ),

         ( "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 NormalSyntax, always)
       ,("=",   ITequal,               always)
       ,("\\",  ITlam,                 always)
       ,("|",   ITvbar,                always)
       ,("<-",  ITlarrow NormalSyntax, always)
       ,("->",  ITrarrow NormalSyntax, always)
       ,("@",   ITat,                  always)
       ,("~",   ITtilde,               always)
       ,("=>",  ITdarrow NormalSyntax, always)
       ,("-",   ITminus,               always)
       ,("!",   ITbang,                always)

       ,("*", ITstar NormalSyntax, starIsTypeEnabled)

        -- For 'forall a . t'
       ,(".", ITdot,  always) -- \i -> explicitForallEnabled i || inRulePrag i)

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

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

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

        -- 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))

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

pop :: Action
pop _span _buf _len = do _ <- popLexState
                         lexToken
-- See Note [Nested comment line pragmas]
failLinePrag1 :: Action
failLinePrag1 span _buf _len = do
  b <- extension inNestedComment
  if b then return (L span ITcomment_line_prag)
       else lexError "lexical error in pragma"

-- See Note [Nested comment line pragmas]
popLinePrag1 :: Action
popLinePrag1 span _buf _len = do
  b <- extension inNestedComment
  if b then return (L span ITcomment_line_prag) else 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` "!#$%&*+./<=>?@\\^|-~")

followedByDigit :: AlexAccPred ExtsBitmap
followedByDigit _ _ _ (AI _ buf)
  = afterOptionalSpace buf (\b -> nextCharIs b (`elem` ['0'..'9']))

ifCurrentChar :: Char -> AlexAccPred ExtsBitmap
ifCurrentChar char _ (AI _ buf) _ _
  = nextCharIs buf (== char)

-- 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

orExtensions :: (ExtsBitmap -> Bool) -> (ExtsBitmap -> Bool) -> AlexAccPred ExtsBitmap
orExtensions pred1 pred2 bits _ _ _ = pred1 bits || pred2 bits

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