1 -----------------------------------------------------------------------------
2 -- (c) The University of Glasgow, 2006
6 -- This is a combination of an Alex-generated lexer from a regex
7 -- definition, with some hand-coded bits.
9 -- Completely accurate information about token-spans within the source
10 -- file is maintained. Every token has a start and end SrcLoc attached to it.
12 -----------------------------------------------------------------------------
15 -- - parsing integers is a bit slow
16 -- - readRational is a bit slow
18 -- Known bugs, that were also in the previous version:
19 -- - M... should be 3 tokens, not 1.
20 -- - pragma-end should be only valid in a pragma
22 -- qualified operator NOTES.
24 -- - If M.(+) is a single lexeme, then..
25 -- - Probably (+) should be a single lexeme too, for consistency.
26 -- Otherwise ( + ) would be a prefix operator, but M.( + ) would not be.
27 -- - But we have to rule out reserved operators, otherwise (..) becomes
28 -- a different lexeme.
29 -- - Should we therefore also rule out reserved operators in the qualified
30 -- form? This is quite difficult to achieve. We don't do it for
34 -- XXX The above flags turn off warnings in the generated code:
35 {-# OPTIONS_GHC -fno-warn-unused-matches #-}
36 {-# OPTIONS_GHC -fno-warn-unused-binds #-}
37 {-# OPTIONS_GHC -fno-warn-unused-imports #-}
38 {-# OPTIONS_GHC -fno-warn-missing-signatures #-}
39 -- But alex still generates some code that causes the "lazy unlifted bindings"
40 -- warning, and old compilers don't know about it so we can't easily turn
41 -- it off, so for now we use the sledge hammer:
42 {-# OPTIONS_GHC -w #-}
44 {-# OPTIONS_GHC -funbox-strict-fields #-}
47 Token(..), lexer, pragState, mkPState, PState(..),
48 P(..), ParseResult(..), getSrcLoc,
49 getPState, getDynFlags, withThisPackage,
50 failLocMsgP, failSpanMsgP, srcParseFail,
52 popContext, pushCurrentContext, setLastToken, setSrcLoc,
53 getLexState, popLexState, pushLexState,
54 extension, standaloneDerivingEnabled, bangPatEnabled,
69 import Util ( readRational )
77 import qualified Data.Map as Map
81 $unispace = \x05 -- Trick Alex into handling Unicode. See alexGetChar.
82 $whitechar = [\ \n\r\f\v $unispace]
83 $white_no_nl = $whitechar # \n
87 $unidigit = \x03 -- Trick Alex into handling Unicode. See alexGetChar.
88 $decdigit = $ascdigit -- for now, should really be $digit (ToDo)
89 $digit = [$ascdigit $unidigit]
91 $special = [\(\)\,\;\[\]\`\{\}]
92 $ascsymbol = [\!\#\$\%\&\*\+\.\/\<\=\>\?\@\\\^\|\-\~]
93 $unisymbol = \x04 -- Trick Alex into handling Unicode. See alexGetChar.
94 $symbol = [$ascsymbol $unisymbol] # [$special \_\:\"\']
96 $unilarge = \x01 -- Trick Alex into handling Unicode. See alexGetChar.
98 $large = [$asclarge $unilarge]
100 $unismall = \x02 -- Trick Alex into handling Unicode. See alexGetChar.
102 $small = [$ascsmall $unismall \_]
104 $unigraphic = \x06 -- Trick Alex into handling Unicode. See alexGetChar.
105 $graphic = [$small $large $symbol $digit $special $unigraphic \:\"\']
108 $hexit = [$decdigit A-F a-f]
109 $symchar = [$symbol \:]
111 $idchar = [$small $large $digit \']
113 $pragmachar = [$small $large $digit]
115 $docsym = [\| \^ \* \$]
117 @varid = $small $idchar*
118 @conid = $large $idchar*
120 @varsym = $symbol $symchar*
121 @consym = \: $symchar*
123 @decimal = $decdigit+
125 @hexadecimal = $hexit+
126 @exponent = [eE] [\-\+]? @decimal
128 -- we support the hierarchical module name extension:
131 @floating_point = @decimal \. @decimal @exponent? | @decimal @exponent
133 -- normal signed numerical literals can only be explicitly negative,
134 -- not explicitly positive (contrast @exponent)
136 @signed = @negative ?
140 -- everywhere: skip whitespace and comments
142 $tab+ { warn Opt_WarnTabs (text "Tab character") }
144 -- Everywhere: deal with nested comments. We explicitly rule out
145 -- pragmas, "{-#", so that we don't accidentally treat them as comments.
146 -- (this can happen even though pragmas will normally take precedence due to
147 -- longest-match, because pragmas aren't valid in every state, but comments
148 -- are). We also rule out nested Haddock comments, if the -haddock flag is
151 "{-" / { isNormalComment } { nested_comment lexToken }
153 -- Single-line comments are a bit tricky. Haskell 98 says that two or
154 -- more dashes followed by a symbol should be parsed as a varsym, so we
155 -- have to exclude those.
157 -- Since Haddock comments aren't valid in every state, we need to rule them
160 -- The following two rules match comments that begin with two dashes, but
161 -- continue with a different character. The rules test that this character
162 -- is not a symbol (in which case we'd have a varsym), and that it's not a
163 -- space followed by a Haddock comment symbol (docsym) (in which case we'd
164 -- have a Haddock comment). The rules then munch the rest of the line.
166 "-- " ~[$docsym \#] .* { lineCommentToken }
167 "--" [^$symbol : \ ] .* { lineCommentToken }
169 -- Next, match Haddock comments if no -haddock flag
171 "-- " [$docsym \#] .* / { ifExtension (not . haddockEnabled) } { lineCommentToken }
173 -- Now, when we've matched comments that begin with 2 dashes and continue
174 -- with a different character, we need to match comments that begin with three
175 -- or more dashes (which clearly can't be Haddock comments). We only need to
176 -- make sure that the first non-dash character isn't a symbol, and munch the
179 "---"\-* [^$symbol :] .* { lineCommentToken }
181 -- Since the previous rules all match dashes followed by at least one
182 -- character, we also need to match a whole line filled with just dashes.
184 "--"\-* / { atEOL } { lineCommentToken }
186 -- We need this rule since none of the other single line comment rules
187 -- actually match this case.
189 "-- " / { atEOL } { lineCommentToken }
191 -- 'bol' state: beginning of a line. Slurp up all the whitespace (including
192 -- blank lines) until we find a non-whitespace character, then do layout
195 -- One slight wibble here: what if the line begins with {-#? In
196 -- theory, we have to lex the pragma to see if it's one we recognise,
197 -- and if it is, then we backtrack and do_bol, otherwise we treat it
198 -- as a nested comment. We don't bother with this: if the line begins
199 -- with {-#, then we'll assume it's a pragma we know about and go for do_bol.
202 ^\# (line)? { begin line_prag1 }
203 ^\# pragma .* \n ; -- GCC 3.3 CPP generated, apparently
204 ^\# \! .* \n ; -- #!, for scripts
208 -- after a layout keyword (let, where, do, of), we begin a new layout
209 -- context if the curly brace is missing.
210 -- Careful! This stuff is quite delicate.
211 <layout, layout_do> {
212 \{ / { notFollowedBy '-' } { pop_and open_brace }
213 -- we might encounter {-# here, but {- has been handled already
215 ^\# (line)? { begin line_prag1 }
218 -- do is treated in a subtly different way, see new_layout_context
219 <layout> () { new_layout_context True }
220 <layout_do> () { new_layout_context False }
222 -- after a new layout context which was found to be to the left of the
223 -- previous context, we have generated a '{' token, and we now need to
224 -- generate a matching '}' token.
225 <layout_left> () { do_layout_left }
227 <0,option_prags> \n { begin bol }
229 "{-#" $whitechar* $pragmachar+ / { known_pragma linePrags }
230 { dispatch_pragmas linePrags }
232 -- single-line line pragmas, of the form
233 -- # <line> "<file>" <extra-stuff> \n
234 <line_prag1> $decdigit+ { setLine line_prag1a }
235 <line_prag1a> \" [$graphic \ ]* \" { setFile line_prag1b }
236 <line_prag1b> .* { pop }
238 -- Haskell-style line pragmas, of the form
239 -- {-# LINE <line> "<file>" #-}
240 <line_prag2> $decdigit+ { setLine line_prag2a }
241 <line_prag2a> \" [$graphic \ ]* \" { setFile line_prag2b }
242 <line_prag2b> "#-}"|"-}" { pop }
243 -- NOTE: accept -} at the end of a LINE pragma, for compatibility
244 -- with older versions of GHC which generated these.
247 "{-#" $whitechar* $pragmachar+
248 $whitechar+ $pragmachar+ / { known_pragma twoWordPrags }
249 { dispatch_pragmas twoWordPrags }
251 "{-#" $whitechar* $pragmachar+ / { known_pragma oneWordPrags }
252 { dispatch_pragmas oneWordPrags }
254 -- We ignore all these pragmas, but don't generate a warning for them
255 "{-#" $whitechar* $pragmachar+ / { known_pragma ignoredPrags }
256 { dispatch_pragmas ignoredPrags }
258 -- ToDo: should only be valid inside a pragma:
263 "{-#" $whitechar* $pragmachar+ / { known_pragma fileHeaderPrags }
264 { dispatch_pragmas fileHeaderPrags }
266 "-- #" { multiline_doc_comment }
270 -- In the "0" mode we ignore these pragmas
271 "{-#" $whitechar* $pragmachar+ / { known_pragma fileHeaderPrags }
272 { nested_comment lexToken }
276 "-- #" .* { lineCommentToken }
280 "{-#" { warnThen Opt_WarnUnrecognisedPragmas (text "Unrecognised pragma")
281 (nested_comment lexToken) }
284 -- '0' state: ordinary lexemes
289 "-- " $docsym / { ifExtension haddockEnabled } { multiline_doc_comment }
290 "{-" \ ? $docsym / { ifExtension haddockEnabled } { nested_doc_comment }
296 "[:" / { ifExtension parrEnabled } { token ITopabrack }
297 ":]" / { ifExtension parrEnabled } { token ITcpabrack }
301 "[|" / { ifExtension thEnabled } { token ITopenExpQuote }
302 "[e|" / { ifExtension thEnabled } { token ITopenExpQuote }
303 "[p|" / { ifExtension thEnabled } { token ITopenPatQuote }
304 "[d|" / { ifExtension thEnabled } { layout_token ITopenDecQuote }
305 "[t|" / { ifExtension thEnabled } { token ITopenTypQuote }
306 "|]" / { ifExtension thEnabled } { token ITcloseQuote }
307 \$ @varid / { ifExtension thEnabled } { skip_one_varid ITidEscape }
308 "$(" / { ifExtension thEnabled } { token ITparenEscape }
310 "[$" @varid "|" / { ifExtension qqEnabled }
311 { lex_quasiquote_tok }
315 "(|" / { ifExtension arrowsEnabled `alexAndPred` notFollowedBySymbol }
316 { special IToparenbar }
317 "|)" / { ifExtension arrowsEnabled } { special ITcparenbar }
321 \? @varid / { ifExtension ipEnabled } { skip_one_varid ITdupipvarid }
325 "(#" / { ifExtension unboxedTuplesEnabled `alexAndPred` notFollowedBySymbol }
326 { token IToubxparen }
327 "#)" / { ifExtension unboxedTuplesEnabled }
328 { token ITcubxparen }
332 "{|" / { ifExtension genericsEnabled } { token ITocurlybar }
333 "|}" / { ifExtension genericsEnabled } { token ITccurlybar }
337 \( { special IToparen }
338 \) { special ITcparen }
339 \[ { special ITobrack }
340 \] { special ITcbrack }
341 \, { special ITcomma }
342 \; { special ITsemi }
343 \` { special ITbackquote }
350 @qual @varid { idtoken qvarid }
351 @qual @conid { idtoken qconid }
353 @conid { idtoken conid }
357 @qual @varid "#"+ / { ifExtension magicHashEnabled } { idtoken qvarid }
358 @qual @conid "#"+ / { ifExtension magicHashEnabled } { idtoken qconid }
359 @varid "#"+ / { ifExtension magicHashEnabled } { varid }
360 @conid "#"+ / { ifExtension magicHashEnabled } { idtoken conid }
363 -- ToDo: - move `var` and (sym) into lexical syntax?
364 -- - remove backquote from $special?
366 @qual @varsym / { ifExtension oldQualOps } { idtoken qvarsym }
367 @qual @consym / { ifExtension oldQualOps } { idtoken qconsym }
368 @qual \( @varsym \) / { ifExtension newQualOps } { idtoken prefixqvarsym }
369 @qual \( @consym \) / { ifExtension newQualOps } { idtoken prefixqconsym }
374 -- For the normal boxed literals we need to be careful
375 -- when trying to be close to Haskell98
377 -- Normal integral literals (:: Num a => a, from Integer)
378 @decimal { tok_num positive 0 0 decimal }
379 0[oO] @octal { tok_num positive 2 2 octal }
380 0[xX] @hexadecimal { tok_num positive 2 2 hexadecimal }
382 -- Normal rational literals (:: Fractional a => a, from Rational)
383 @floating_point { strtoken tok_float }
387 -- Unboxed ints (:: Int#) and words (:: Word#)
388 -- It's simpler (and faster?) to give separate cases to the negatives,
389 -- especially considering octal/hexadecimal prefixes.
390 @decimal \# / { ifExtension magicHashEnabled } { tok_primint positive 0 1 decimal }
391 0[oO] @octal \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 octal }
392 0[xX] @hexadecimal \# / { ifExtension magicHashEnabled } { tok_primint positive 2 3 hexadecimal }
393 @negative @decimal \# / { ifExtension magicHashEnabled } { tok_primint negative 1 2 decimal }
394 @negative 0[oO] @octal \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 octal }
395 @negative 0[xX] @hexadecimal \# / { ifExtension magicHashEnabled } { tok_primint negative 3 4 hexadecimal }
397 @decimal \# \# / { ifExtension magicHashEnabled } { tok_primword 0 2 decimal }
398 0[oO] @octal \# \# / { ifExtension magicHashEnabled } { tok_primword 2 4 octal }
399 0[xX] @hexadecimal \# \# / { ifExtension magicHashEnabled } { tok_primword 2 4 hexadecimal }
401 -- Unboxed floats and doubles (:: Float#, :: Double#)
402 -- prim_{float,double} work with signed literals
403 @signed @floating_point \# / { ifExtension magicHashEnabled } { init_strtoken 1 tok_primfloat }
404 @signed @floating_point \# \# / { ifExtension magicHashEnabled } { init_strtoken 2 tok_primdouble }
407 -- Strings and chars are lexed by hand-written code. The reason is
408 -- that even if we recognise the string or char here in the regex
409 -- lexer, we would still have to parse the string afterward in order
410 -- to convert it to a String.
413 \" { lex_string_tok }
417 -- -----------------------------------------------------------------------------
421 = ITas -- Haskell keywords
445 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
447 | ITforall -- GHC extension keywords
465 | ITinline_prag Bool -- True <=> INLINE, False <=> NOINLINE
466 | ITinline_conlike_prag Bool -- same
467 | ITspec_prag -- SPECIALISE
468 | ITspec_inline_prag Bool -- SPECIALISE INLINE (or NOINLINE)
476 | ITcore_prag -- hdaume: core annotations
480 | IToptions_prag String
481 | ITinclude_prag String
484 | ITdotdot -- reserved symbols
500 | ITbiglam -- GHC-extension symbols
502 | ITocurly -- special symbols
504 | ITocurlybar -- {|, for type applications
505 | ITccurlybar -- |}, for type applications
509 | ITopabrack -- [:, for parallel arrays with -XParr
510 | ITcpabrack -- :], for parallel arrays with -XParr
521 | ITvarid FastString -- identifiers
523 | ITvarsym FastString
524 | ITconsym FastString
525 | ITqvarid (FastString,FastString)
526 | ITqconid (FastString,FastString)
527 | ITqvarsym (FastString,FastString)
528 | ITqconsym (FastString,FastString)
529 | ITprefixqvarsym (FastString,FastString)
530 | ITprefixqconsym (FastString,FastString)
532 | ITdupipvarid FastString -- GHC extension: implicit param: ?x
535 | ITstring FastString
537 | ITrational Rational
540 | ITprimstring FastString
543 | ITprimfloat Rational
544 | ITprimdouble Rational
546 -- Template Haskell extension tokens
547 | ITopenExpQuote -- [| or [e|
548 | ITopenPatQuote -- [p|
549 | ITopenDecQuote -- [d|
550 | ITopenTypQuote -- [t|
552 | ITidEscape FastString -- $x
553 | ITparenEscape -- $(
556 | ITquasiQuote (FastString,FastString,SrcSpan) -- [:...|...|]
558 -- Arrow notation extension
565 | ITLarrowtail -- -<<
566 | ITRarrowtail -- >>-
568 | ITunknown String -- Used when the lexer can't make sense of it
569 | ITeof -- end of file token
571 -- Documentation annotations
572 | ITdocCommentNext String -- something beginning '-- |'
573 | ITdocCommentPrev String -- something beginning '-- ^'
574 | ITdocCommentNamed String -- something beginning '-- $'
575 | ITdocSection Int String -- a section heading
576 | ITdocOptions String -- doc options (prune, ignore-exports, etc)
577 | ITdocOptionsOld String -- doc options declared "-- # ..."-style
578 | ITlineComment String -- comment starting by "--"
579 | ITblockComment String -- comment in {- -}
582 deriving Show -- debugging
586 isSpecial :: Token -> Bool
587 -- If we see M.x, where x is a keyword, but
588 -- is special, we treat is as just plain M.x,
590 isSpecial ITas = True
591 isSpecial IThiding = True
592 isSpecial ITqualified = True
593 isSpecial ITforall = True
594 isSpecial ITexport = True
595 isSpecial ITlabel = True
596 isSpecial ITdynamic = True
597 isSpecial ITsafe = True
598 isSpecial ITthreadsafe = True
599 isSpecial ITunsafe = True
600 isSpecial ITccallconv = True
601 isSpecial ITstdcallconv = True
602 isSpecial ITprimcallconv = True
603 isSpecial ITmdo = True
604 isSpecial ITfamily = True
605 isSpecial ITgroup = True
606 isSpecial ITby = True
607 isSpecial ITusing = True
611 -- the bitmap provided as the third component indicates whether the
612 -- corresponding extension keyword is valid under the extension options
613 -- provided to the compiler; if the extension corresponding to *any* of the
614 -- bits set in the bitmap is enabled, the keyword is valid (this setup
615 -- facilitates using a keyword in two different extensions that can be
616 -- activated independently)
618 reservedWordsFM :: UniqFM (Token, Int)
619 reservedWordsFM = listToUFM $
620 map (\(x, y, z) -> (mkFastString x, (y, z)))
621 [( "_", ITunderscore, 0 ),
623 ( "case", ITcase, 0 ),
624 ( "class", ITclass, 0 ),
625 ( "data", ITdata, 0 ),
626 ( "default", ITdefault, 0 ),
627 ( "deriving", ITderiving, 0 ),
629 ( "else", ITelse, 0 ),
630 ( "hiding", IThiding, 0 ),
632 ( "import", ITimport, 0 ),
634 ( "infix", ITinfix, 0 ),
635 ( "infixl", ITinfixl, 0 ),
636 ( "infixr", ITinfixr, 0 ),
637 ( "instance", ITinstance, 0 ),
639 ( "module", ITmodule, 0 ),
640 ( "newtype", ITnewtype, 0 ),
642 ( "qualified", ITqualified, 0 ),
643 ( "then", ITthen, 0 ),
644 ( "type", ITtype, 0 ),
645 ( "where", ITwhere, 0 ),
646 ( "_scc_", ITscc, 0 ), -- ToDo: remove
648 ( "forall", ITforall, bit explicitForallBit .|. bit inRulePragBit),
649 ( "mdo", ITmdo, bit recursiveDoBit),
650 ( "family", ITfamily, bit tyFamBit),
651 ( "group", ITgroup, bit transformComprehensionsBit),
652 ( "by", ITby, bit transformComprehensionsBit),
653 ( "using", ITusing, bit transformComprehensionsBit),
655 ( "foreign", ITforeign, bit ffiBit),
656 ( "export", ITexport, bit ffiBit),
657 ( "label", ITlabel, bit ffiBit),
658 ( "dynamic", ITdynamic, bit ffiBit),
659 ( "safe", ITsafe, bit ffiBit),
660 ( "threadsafe", ITthreadsafe, bit ffiBit), -- ToDo: remove
661 ( "unsafe", ITunsafe, bit ffiBit),
662 ( "stdcall", ITstdcallconv, bit ffiBit),
663 ( "ccall", ITccallconv, bit ffiBit),
664 ( "prim", ITprimcallconv, bit ffiBit),
666 ( "rec", ITrec, bit recBit),
667 ( "proc", ITproc, bit arrowsBit)
670 reservedSymsFM :: UniqFM (Token, Int -> Bool)
671 reservedSymsFM = listToUFM $
672 map (\ (x,y,z) -> (mkFastString x,(y,z)))
673 [ ("..", ITdotdot, always)
674 -- (:) is a reserved op, meaning only list cons
675 ,(":", ITcolon, always)
676 ,("::", ITdcolon, always)
677 ,("=", ITequal, always)
678 ,("\\", ITlam, always)
679 ,("|", ITvbar, always)
680 ,("<-", ITlarrow, always)
681 ,("->", ITrarrow, always)
683 ,("~", ITtilde, always)
684 ,("=>", ITdarrow, always)
685 ,("-", ITminus, always)
686 ,("!", ITbang, always)
688 -- For data T (a::*) = MkT
689 ,("*", ITstar, always) -- \i -> kindSigsEnabled i || tyFamEnabled i)
690 -- For 'forall a . t'
691 ,(".", ITdot, always) -- \i -> explicitForallEnabled i || inRulePrag i)
693 ,("-<", ITlarrowtail, arrowsEnabled)
694 ,(">-", ITrarrowtail, arrowsEnabled)
695 ,("-<<", ITLarrowtail, arrowsEnabled)
696 ,(">>-", ITRarrowtail, arrowsEnabled)
698 ,("∷", ITdcolon, unicodeSyntaxEnabled)
699 ,("⇒", ITdarrow, unicodeSyntaxEnabled)
700 ,("∀", ITforall, \i -> unicodeSyntaxEnabled i &&
701 explicitForallEnabled i)
702 ,("→", ITrarrow, unicodeSyntaxEnabled)
703 ,("←", ITlarrow, unicodeSyntaxEnabled)
704 ,("⋯", ITdotdot, unicodeSyntaxEnabled)
706 ,("⤙", ITlarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
707 ,("⤚", ITrarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
708 ,("⤛", ITLarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
709 ,("⤜", ITRarrowtail, \i -> unicodeSyntaxEnabled i && arrowsEnabled i)
711 ,("★", ITstar, unicodeSyntaxEnabled)
713 -- ToDo: ideally, → and ∷ should be "specials", so that they cannot
714 -- form part of a large operator. This would let us have a better
715 -- syntax for kinds: ɑ∷*→* would be a legal kind signature. (maybe).
718 -- -----------------------------------------------------------------------------
721 type Action = SrcSpan -> StringBuffer -> Int -> P (Located Token)
723 special :: Token -> Action
724 special tok span _buf _len = return (L span tok)
726 token, layout_token :: Token -> Action
727 token t span _buf _len = return (L span t)
728 layout_token t span _buf _len = pushLexState layout >> return (L span t)
730 idtoken :: (StringBuffer -> Int -> Token) -> Action
731 idtoken f span buf len = return (L span $! (f buf len))
733 skip_one_varid :: (FastString -> Token) -> Action
734 skip_one_varid f span buf len
735 = return (L span $! f (lexemeToFastString (stepOn buf) (len-1)))
737 strtoken :: (String -> Token) -> Action
738 strtoken f span buf len =
739 return (L span $! (f $! lexemeToString buf len))
741 init_strtoken :: Int -> (String -> Token) -> Action
742 -- like strtoken, but drops the last N character(s)
743 init_strtoken drop f span buf len =
744 return (L span $! (f $! lexemeToString buf (len-drop)))
746 begin :: Int -> Action
747 begin code _span _str _len = do pushLexState code; lexToken
750 pop _span _buf _len = do _ <- popLexState
753 pop_and :: Action -> Action
754 pop_and act span buf len = do _ <- popLexState
757 {-# INLINE nextCharIs #-}
758 nextCharIs :: StringBuffer -> (Char -> Bool) -> Bool
759 nextCharIs buf p = not (atEnd buf) && p (currentChar buf)
761 notFollowedBy :: Char -> AlexAccPred Int
762 notFollowedBy char _ _ _ (AI _ buf)
763 = nextCharIs buf (/=char)
765 notFollowedBySymbol :: AlexAccPred Int
766 notFollowedBySymbol _ _ _ (AI _ buf)
767 = nextCharIs buf (`notElem` "!#$%&*+./<=>?@\\^|-~")
769 -- We must reject doc comments as being ordinary comments everywhere.
770 -- In some cases the doc comment will be selected as the lexeme due to
771 -- maximal munch, but not always, because the nested comment rule is
772 -- valid in all states, but the doc-comment rules are only valid in
773 -- the non-layout states.
774 isNormalComment :: AlexAccPred Int
775 isNormalComment bits _ _ (AI _ buf)
776 | haddockEnabled bits = notFollowedByDocOrPragma
777 | otherwise = nextCharIs buf (/='#')
779 notFollowedByDocOrPragma
780 = not $ spaceAndP buf (`nextCharIs` (`elem` "|^*$#"))
782 spaceAndP :: StringBuffer -> (StringBuffer -> Bool) -> Bool
783 spaceAndP buf p = p buf || nextCharIs buf (==' ') && p (snd (nextChar buf))
786 haddockDisabledAnd p bits _ _ (AI _ buf)
787 = if haddockEnabled bits then False else (p buf)
790 atEOL :: AlexAccPred Int
791 atEOL _ _ _ (AI _ buf) = atEnd buf || currentChar buf == '\n'
793 ifExtension :: (Int -> Bool) -> AlexAccPred Int
794 ifExtension pred bits _ _ _ = pred bits
796 multiline_doc_comment :: Action
797 multiline_doc_comment span buf _len = withLexedDocType (worker "")
799 worker commentAcc input docType oneLine = case alexGetChar input of
801 | oneLine -> docCommentEnd input commentAcc docType buf span
802 | otherwise -> case checkIfCommentLine input' of
803 Just input -> worker ('\n':commentAcc) input docType False
804 Nothing -> docCommentEnd input commentAcc docType buf span
805 Just (c, input) -> worker (c:commentAcc) input docType oneLine
806 Nothing -> docCommentEnd input commentAcc docType buf span
808 checkIfCommentLine input = check (dropNonNewlineSpace input)
810 check input = case alexGetChar input of
811 Just ('-', input) -> case alexGetChar input of
812 Just ('-', input) -> case alexGetChar input of
813 Just (c, _) | c /= '-' -> Just input
818 dropNonNewlineSpace input = case alexGetChar input of
820 | isSpace c && c /= '\n' -> dropNonNewlineSpace input'
824 lineCommentToken :: Action
825 lineCommentToken span buf len = do
826 b <- extension rawTokenStreamEnabled
827 if b then strtoken ITlineComment span buf len else lexToken
830 nested comments require traversing by hand, they can't be parsed
831 using regular expressions.
833 nested_comment :: P (Located Token) -> Action
834 nested_comment cont span _str _len = do
838 go commentAcc 0 input = do setInput input
839 b <- extension rawTokenStreamEnabled
841 then docCommentEnd input commentAcc ITblockComment _str span
843 go commentAcc n input = case alexGetChar input of
844 Nothing -> errBrace input span
845 Just ('-',input) -> case alexGetChar input of
846 Nothing -> errBrace input span
847 Just ('\125',input) -> go commentAcc (n-1) input
848 Just (_,_) -> go ('-':commentAcc) n input
849 Just ('\123',input) -> case alexGetChar input of
850 Nothing -> errBrace input span
851 Just ('-',input) -> go ('-':'\123':commentAcc) (n+1) input
852 Just (_,_) -> go ('\123':commentAcc) n input
853 Just (c,input) -> go (c:commentAcc) n input
855 nested_doc_comment :: Action
856 nested_doc_comment span buf _len = withLexedDocType (go "")
858 go commentAcc input docType _ = case alexGetChar input of
859 Nothing -> errBrace input span
860 Just ('-',input) -> case alexGetChar input of
861 Nothing -> errBrace input span
862 Just ('\125',input) ->
863 docCommentEnd input commentAcc docType buf span
864 Just (_,_) -> go ('-':commentAcc) input docType False
865 Just ('\123', input) -> case alexGetChar input of
866 Nothing -> errBrace input span
867 Just ('-',input) -> do
869 let cont = do input <- getInput; go commentAcc input docType False
870 nested_comment cont span buf _len
871 Just (_,_) -> go ('\123':commentAcc) input docType False
872 Just (c,input) -> go (c:commentAcc) input docType False
874 withLexedDocType :: (AlexInput -> (String -> Token) -> Bool -> P (Located Token))
876 withLexedDocType lexDocComment = do
877 input@(AI _ buf) <- getInput
878 case prevChar buf ' ' of
879 '|' -> lexDocComment input ITdocCommentNext False
880 '^' -> lexDocComment input ITdocCommentPrev False
881 '$' -> lexDocComment input ITdocCommentNamed False
882 '*' -> lexDocSection 1 input
883 '#' -> lexDocComment input ITdocOptionsOld False
884 _ -> panic "withLexedDocType: Bad doc type"
886 lexDocSection n input = case alexGetChar input of
887 Just ('*', input) -> lexDocSection (n+1) input
888 Just (_, _) -> lexDocComment input (ITdocSection n) True
889 Nothing -> do setInput input; lexToken -- eof reached, lex it normally
891 -- RULES pragmas turn on the forall and '.' keywords, and we turn them
892 -- off again at the end of the pragma.
894 rulePrag span _buf _len = do
895 setExts (.|. bit inRulePragBit)
896 return (L span ITrules_prag)
899 endPrag span _buf _len = do
900 setExts (.&. complement (bit inRulePragBit))
901 return (L span ITclose_prag)
904 -------------------------------------------------------------------------------
905 -- This function is quite tricky. We can't just return a new token, we also
906 -- need to update the state of the parser. Why? Because the token is longer
907 -- than what was lexed by Alex, and the lexToken function doesn't know this, so
908 -- it writes the wrong token length to the parser state. This function is
909 -- called afterwards, so it can just update the state.
911 docCommentEnd :: AlexInput -> String -> (String -> Token) -> StringBuffer ->
912 SrcSpan -> P (Located Token)
913 docCommentEnd input commentAcc docType buf span = do
915 let (AI loc nextBuf) = input
916 comment = reverse commentAcc
917 span' = mkSrcSpan (srcSpanStart span) loc
918 last_len = byteDiff buf nextBuf
920 span `seq` setLastToken span' last_len
921 return (L span' (docType comment))
923 errBrace :: AlexInput -> SrcSpan -> P a
924 errBrace (AI end _) span = failLocMsgP (srcSpanStart span) end "unterminated `{-'"
926 open_brace, close_brace :: Action
927 open_brace span _str _len = do
929 setContext (NoLayout:ctx)
930 return (L span ITocurly)
931 close_brace span _str _len = do
933 return (L span ITccurly)
935 qvarid, qconid :: StringBuffer -> Int -> Token
936 qvarid buf len = ITqvarid $! splitQualName buf len False
937 qconid buf len = ITqconid $! splitQualName buf len False
939 splitQualName :: StringBuffer -> Int -> Bool -> (FastString,FastString)
940 -- takes a StringBuffer and a length, and returns the module name
941 -- and identifier parts of a qualified name. Splits at the *last* dot,
942 -- because of hierarchical module names.
943 splitQualName orig_buf len parens = split orig_buf orig_buf
946 | orig_buf `byteDiff` buf >= len = done dot_buf
947 | c == '.' = found_dot buf'
948 | otherwise = split buf' dot_buf
950 (c,buf') = nextChar buf
952 -- careful, we might get names like M....
953 -- so, if the character after the dot is not upper-case, this is
954 -- the end of the qualifier part.
955 found_dot buf -- buf points after the '.'
956 | isUpper c = split buf' buf
957 | otherwise = done buf
959 (c,buf') = nextChar buf
962 (lexemeToFastString orig_buf (qual_size - 1),
963 if parens -- Prelude.(+)
964 then lexemeToFastString (stepOn dot_buf) (len - qual_size - 2)
965 else lexemeToFastString dot_buf (len - qual_size))
967 qual_size = orig_buf `byteDiff` dot_buf
972 case lookupUFM reservedWordsFM fs of
973 Just (keyword,0) -> do
975 return (L span keyword)
976 Just (keyword,exts) -> do
977 b <- extension (\i -> exts .&. i /= 0)
978 if b then do maybe_layout keyword
979 return (L span keyword)
980 else return (L span (ITvarid fs))
981 _other -> return (L span (ITvarid fs))
983 fs = lexemeToFastString buf len
985 conid :: StringBuffer -> Int -> Token
986 conid buf len = ITconid fs
987 where fs = lexemeToFastString buf len
989 qvarsym, qconsym, prefixqvarsym, prefixqconsym :: StringBuffer -> Int -> Token
990 qvarsym buf len = ITqvarsym $! splitQualName buf len False
991 qconsym buf len = ITqconsym $! splitQualName buf len False
992 prefixqvarsym buf len = ITprefixqvarsym $! splitQualName buf len True
993 prefixqconsym buf len = ITprefixqconsym $! splitQualName buf len True
995 varsym, consym :: Action
996 varsym = sym ITvarsym
997 consym = sym ITconsym
999 sym :: (FastString -> Token) -> SrcSpan -> StringBuffer -> Int
1000 -> P (Located Token)
1001 sym con span buf len =
1002 case lookupUFM reservedSymsFM fs of
1003 Just (keyword,exts) -> do
1005 if b then return (L span keyword)
1006 else return (L span $! con fs)
1007 _other -> return (L span $! con fs)
1009 fs = lexemeToFastString buf len
1011 -- Variations on the integral numeric literal.
1012 tok_integral :: (Integer -> Token)
1013 -> (Integer -> Integer)
1014 -- -> (StringBuffer -> StringBuffer) -> (Int -> Int)
1016 -> (Integer, (Char->Int)) -> Action
1017 tok_integral itint transint transbuf translen (radix,char_to_int) span buf len =
1018 return $ L span $ itint $! transint $ parseUnsignedInteger
1019 (offsetBytes transbuf buf) (subtract translen len) radix char_to_int
1021 -- some conveniences for use with tok_integral
1022 tok_num :: (Integer -> Integer)
1024 -> (Integer, (Char->Int)) -> Action
1025 tok_num = tok_integral ITinteger
1026 tok_primint :: (Integer -> Integer)
1028 -> (Integer, (Char->Int)) -> Action
1029 tok_primint = tok_integral ITprimint
1030 tok_primword :: Int -> Int
1031 -> (Integer, (Char->Int)) -> Action
1032 tok_primword = tok_integral ITprimword positive
1033 positive, negative :: (Integer -> Integer)
1036 decimal, octal, hexadecimal :: (Integer, Char -> Int)
1037 decimal = (10,octDecDigit)
1038 octal = (8,octDecDigit)
1039 hexadecimal = (16,hexDigit)
1041 -- readRational can understand negative rationals, exponents, everything.
1042 tok_float, tok_primfloat, tok_primdouble :: String -> Token
1043 tok_float str = ITrational $! readRational str
1044 tok_primfloat str = ITprimfloat $! readRational str
1045 tok_primdouble str = ITprimdouble $! readRational str
1047 -- -----------------------------------------------------------------------------
1048 -- Layout processing
1050 -- we're at the first token on a line, insert layout tokens if necessary
1052 do_bol span _str _len = do
1056 --trace "layout: inserting '}'" $ do
1058 -- do NOT pop the lex state, we might have a ';' to insert
1059 return (L span ITvccurly)
1061 --trace "layout: inserting ';'" $ do
1063 return (L span ITsemi)
1068 -- certain keywords put us in the "layout" state, where we might
1069 -- add an opening curly brace.
1070 maybe_layout :: Token -> P ()
1071 maybe_layout t = do -- If the alternative layout rule is enabled then
1072 -- we never create an implicit layout context here.
1073 -- Layout is handled XXX instead.
1074 -- The code for closing implicit contexts, or
1075 -- inserting implicit semi-colons, is therefore
1076 -- irrelevant as it only applies in an implicit
1078 alr <- extension alternativeLayoutRule
1080 where f ITdo = pushLexState layout_do
1081 f ITmdo = pushLexState layout_do
1082 f ITof = pushLexState layout
1083 f ITlet = pushLexState layout
1084 f ITwhere = pushLexState layout
1085 f ITrec = pushLexState layout
1088 -- Pushing a new implicit layout context. If the indentation of the
1089 -- next token is not greater than the previous layout context, then
1090 -- Haskell 98 says that the new layout context should be empty; that is
1091 -- the lexer must generate {}.
1093 -- We are slightly more lenient than this: when the new context is started
1094 -- by a 'do', then we allow the new context to be at the same indentation as
1095 -- the previous context. This is what the 'strict' argument is for.
1097 new_layout_context :: Bool -> Action
1098 new_layout_context strict span _buf _len = do
1100 (AI l _) <- getInput
1101 let offset = srcLocCol l
1104 Layout prev_off : _ |
1105 (strict && prev_off >= offset ||
1106 not strict && prev_off > offset) -> do
1107 -- token is indented to the left of the previous context.
1108 -- we must generate a {} sequence now.
1109 pushLexState layout_left
1110 return (L span ITvocurly)
1112 setContext (Layout offset : ctx)
1113 return (L span ITvocurly)
1115 do_layout_left :: Action
1116 do_layout_left span _buf _len = do
1118 pushLexState bol -- we must be at the start of a line
1119 return (L span ITvccurly)
1121 -- -----------------------------------------------------------------------------
1124 setLine :: Int -> Action
1125 setLine code span buf len = do
1126 let line = parseUnsignedInteger buf len 10 octDecDigit
1127 setSrcLoc (mkSrcLoc (srcSpanFile span) (fromIntegral line - 1) 1)
1128 -- subtract one: the line number refers to the *following* line
1133 setFile :: Int -> Action
1134 setFile code span buf len = do
1135 let file = lexemeToFastString (stepOn buf) (len-2)
1136 setAlrLastLoc noSrcSpan
1137 setSrcLoc (mkSrcLoc file (srcSpanEndLine span) (srcSpanEndCol span))
1143 -- -----------------------------------------------------------------------------
1144 -- Options, includes and language pragmas.
1146 lex_string_prag :: (String -> Token) -> Action
1147 lex_string_prag mkTok span _buf _len
1148 = do input <- getInput
1152 return (L (mkSrcSpan start end) tok)
1154 = if isString input "#-}"
1155 then do setInput input
1156 return (mkTok (reverse acc))
1157 else case alexGetChar input of
1158 Just (c,i) -> go (c:acc) i
1159 Nothing -> err input
1160 isString _ [] = True
1162 = case alexGetChar i of
1163 Just (c,i') | c == x -> isString i' xs
1165 err (AI end _) = failLocMsgP (srcSpanStart span) end "unterminated options pragma"
1168 -- -----------------------------------------------------------------------------
1171 -- This stuff is horrible. I hates it.
1173 lex_string_tok :: Action
1174 lex_string_tok span _buf _len = do
1175 tok <- lex_string ""
1177 return (L (mkSrcSpan (srcSpanStart span) end) tok)
1179 lex_string :: String -> P Token
1182 case alexGetChar' i of
1183 Nothing -> lit_error i
1187 magicHash <- extension magicHashEnabled
1191 case alexGetChar' i of
1195 then failMsgP "primitive string literal must contain only characters <= \'\\xFF\'"
1196 else let s' = mkZFastString (reverse s) in
1197 return (ITprimstring s')
1198 -- mkZFastString is a hack to avoid encoding the
1199 -- string in UTF-8. We just want the exact bytes.
1201 return (ITstring (mkFastString (reverse s)))
1203 return (ITstring (mkFastString (reverse s)))
1206 | Just ('&',i) <- next -> do
1207 setInput i; lex_string s
1208 | Just (c,i) <- next, c <= '\x7f' && is_space c -> do
1209 -- is_space only works for <= '\x7f' (#3751)
1210 setInput i; lex_stringgap s
1211 where next = alexGetChar' i
1215 '\\' -> do setInput i1; c' <- lex_escape; lex_string (c':s)
1216 c | isAny c -> do setInput i1; lex_string (c:s)
1217 _other -> lit_error i
1219 lex_stringgap :: String -> P Token
1220 lex_stringgap s = do
1222 c <- getCharOrFail i
1224 '\\' -> lex_string s
1225 c | is_space c -> lex_stringgap s
1226 _other -> lit_error i
1229 lex_char_tok :: Action
1230 -- Here we are basically parsing character literals, such as 'x' or '\n'
1231 -- but, when Template Haskell is on, we additionally spot
1232 -- 'x and ''T, returning ITvarQuote and ITtyQuote respectively,
1233 -- but WITHOUT CONSUMING the x or T part (the parser does that).
1234 -- So we have to do two characters of lookahead: when we see 'x we need to
1235 -- see if there's a trailing quote
1236 lex_char_tok span _buf _len = do -- We've seen '
1237 i1 <- getInput -- Look ahead to first character
1238 let loc = srcSpanStart span
1239 case alexGetChar' i1 of
1240 Nothing -> lit_error i1
1242 Just ('\'', i2@(AI end2 _)) -> do -- We've seen ''
1243 th_exts <- extension thEnabled
1246 return (L (mkSrcSpan loc end2) ITtyQuote)
1249 Just ('\\', i2@(AI _end2 _)) -> do -- We've seen 'backslash
1251 lit_ch <- lex_escape
1253 mc <- getCharOrFail i3 -- Trailing quote
1254 if mc == '\'' then finish_char_tok loc lit_ch
1257 Just (c, i2@(AI _end2 _))
1258 | not (isAny c) -> lit_error i1
1261 -- We've seen 'x, where x is a valid character
1262 -- (i.e. not newline etc) but not a quote or backslash
1263 case alexGetChar' i2 of -- Look ahead one more character
1264 Just ('\'', i3) -> do -- We've seen 'x'
1266 finish_char_tok loc c
1267 _other -> do -- We've seen 'x not followed by quote
1268 -- (including the possibility of EOF)
1269 -- If TH is on, just parse the quote only
1270 th_exts <- extension thEnabled
1272 if th_exts then return (L (mkSrcSpan loc end) ITvarQuote)
1275 finish_char_tok :: SrcLoc -> Char -> P (Located Token)
1276 finish_char_tok loc ch -- We've already seen the closing quote
1277 -- Just need to check for trailing #
1278 = do magicHash <- extension magicHashEnabled
1279 i@(AI end _) <- getInput
1280 if magicHash then do
1281 case alexGetChar' i of
1282 Just ('#',i@(AI end _)) -> do
1284 return (L (mkSrcSpan loc end) (ITprimchar ch))
1286 return (L (mkSrcSpan loc end) (ITchar ch))
1288 return (L (mkSrcSpan loc end) (ITchar ch))
1290 isAny :: Char -> Bool
1291 isAny c | c > '\x7f' = isPrint c
1292 | otherwise = is_any c
1294 lex_escape :: P Char
1297 c <- getCharOrFail i0
1309 '^' -> do i1 <- getInput
1310 c <- getCharOrFail i1
1311 if c >= '@' && c <= '_'
1312 then return (chr (ord c - ord '@'))
1315 'x' -> readNum is_hexdigit 16 hexDigit
1316 'o' -> readNum is_octdigit 8 octDecDigit
1317 x | is_decdigit x -> readNum2 is_decdigit 10 octDecDigit (octDecDigit x)
1321 case alexGetChar' i of
1322 Nothing -> lit_error i0
1324 case alexGetChar' i2 of
1325 Nothing -> do lit_error i0
1327 let str = [c1,c2,c3] in
1328 case [ (c,rest) | (p,c) <- silly_escape_chars,
1329 Just rest <- [stripPrefix p str] ] of
1330 (escape_char,[]):_ -> do
1333 (escape_char,_:_):_ -> do
1338 readNum :: (Char -> Bool) -> Int -> (Char -> Int) -> P Char
1339 readNum is_digit base conv = do
1341 c <- getCharOrFail i
1343 then readNum2 is_digit base conv (conv c)
1346 readNum2 :: (Char -> Bool) -> Int -> (Char -> Int) -> Int -> P Char
1347 readNum2 is_digit base conv i = do
1350 where read i input = do
1351 case alexGetChar' input of
1352 Just (c,input') | is_digit c -> do
1353 read (i*base + conv c) input'
1355 if i >= 0 && i <= 0x10FFFF
1356 then do setInput input; return (chr i)
1357 else lit_error input
1359 silly_escape_chars :: [(String, Char)]
1360 silly_escape_chars = [
1397 -- before calling lit_error, ensure that the current input is pointing to
1398 -- the position of the error in the buffer. This is so that we can report
1399 -- a correct location to the user, but also so we can detect UTF-8 decoding
1400 -- errors if they occur.
1401 lit_error :: AlexInput -> P a
1402 lit_error i = do setInput i; lexError "lexical error in string/character literal"
1404 getCharOrFail :: AlexInput -> P Char
1405 getCharOrFail i = do
1406 case alexGetChar' i of
1407 Nothing -> lexError "unexpected end-of-file in string/character literal"
1408 Just (c,i) -> do setInput i; return c
1410 -- -----------------------------------------------------------------------------
1413 lex_quasiquote_tok :: Action
1414 lex_quasiquote_tok span buf len = do
1415 let quoter = reverse $ takeWhile (/= '$')
1416 $ reverse $ lexemeToString buf (len - 1)
1417 quoteStart <- getSrcLoc
1418 quote <- lex_quasiquote ""
1420 return (L (mkSrcSpan (srcSpanStart span) end)
1421 (ITquasiQuote (mkFastString quoter,
1422 mkFastString (reverse quote),
1423 mkSrcSpan quoteStart end)))
1425 lex_quasiquote :: String -> P String
1426 lex_quasiquote s = do
1428 case alexGetChar' i of
1429 Nothing -> lit_error i
1432 | Just ('|',i) <- next -> do
1433 setInput i; lex_quasiquote ('|' : s)
1434 | Just (']',i) <- next -> do
1435 setInput i; lex_quasiquote (']' : s)
1436 where next = alexGetChar' i
1439 | Just (']',i) <- next -> do
1440 setInput i; return s
1441 where next = alexGetChar' i
1444 setInput i; lex_quasiquote (c : s)
1446 -- -----------------------------------------------------------------------------
1449 warn :: DynFlag -> SDoc -> Action
1450 warn option warning srcspan _buf _len = do
1451 addWarning option srcspan warning
1454 warnThen :: DynFlag -> SDoc -> Action -> Action
1455 warnThen option warning action srcspan buf len = do
1456 addWarning option srcspan warning
1457 action srcspan buf len
1459 -- -----------------------------------------------------------------------------
1470 SrcSpan -- The start and end of the text span related to
1471 -- the error. Might be used in environments which can
1472 -- show this span, e.g. by highlighting it.
1473 Message -- The error message
1475 data PState = PState {
1476 buffer :: StringBuffer,
1478 messages :: Messages,
1479 last_loc :: SrcSpan, -- pos of previous token
1480 last_len :: !Int, -- len of previous token
1481 loc :: SrcLoc, -- current loc (end of prev token + 1)
1482 extsBitmap :: !Int, -- bitmap that determines permitted extensions
1483 context :: [LayoutContext],
1485 -- Used in the alternative layout rule:
1486 -- These tokens are the next ones to be sent out. They are
1487 -- just blindly emitted, without the rule looking at them again:
1488 alr_pending_implicit_tokens :: [Located Token],
1489 -- This is the next token to be considered or, if it is Nothing,
1490 -- we need to get the next token from the input stream:
1491 alr_next_token :: Maybe (Located Token),
1492 -- This is what we consider to be the locatino of the last token
1494 alr_last_loc :: SrcSpan,
1495 -- The stack of layout contexts:
1496 alr_context :: [ALRContext],
1497 -- Are we expecting a '{'? If it's Just, then the ALRLayout tells
1498 -- us what sort of layout the '{' will open:
1499 alr_expecting_ocurly :: Maybe ALRLayout
1501 -- last_loc and last_len are used when generating error messages,
1502 -- and in pushCurrentContext only. Sigh, if only Happy passed the
1503 -- current token to happyError, we could at least get rid of last_len.
1504 -- Getting rid of last_loc would require finding another way to
1505 -- implement pushCurrentContext (which is only called from one place).
1507 data ALRContext = ALRNoLayout Bool{- does it contain commas? -}
1508 | ALRLayout ALRLayout Int
1509 data ALRLayout = ALRLayoutLet
1514 newtype P a = P { unP :: PState -> ParseResult a }
1516 instance Monad P where
1522 returnP a = a `seq` (P $ \s -> POk s a)
1524 thenP :: P a -> (a -> P b) -> P b
1525 (P m) `thenP` k = P $ \ s ->
1527 POk s1 a -> (unP (k a)) s1
1528 PFailed span err -> PFailed span err
1530 failP :: String -> P a
1531 failP msg = P $ \s -> PFailed (last_loc s) (text msg)
1533 failMsgP :: String -> P a
1534 failMsgP msg = P $ \s -> PFailed (last_loc s) (text msg)
1536 failLocMsgP :: SrcLoc -> SrcLoc -> String -> P a
1537 failLocMsgP loc1 loc2 str = P $ \_ -> PFailed (mkSrcSpan loc1 loc2) (text str)
1539 failSpanMsgP :: SrcSpan -> SDoc -> P a
1540 failSpanMsgP span msg = P $ \_ -> PFailed span msg
1542 getPState :: P PState
1543 getPState = P $ \s -> POk s s
1545 getDynFlags :: P DynFlags
1546 getDynFlags = P $ \s -> POk s (dflags s)
1548 withThisPackage :: (PackageId -> a) -> P a
1550 = do pkg <- liftM thisPackage getDynFlags
1553 extension :: (Int -> Bool) -> P Bool
1554 extension p = P $ \s -> POk s (p $! extsBitmap s)
1557 getExts = P $ \s -> POk s (extsBitmap s)
1559 setExts :: (Int -> Int) -> P ()
1560 setExts f = P $ \s -> POk s{ extsBitmap = f (extsBitmap s) } ()
1562 setSrcLoc :: SrcLoc -> P ()
1563 setSrcLoc new_loc = P $ \s -> POk s{loc=new_loc} ()
1565 getSrcLoc :: P SrcLoc
1566 getSrcLoc = P $ \s@(PState{ loc=loc }) -> POk s loc
1568 setLastToken :: SrcSpan -> Int -> P ()
1569 setLastToken loc len = P $ \s -> POk s {
1574 data AlexInput = AI SrcLoc StringBuffer
1576 alexInputPrevChar :: AlexInput -> Char
1577 alexInputPrevChar (AI _ buf) = prevChar buf '\n'
1579 alexGetChar :: AlexInput -> Maybe (Char,AlexInput)
1580 alexGetChar (AI loc s)
1582 | otherwise = adj_c `seq` loc' `seq` s' `seq`
1583 --trace (show (ord c)) $
1584 Just (adj_c, (AI loc' s'))
1585 where (c,s') = nextChar s
1586 loc' = advanceSrcLoc loc c
1594 other_graphic = '\x6'
1597 | c <= '\x06' = non_graphic
1599 -- Alex doesn't handle Unicode, so when Unicode
1600 -- character is encountered we output these values
1601 -- with the actual character value hidden in the state.
1603 case generalCategory c of
1604 UppercaseLetter -> upper
1605 LowercaseLetter -> lower
1606 TitlecaseLetter -> upper
1607 ModifierLetter -> other_graphic
1608 OtherLetter -> lower -- see #1103
1609 NonSpacingMark -> other_graphic
1610 SpacingCombiningMark -> other_graphic
1611 EnclosingMark -> other_graphic
1612 DecimalNumber -> digit
1613 LetterNumber -> other_graphic
1614 OtherNumber -> other_graphic
1615 ConnectorPunctuation -> symbol
1616 DashPunctuation -> symbol
1617 OpenPunctuation -> other_graphic
1618 ClosePunctuation -> other_graphic
1619 InitialQuote -> other_graphic
1620 FinalQuote -> other_graphic
1621 OtherPunctuation -> symbol
1622 MathSymbol -> symbol
1623 CurrencySymbol -> symbol
1624 ModifierSymbol -> symbol
1625 OtherSymbol -> symbol
1627 _other -> non_graphic
1629 -- This version does not squash unicode characters, it is used when
1631 alexGetChar' :: AlexInput -> Maybe (Char,AlexInput)
1632 alexGetChar' (AI loc s)
1634 | otherwise = c `seq` loc' `seq` s' `seq`
1635 --trace (show (ord c)) $
1636 Just (c, (AI loc' s'))
1637 where (c,s') = nextChar s
1638 loc' = advanceSrcLoc loc c
1640 getInput :: P AlexInput
1641 getInput = P $ \s@PState{ loc=l, buffer=b } -> POk s (AI l b)
1643 setInput :: AlexInput -> P ()
1644 setInput (AI l b) = P $ \s -> POk s{ loc=l, buffer=b } ()
1646 pushLexState :: Int -> P ()
1647 pushLexState ls = P $ \s@PState{ lex_state=l } -> POk s{lex_state=ls:l} ()
1649 popLexState :: P Int
1650 popLexState = P $ \s@PState{ lex_state=ls:l } -> POk s{ lex_state=l } ls
1652 getLexState :: P Int
1653 getLexState = P $ \s@PState{ lex_state=ls:_ } -> POk s ls
1655 popNextToken :: P (Maybe (Located Token))
1657 = P $ \s@PState{ alr_next_token = m } ->
1658 POk (s {alr_next_token = Nothing}) m
1660 setAlrLastLoc :: SrcSpan -> P ()
1661 setAlrLastLoc l = P $ \s -> POk (s {alr_last_loc = l}) ()
1663 getAlrLastLoc :: P SrcSpan
1664 getAlrLastLoc = P $ \s@(PState {alr_last_loc = l}) -> POk s l
1666 getALRContext :: P [ALRContext]
1667 getALRContext = P $ \s@(PState {alr_context = cs}) -> POk s cs
1669 setALRContext :: [ALRContext] -> P ()
1670 setALRContext cs = P $ \s -> POk (s {alr_context = cs}) ()
1672 setNextToken :: Located Token -> P ()
1673 setNextToken t = P $ \s -> POk (s {alr_next_token = Just t}) ()
1675 popPendingImplicitToken :: P (Maybe (Located Token))
1676 popPendingImplicitToken
1677 = P $ \s@PState{ alr_pending_implicit_tokens = ts } ->
1680 (t : ts') -> POk (s {alr_pending_implicit_tokens = ts'}) (Just t)
1682 setPendingImplicitTokens :: [Located Token] -> P ()
1683 setPendingImplicitTokens ts = P $ \s -> POk (s {alr_pending_implicit_tokens = ts}) ()
1685 getAlrExpectingOCurly :: P (Maybe ALRLayout)
1686 getAlrExpectingOCurly = P $ \s@(PState {alr_expecting_ocurly = b}) -> POk s b
1688 setAlrExpectingOCurly :: Maybe ALRLayout -> P ()
1689 setAlrExpectingOCurly b = P $ \s -> POk (s {alr_expecting_ocurly = b}) ()
1691 -- for reasons of efficiency, flags indicating language extensions (eg,
1692 -- -fglasgow-exts or -XParr) are represented by a bitmap stored in an unboxed
1696 genericsBit = 0 -- {| and |}
1707 explicitForallBit :: Int
1708 explicitForallBit = 7 -- the 'forall' keyword and '.' symbol
1710 bangPatBit = 8 -- Tells the parser to understand bang-patterns
1711 -- (doesn't affect the lexer)
1713 tyFamBit = 9 -- indexed type families: 'family' keyword and kind sigs
1715 haddockBit = 10 -- Lex and parse Haddock comments
1717 magicHashBit = 11 -- "#" in both functions and operators
1719 kindSigsBit = 12 -- Kind signatures on type variables
1720 recursiveDoBit :: Int
1721 recursiveDoBit = 13 -- mdo
1722 unicodeSyntaxBit :: Int
1723 unicodeSyntaxBit = 14 -- the forall symbol, arrow symbols, etc
1724 unboxedTuplesBit :: Int
1725 unboxedTuplesBit = 15 -- (# and #)
1726 standaloneDerivingBit :: Int
1727 standaloneDerivingBit = 16 -- standalone instance deriving declarations
1728 transformComprehensionsBit :: Int
1729 transformComprehensionsBit = 17
1731 qqBit = 18 -- enable quasiquoting
1732 inRulePragBit :: Int
1734 rawTokenStreamBit :: Int
1735 rawTokenStreamBit = 20 -- producing a token stream with all comments included
1736 newQualOpsBit :: Int
1737 newQualOpsBit = 21 -- Haskell' qualified operator syntax, e.g. Prelude.(+)
1740 alternativeLayoutRuleBit :: Int
1741 alternativeLayoutRuleBit = 23
1743 always :: Int -> Bool
1745 genericsEnabled :: Int -> Bool
1746 genericsEnabled flags = testBit flags genericsBit
1747 parrEnabled :: Int -> Bool
1748 parrEnabled flags = testBit flags parrBit
1749 arrowsEnabled :: Int -> Bool
1750 arrowsEnabled flags = testBit flags arrowsBit
1751 thEnabled :: Int -> Bool
1752 thEnabled flags = testBit flags thBit
1753 ipEnabled :: Int -> Bool
1754 ipEnabled flags = testBit flags ipBit
1755 explicitForallEnabled :: Int -> Bool
1756 explicitForallEnabled flags = testBit flags explicitForallBit
1757 bangPatEnabled :: Int -> Bool
1758 bangPatEnabled flags = testBit flags bangPatBit
1759 -- tyFamEnabled :: Int -> Bool
1760 -- tyFamEnabled flags = testBit flags tyFamBit
1761 haddockEnabled :: Int -> Bool
1762 haddockEnabled flags = testBit flags haddockBit
1763 magicHashEnabled :: Int -> Bool
1764 magicHashEnabled flags = testBit flags magicHashBit
1765 -- kindSigsEnabled :: Int -> Bool
1766 -- kindSigsEnabled flags = testBit flags kindSigsBit
1767 unicodeSyntaxEnabled :: Int -> Bool
1768 unicodeSyntaxEnabled flags = testBit flags unicodeSyntaxBit
1769 unboxedTuplesEnabled :: Int -> Bool
1770 unboxedTuplesEnabled flags = testBit flags unboxedTuplesBit
1771 standaloneDerivingEnabled :: Int -> Bool
1772 standaloneDerivingEnabled flags = testBit flags standaloneDerivingBit
1773 qqEnabled :: Int -> Bool
1774 qqEnabled flags = testBit flags qqBit
1775 -- inRulePrag :: Int -> Bool
1776 -- inRulePrag flags = testBit flags inRulePragBit
1777 rawTokenStreamEnabled :: Int -> Bool
1778 rawTokenStreamEnabled flags = testBit flags rawTokenStreamBit
1779 newQualOps :: Int -> Bool
1780 newQualOps flags = testBit flags newQualOpsBit
1781 oldQualOps :: Int -> Bool
1782 oldQualOps flags = not (newQualOps flags)
1783 alternativeLayoutRule :: Int -> Bool
1784 alternativeLayoutRule flags = testBit flags alternativeLayoutRuleBit
1786 -- PState for parsing options pragmas
1788 pragState :: DynFlags -> StringBuffer -> SrcLoc -> PState
1789 pragState dynflags buf loc =
1792 messages = emptyMessages,
1794 last_loc = mkSrcSpan loc loc,
1799 lex_state = [bol, option_prags, 0],
1800 alr_pending_implicit_tokens = [],
1801 alr_next_token = Nothing,
1802 alr_last_loc = noSrcSpan,
1804 alr_expecting_ocurly = Nothing
1808 -- create a parse state
1810 mkPState :: StringBuffer -> SrcLoc -> DynFlags -> PState
1811 mkPState buf loc flags =
1815 messages = emptyMessages,
1816 last_loc = mkSrcSpan loc loc,
1819 extsBitmap = fromIntegral bitmap,
1821 lex_state = [bol, 0],
1822 -- we begin in the layout state if toplev_layout is set
1823 alr_pending_implicit_tokens = [],
1824 alr_next_token = Nothing,
1825 alr_last_loc = noSrcSpan,
1827 alr_expecting_ocurly = Nothing
1830 bitmap = genericsBit `setBitIf` dopt Opt_Generics flags
1831 .|. ffiBit `setBitIf` dopt Opt_ForeignFunctionInterface flags
1832 .|. parrBit `setBitIf` dopt Opt_PArr flags
1833 .|. arrowsBit `setBitIf` dopt Opt_Arrows flags
1834 .|. thBit `setBitIf` dopt Opt_TemplateHaskell flags
1835 .|. qqBit `setBitIf` dopt Opt_QuasiQuotes flags
1836 .|. ipBit `setBitIf` dopt Opt_ImplicitParams flags
1837 .|. explicitForallBit `setBitIf` dopt Opt_ExplicitForAll flags
1838 .|. bangPatBit `setBitIf` dopt Opt_BangPatterns flags
1839 .|. tyFamBit `setBitIf` dopt Opt_TypeFamilies flags
1840 .|. haddockBit `setBitIf` dopt Opt_Haddock flags
1841 .|. magicHashBit `setBitIf` dopt Opt_MagicHash flags
1842 .|. kindSigsBit `setBitIf` dopt Opt_KindSignatures flags
1843 .|. recursiveDoBit `setBitIf` dopt Opt_RecursiveDo flags
1844 .|. recBit `setBitIf` dopt Opt_DoRec flags
1845 .|. recBit `setBitIf` dopt Opt_Arrows flags
1846 .|. unicodeSyntaxBit `setBitIf` dopt Opt_UnicodeSyntax flags
1847 .|. unboxedTuplesBit `setBitIf` dopt Opt_UnboxedTuples flags
1848 .|. standaloneDerivingBit `setBitIf` dopt Opt_StandaloneDeriving flags
1849 .|. transformComprehensionsBit `setBitIf` dopt Opt_TransformListComp flags
1850 .|. rawTokenStreamBit `setBitIf` dopt Opt_KeepRawTokenStream flags
1851 .|. newQualOpsBit `setBitIf` dopt Opt_NewQualifiedOperators flags
1852 .|. alternativeLayoutRuleBit `setBitIf` dopt Opt_AlternativeLayoutRule flags
1854 setBitIf :: Int -> Bool -> Int
1855 b `setBitIf` cond | cond = bit b
1858 addWarning :: DynFlag -> SrcSpan -> SDoc -> P ()
1859 addWarning option srcspan warning
1860 = P $ \s@PState{messages=(ws,es), dflags=d} ->
1861 let warning' = mkWarnMsg srcspan alwaysQualify warning
1862 ws' = if dopt option d then ws `snocBag` warning' else ws
1863 in POk s{messages=(ws', es)} ()
1865 getMessages :: PState -> Messages
1866 getMessages PState{messages=ms} = ms
1868 getContext :: P [LayoutContext]
1869 getContext = P $ \s@PState{context=ctx} -> POk s ctx
1871 setContext :: [LayoutContext] -> P ()
1872 setContext ctx = P $ \s -> POk s{context=ctx} ()
1875 popContext = P $ \ s@(PState{ buffer = buf, context = ctx,
1876 last_len = len, last_loc = last_loc }) ->
1878 (_:tl) -> POk s{ context = tl } ()
1879 [] -> PFailed last_loc (srcParseErr buf len)
1881 -- Push a new layout context at the indentation of the last token read.
1882 -- This is only used at the outer level of a module when the 'module'
1883 -- keyword is missing.
1884 pushCurrentContext :: P ()
1885 pushCurrentContext = P $ \ s@PState{ last_loc=loc, context=ctx } ->
1886 POk s{context = Layout (srcSpanStartCol loc) : ctx} ()
1888 getOffside :: P Ordering
1889 getOffside = P $ \s@PState{last_loc=loc, context=stk} ->
1890 let offs = srcSpanStartCol loc in
1891 let ord = case stk of
1892 (Layout n:_) -> --trace ("layout: " ++ show n ++ ", offs: " ++ show offs) $
1897 -- ---------------------------------------------------------------------------
1898 -- Construct a parse error
1901 :: StringBuffer -- current buffer (placed just after the last token)
1902 -> Int -- length of the previous token
1905 = hcat [ if null token
1906 then ptext (sLit "parse error (possibly incorrect indentation)")
1907 else hcat [ptext (sLit "parse error on input "),
1908 char '`', text token, char '\'']
1910 where token = lexemeToString (offsetBytes (-len) buf) len
1912 -- Report a parse failure, giving the span of the previous token as
1913 -- the location of the error. This is the entry point for errors
1914 -- detected during parsing.
1916 srcParseFail = P $ \PState{ buffer = buf, last_len = len,
1917 last_loc = last_loc } ->
1918 PFailed last_loc (srcParseErr buf len)
1920 -- A lexical error is reported at a particular position in the source file,
1921 -- not over a token range.
1922 lexError :: String -> P a
1925 (AI end buf) <- getInput
1926 reportLexError loc end buf str
1928 -- -----------------------------------------------------------------------------
1929 -- This is the top-level function: called from the parser each time a
1930 -- new token is to be read from the input.
1932 lexer :: (Located Token -> P a) -> P a
1934 alr <- extension alternativeLayoutRule
1935 let lexTokenFun = if alr then lexTokenAlr else lexToken
1936 tok@(L _span _tok__) <- lexTokenFun
1937 --trace ("token: " ++ show _tok__) $ do
1940 lexTokenAlr :: P (Located Token)
1941 lexTokenAlr = do mPending <- popPendingImplicitToken
1942 t <- case mPending of
1944 do mNext <- popNextToken
1947 Just next -> return next
1948 alternativeLayoutRuleToken t
1951 setAlrLastLoc (getLoc t)
1953 ITwhere -> setAlrExpectingOCurly (Just ALRLayoutWhere)
1954 ITlet -> setAlrExpectingOCurly (Just ALRLayoutLet)
1955 ITof -> setAlrExpectingOCurly (Just ALRLayoutOf)
1956 ITdo -> setAlrExpectingOCurly (Just ALRLayoutDo)
1960 alternativeLayoutRuleToken :: Located Token -> P (Located Token)
1961 alternativeLayoutRuleToken t
1962 = do context <- getALRContext
1963 lastLoc <- getAlrLastLoc
1964 mExpectingOCurly <- getAlrExpectingOCurly
1965 let thisLoc = getLoc t
1966 thisCol = srcSpanStartCol thisLoc
1967 newLine = (lastLoc == noSrcSpan)
1968 || (srcSpanStartLine thisLoc > srcSpanEndLine lastLoc)
1969 case (unLoc t, context, mExpectingOCurly) of
1970 -- This case handles a GHC extension to the original H98
1972 (ITocurly, _, Just _) ->
1973 do setAlrExpectingOCurly Nothing
1974 setALRContext (ALRNoLayout (containsCommas ITocurly) : context)
1976 -- ...and makes this case unnecessary
1978 -- I think our implicit open-curly handling is slightly
1979 -- different to John's, in how it interacts with newlines
1981 (ITocurly, _, Just _) ->
1982 do setAlrExpectingOCurly Nothing
1986 (_, ALRLayout _ col : ls, Just expectingOCurly)
1987 | (thisCol > col) ||
1989 isNonDecreasingIntentation expectingOCurly) ->
1990 do setAlrExpectingOCurly Nothing
1991 setALRContext (ALRLayout expectingOCurly thisCol : context)
1993 return (L thisLoc ITocurly)
1995 do setAlrExpectingOCurly Nothing
1996 setPendingImplicitTokens [L lastLoc ITccurly]
1998 return (L lastLoc ITocurly)
1999 (_, _, Just expectingOCurly) ->
2000 do setAlrExpectingOCurly Nothing
2001 setALRContext (ALRLayout expectingOCurly thisCol : context)
2003 return (L thisLoc ITocurly)
2004 -- We do the [] cases earlier than in the spec, as we
2005 -- have an actual EOF token
2006 (ITeof, ALRLayout _ _ : ls, _) ->
2009 return (L thisLoc ITccurly)
2012 -- the other ITeof case omitted; general case below covers it
2013 (ITin, ALRLayout ALRLayoutLet _ : ls, _)
2015 do setPendingImplicitTokens [t]
2017 return (L thisLoc ITccurly)
2018 (_, ALRLayout _ col : ls, _)
2019 | newLine && thisCol == col ->
2021 return (L thisLoc ITsemi)
2022 | newLine && thisCol < col ->
2025 -- Note that we use lastLoc, as we may need to close
2026 -- more layouts, or give a semicolon
2027 return (L lastLoc ITccurly)
2030 do setALRContext (ALRNoLayout (containsCommas u) : context)
2035 ALRLayout _ _ : ls ->
2038 return (L thisLoc ITccurly)
2039 ALRNoLayout _ : ls ->
2043 -- XXX This is an error in John's code, but
2044 -- it looks reachable to me at first glance
2046 (ITin, ALRLayout ALRLayoutLet _ : ls, _) ->
2048 setPendingImplicitTokens [t]
2049 return (L thisLoc ITccurly)
2050 (ITin, ALRLayout _ _ : ls, _) ->
2053 return (L thisLoc ITccurly)
2054 -- the other ITin case omitted; general case below covers it
2055 (ITcomma, ALRLayout _ _ : ls, _)
2056 | topNoLayoutContainsCommas ls ->
2059 return (L thisLoc ITccurly)
2060 (ITwhere, ALRLayout ALRLayoutDo _ : ls, _) ->
2062 setPendingImplicitTokens [t]
2063 return (L thisLoc ITccurly)
2064 -- the other ITwhere case omitted; general case below covers it
2065 (_, _, _) -> return t
2067 isALRopen :: Token -> Bool
2068 isALRopen ITcase = True
2069 isALRopen ITif = True
2070 isALRopen IToparen = True
2071 isALRopen ITobrack = True
2072 isALRopen ITocurly = True
2074 isALRopen IToubxparen = True
2077 isALRclose :: Token -> Bool
2078 isALRclose ITof = True
2079 isALRclose ITthen = True
2080 isALRclose ITcparen = True
2081 isALRclose ITcbrack = True
2082 isALRclose ITccurly = True
2084 isALRclose ITcubxparen = True
2085 isALRclose _ = False
2087 isNonDecreasingIntentation :: ALRLayout -> Bool
2088 isNonDecreasingIntentation ALRLayoutDo = True
2089 isNonDecreasingIntentation _ = False
2091 containsCommas :: Token -> Bool
2092 containsCommas IToparen = True
2093 containsCommas ITobrack = True
2094 -- John doesn't have {} as containing commas, but records contain them,
2095 -- which caused a problem parsing Cabal's Distribution.Simple.InstallDirs
2096 -- (defaultInstallDirs).
2097 containsCommas ITocurly = True
2099 containsCommas IToubxparen = True
2100 containsCommas _ = False
2102 topNoLayoutContainsCommas :: [ALRContext] -> Bool
2103 topNoLayoutContainsCommas [] = False
2104 topNoLayoutContainsCommas (ALRLayout _ _ : ls) = topNoLayoutContainsCommas ls
2105 topNoLayoutContainsCommas (ALRNoLayout b : _) = b
2107 lexToken :: P (Located Token)
2109 inp@(AI loc1 buf) <- getInput
2112 case alexScanUser exts inp sc of
2114 let span = mkSrcSpan loc1 loc1
2116 return (L span ITeof)
2117 AlexError (AI loc2 buf) ->
2118 reportLexError loc1 loc2 buf "lexical error"
2119 AlexSkip inp2 _ -> do
2122 AlexToken inp2@(AI end buf2) _ t -> do
2124 let span = mkSrcSpan loc1 end
2125 let bytes = byteDiff buf buf2
2126 span `seq` setLastToken span bytes
2129 reportLexError :: SrcLoc -> SrcLoc -> StringBuffer -> [Char] -> P a
2130 reportLexError loc1 loc2 buf str
2131 | atEnd buf = failLocMsgP loc1 loc2 (str ++ " at end of input")
2134 c = fst (nextChar buf)
2136 if c == '\0' -- decoding errors are mapped to '\0', see utf8DecodeChar#
2137 then failLocMsgP loc2 loc2 (str ++ " (UTF-8 decoding error)")
2138 else failLocMsgP loc1 loc2 (str ++ " at character " ++ show c)
2140 lexTokenStream :: StringBuffer -> SrcLoc -> DynFlags -> ParseResult [Located Token]
2141 lexTokenStream buf loc dflags = unP go initState
2142 where initState = mkPState buf loc (dopt_set (dopt_unset dflags Opt_Haddock) Opt_KeepRawTokenStream)
2144 ltok <- lexer return
2146 L _ ITeof -> return []
2147 _ -> liftM (ltok:) go
2149 linePrags = Map.singleton "line" (begin line_prag2)
2151 fileHeaderPrags = Map.fromList([("options", lex_string_prag IToptions_prag),
2152 ("options_ghc", lex_string_prag IToptions_prag),
2153 ("options_haddock", lex_string_prag ITdocOptions),
2154 ("language", token ITlanguage_prag),
2155 ("include", lex_string_prag ITinclude_prag)])
2157 ignoredPrags = Map.fromList (map ignored pragmas)
2158 where ignored opt = (opt, nested_comment lexToken)
2159 impls = ["hugs", "nhc98", "jhc", "yhc", "catch", "derive"]
2160 options_pragmas = map ("options_" ++) impls
2161 -- CFILES is a hugs-only thing.
2162 pragmas = options_pragmas ++ ["cfiles", "contract"]
2164 oneWordPrags = Map.fromList([("rules", rulePrag),
2165 ("inline", token (ITinline_prag True)),
2166 ("notinline", token (ITinline_prag False)),
2167 ("specialize", token ITspec_prag),
2168 ("source", token ITsource_prag),
2169 ("warning", token ITwarning_prag),
2170 ("deprecated", token ITdeprecated_prag),
2171 ("scc", token ITscc_prag),
2172 ("generated", token ITgenerated_prag),
2173 ("core", token ITcore_prag),
2174 ("unpack", token ITunpack_prag),
2175 ("ann", token ITann_prag)])
2177 twoWordPrags = Map.fromList([("inline conlike", token (ITinline_conlike_prag True)),
2178 ("notinline conlike", token (ITinline_conlike_prag False)),
2179 ("specialize inline", token (ITspec_inline_prag True)),
2180 ("specialize notinline", token (ITspec_inline_prag False))])
2183 dispatch_pragmas :: Map String Action -> Action
2184 dispatch_pragmas prags span buf len = case Map.lookup (clean_pragma (lexemeToString buf len)) prags of
2185 Just found -> found span buf len
2186 Nothing -> lexError "unknown pragma"
2188 known_pragma :: Map String Action -> AlexAccPred Int
2189 known_pragma prags _ _ len (AI _ buf) = (isJust $ Map.lookup (clean_pragma (lexemeToString (offsetBytes (- len) buf) len)) prags)
2190 && (nextCharIs buf (\c -> not (isAlphaNum c || c == '_')))
2192 clean_pragma :: String -> String
2193 clean_pragma prag = canon_ws (map toLower (unprefix prag))
2194 where unprefix prag' = case stripPrefix "{-#" prag' of
2197 canonical prag' = case prag' of
2198 "noinline" -> "notinline"
2199 "specialise" -> "specialize"
2200 "constructorlike" -> "conlike"
2202 canon_ws s = unwords (map canonical (words s))