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 -----------------------------------------------------------------------------
16 -- - parsing integers is a bit slow
17 -- - readRational is a bit slow
19 -- Known bugs, that were also in the previous version:
20 -- - M... should be 3 tokens, not 1.
21 -- - pragma-end should be only valid in a pragma
25 -- The above warning supression flag is a temporary kludge.
26 -- While working on this module you are encouraged to remove it and fix
27 -- any warnings in the module. See
28 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
32 Token(..), lexer, pragState, mkPState, PState(..),
33 P(..), ParseResult(..), getSrcLoc,
34 failLocMsgP, failSpanMsgP, srcParseFail,
36 popContext, pushCurrentContext, setLastToken, setSrcLoc,
37 getLexState, popLexState, pushLexState,
38 extension, standaloneDerivingEnabled, bangPatEnabled,
42 #include "HsVersions.h"
54 import Util ( maybePrefixMatch, readRational )
58 import Data.Char ( chr, ord, isSpace )
62 #if __GLASGOW_HASKELL__ >= 605
63 import Data.Char ( GeneralCategory(..), generalCategory, isPrint, isUpper )
65 import Compat.Unicode ( GeneralCategory(..), generalCategory, isPrint, isUpper )
69 $unispace = \x05 -- Trick Alex into handling Unicode. See alexGetChar.
70 $whitechar = [\ \n\r\f\v\xa0 $unispace]
71 $white_no_nl = $whitechar # \n
75 $unidigit = \x03 -- Trick Alex into handling Unicode. See alexGetChar.
76 $decdigit = $ascdigit -- for now, should really be $digit (ToDo)
77 $digit = [$ascdigit $unidigit]
79 $special = [\(\)\,\;\[\]\`\{\}]
80 $ascsymbol = [\!\#\$\%\&\*\+\.\/\<\=\>\?\@\\\^\|\-\~ \xa1-\xbf \xd7 \xf7]
81 $unisymbol = \x04 -- Trick Alex into handling Unicode. See alexGetChar.
82 $symbol = [$ascsymbol $unisymbol] # [$special \_\:\"\']
84 $unilarge = \x01 -- Trick Alex into handling Unicode. See alexGetChar.
85 $asclarge = [A-Z \xc0-\xd6 \xd8-\xde]
86 $large = [$asclarge $unilarge]
88 $unismall = \x02 -- Trick Alex into handling Unicode. See alexGetChar.
89 $ascsmall = [a-z \xdf-\xf6 \xf8-\xff]
90 $small = [$ascsmall $unismall \_]
92 $unigraphic = \x06 -- Trick Alex into handling Unicode. See alexGetChar.
93 $graphic = [$small $large $symbol $digit $special $unigraphic \:\"\']
96 $hexit = [$decdigit A-F a-f]
97 $symchar = [$symbol \:]
99 $idchar = [$small $large $digit \']
101 $docsym = [\| \^ \* \$]
103 @varid = $small $idchar*
104 @conid = $large $idchar*
106 @varsym = $symbol $symchar*
107 @consym = \: $symchar*
109 @decimal = $decdigit+
111 @hexadecimal = $hexit+
112 @exponent = [eE] [\-\+]? @decimal
114 -- we support the hierarchical module name extension:
117 @floating_point = @decimal \. @decimal @exponent? | @decimal @exponent
119 -- normal signed numerical literals can only be explicitly negative,
120 -- not explicitly positive (contrast @exponent)
122 @signed = @negative ?
126 -- everywhere: skip whitespace and comments
128 $tab+ { warn Opt_WarnTabs (text "Tab character") }
130 -- Everywhere: deal with nested comments. We explicitly rule out
131 -- pragmas, "{-#", so that we don't accidentally treat them as comments.
132 -- (this can happen even though pragmas will normally take precedence due to
133 -- longest-match, because pragmas aren't valid in every state, but comments
134 -- are). We also rule out nested Haddock comments, if the -haddock flag is
137 "{-" / { isNormalComment } { nested_comment lexToken }
139 -- Single-line comments are a bit tricky. Haskell 98 says that two or
140 -- more dashes followed by a symbol should be parsed as a varsym, so we
141 -- have to exclude those.
143 -- Since Haddock comments aren't valid in every state, we need to rule them
146 -- The following two rules match comments that begin with two dashes, but
147 -- continue with a different character. The rules test that this character
148 -- is not a symbol (in which case we'd have a varsym), and that it's not a
149 -- space followed by a Haddock comment symbol (docsym) (in which case we'd
150 -- have a Haddock comment). The rules then munch the rest of the line.
152 "-- " ~[$docsym \#] .* ;
153 "--" [^$symbol : \ ] .* ;
155 -- Next, match Haddock comments if no -haddock flag
157 "-- " [$docsym \#] .* / { ifExtension (not . haddockEnabled) } ;
159 -- Now, when we've matched comments that begin with 2 dashes and continue
160 -- with a different character, we need to match comments that begin with three
161 -- or more dashes (which clearly can't be Haddock comments). We only need to
162 -- make sure that the first non-dash character isn't a symbol, and munch the
165 "---"\-* [^$symbol :] .* ;
167 -- Since the previous rules all match dashes followed by at least one
168 -- character, we also need to match a whole line filled with just dashes.
170 "--"\-* / { atEOL } ;
172 -- We need this rule since none of the other single line comment rules
173 -- actually match this case.
177 -- 'bol' state: beginning of a line. Slurp up all the whitespace (including
178 -- blank lines) until we find a non-whitespace character, then do layout
181 -- One slight wibble here: what if the line begins with {-#? In
182 -- theory, we have to lex the pragma to see if it's one we recognise,
183 -- and if it is, then we backtrack and do_bol, otherwise we treat it
184 -- as a nested comment. We don't bother with this: if the line begins
185 -- with {-#, then we'll assume it's a pragma we know about and go for do_bol.
188 ^\# (line)? { begin line_prag1 }
189 ^\# pragma .* \n ; -- GCC 3.3 CPP generated, apparently
190 ^\# \! .* \n ; -- #!, for scripts
194 -- after a layout keyword (let, where, do, of), we begin a new layout
195 -- context if the curly brace is missing.
196 -- Careful! This stuff is quite delicate.
197 <layout, layout_do> {
198 \{ / { notFollowedBy '-' } { pop_and open_brace }
199 -- we might encounter {-# here, but {- has been handled already
201 ^\# (line)? { begin line_prag1 }
204 -- do is treated in a subtly different way, see new_layout_context
205 <layout> () { new_layout_context True }
206 <layout_do> () { new_layout_context False }
208 -- after a new layout context which was found to be to the left of the
209 -- previous context, we have generated a '{' token, and we now need to
210 -- generate a matching '}' token.
211 <layout_left> () { do_layout_left }
213 <0,option_prags> \n { begin bol }
215 "{-#" $whitechar* (line|LINE) { begin line_prag2 }
217 -- single-line line pragmas, of the form
218 -- # <line> "<file>" <extra-stuff> \n
219 <line_prag1> $decdigit+ { setLine line_prag1a }
220 <line_prag1a> \" [$graphic \ ]* \" { setFile line_prag1b }
221 <line_prag1b> .* { pop }
223 -- Haskell-style line pragmas, of the form
224 -- {-# LINE <line> "<file>" #-}
225 <line_prag2> $decdigit+ { setLine line_prag2a }
226 <line_prag2a> \" [$graphic \ ]* \" { setFile line_prag2b }
227 <line_prag2b> "#-}"|"-}" { pop }
228 -- NOTE: accept -} at the end of a LINE pragma, for compatibility
229 -- with older versions of GHC which generated these.
231 -- We only want RULES pragmas to be picked up when explicit forall
232 -- syntax is enabled is on, because the contents of the pragma always
233 -- uses it. If it's not on then we're sure to get a parse error.
234 -- (ToDo: we should really emit a warning when ignoring pragmas)
235 -- XXX Now that we can enable this without the -fglasgow-exts hammer,
236 -- is it better just to let the parse error happen?
238 "{-#" $whitechar* (RULES|rules) / { ifExtension explicitForallEnabled } { token ITrules_prag }
241 "{-#" $whitechar* (INLINE|inline) { token (ITinline_prag True) }
242 "{-#" $whitechar* (NO(T?)INLINE|no(t?)inline)
243 { token (ITinline_prag False) }
244 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
245 { token ITspec_prag }
246 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
247 $whitechar* (INLINE|inline) { token (ITspec_inline_prag True) }
248 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
249 $whitechar* (NO(T?)INLINE|no(t?)inline)
250 { token (ITspec_inline_prag False) }
251 "{-#" $whitechar* (SOURCE|source) { token ITsource_prag }
252 "{-#" $whitechar* (DEPRECATED|deprecated)
253 { token ITdeprecated_prag }
254 "{-#" $whitechar* (SCC|scc) { token ITscc_prag }
255 "{-#" $whitechar* (GENERATED|generated)
256 { token ITgenerated_prag }
257 "{-#" $whitechar* (CORE|core) { token ITcore_prag }
258 "{-#" $whitechar* (UNPACK|unpack) { token ITunpack_prag }
260 "{-#" { nested_comment lexToken }
262 -- ToDo: should only be valid inside a pragma:
263 "#-}" { token ITclose_prag}
267 "{-#" $whitechar* (OPTIONS|options) { lex_string_prag IToptions_prag }
268 "{-#" $whitechar* (OPTIONS_GHC|options_ghc)
269 { lex_string_prag IToptions_prag }
270 "{-#" $whitechar* (OPTIONS_HADDOCK|options_haddock)
271 { lex_string_prag ITdocOptions }
272 "-- #" { multiline_doc_comment }
273 "{-#" $whitechar* (LANGUAGE|language) { token ITlanguage_prag }
274 "{-#" $whitechar* (INCLUDE|include) { lex_string_prag ITinclude_prag }
282 -- This is to catch things like {-# OPTIONS OPTIONS_HUGS ...
283 "{-#" $whitechar* $idchar+ { nested_comment lexToken }
286 -- '0' state: ordinary lexemes
291 "-- " $docsym / { ifExtension haddockEnabled } { multiline_doc_comment }
292 "{-" \ ? $docsym / { ifExtension haddockEnabled } { nested_doc_comment }
298 "[:" / { ifExtension parrEnabled } { token ITopabrack }
299 ":]" / { ifExtension parrEnabled } { token ITcpabrack }
303 "[|" / { ifExtension thEnabled } { token ITopenExpQuote }
304 "[e|" / { ifExtension thEnabled } { token ITopenExpQuote }
305 "[p|" / { ifExtension thEnabled } { token ITopenPatQuote }
306 "[d|" / { ifExtension thEnabled } { layout_token ITopenDecQuote }
307 "[t|" / { ifExtension thEnabled } { token ITopenTypQuote }
308 "|]" / { ifExtension thEnabled } { token ITcloseQuote }
309 \$ @varid / { ifExtension thEnabled } { skip_one_varid ITidEscape }
310 "$(" / { ifExtension thEnabled } { token ITparenEscape }
312 "[$" @varid "|" / { ifExtension qqEnabled }
313 { lex_quasiquote_tok }
317 "(|" / { ifExtension arrowsEnabled `alexAndPred` notFollowedBySymbol }
318 { special IToparenbar }
319 "|)" / { ifExtension arrowsEnabled } { special ITcparenbar }
323 \? @varid / { ifExtension ipEnabled } { skip_one_varid ITdupipvarid }
327 "(#" / { ifExtension unboxedTuplesEnabled `alexAndPred` notFollowedBySymbol }
328 { token IToubxparen }
329 "#)" / { ifExtension unboxedTuplesEnabled }
330 { token ITcubxparen }
334 "{|" / { ifExtension genericsEnabled } { token ITocurlybar }
335 "|}" / { ifExtension genericsEnabled } { token ITccurlybar }
339 \( { special IToparen }
340 \) { special ITcparen }
341 \[ { special ITobrack }
342 \] { special ITcbrack }
343 \, { special ITcomma }
344 \; { special ITsemi }
345 \` { special ITbackquote }
352 @qual @varid { idtoken qvarid }
353 @qual @conid { idtoken qconid }
355 @conid { idtoken conid }
359 @qual @varid "#"+ / { ifExtension magicHashEnabled } { idtoken qvarid }
360 @qual @conid "#"+ / { ifExtension magicHashEnabled } { idtoken qconid }
361 @varid "#"+ / { ifExtension magicHashEnabled } { varid }
362 @conid "#"+ / { ifExtension magicHashEnabled } { idtoken conid }
368 @qual @varsym { idtoken qvarsym }
369 @qual @consym { idtoken qconsym }
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#)
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 -- Unboxed floats and doubles (:: Float#, :: Double#)
398 -- prim_{float,double} work with signed literals
399 @signed @floating_point \# / { ifExtension magicHashEnabled } { init_strtoken 1 tok_primfloat }
400 @signed @floating_point \# \# / { ifExtension magicHashEnabled } { init_strtoken 2 tok_primdouble }
403 -- Strings and chars are lexed by hand-written code. The reason is
404 -- that even if we recognise the string or char here in the regex
405 -- lexer, we would still have to parse the string afterward in order
406 -- to convert it to a String.
409 \" { lex_string_tok }
413 -- -----------------------------------------------------------------------------
417 = ITas -- Haskell keywords
441 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
443 | ITforall -- GHC extension keywords
461 | ITinline_prag Bool -- True <=> INLINE, False <=> NOINLINE
462 | ITspec_prag -- SPECIALISE
463 | ITspec_inline_prag Bool -- SPECIALISE INLINE (or NOINLINE)
470 | ITcore_prag -- hdaume: core annotations
473 | IToptions_prag String
474 | ITinclude_prag String
477 | ITdotdot -- reserved symbols
493 | ITbiglam -- GHC-extension symbols
495 | ITocurly -- special symbols
497 | ITocurlybar -- {|, for type applications
498 | ITccurlybar -- |}, for type applications
502 | ITopabrack -- [:, for parallel arrays with -fparr
503 | ITcpabrack -- :], for parallel arrays with -fparr
514 | ITvarid FastString -- identifiers
516 | ITvarsym FastString
517 | ITconsym FastString
518 | ITqvarid (FastString,FastString)
519 | ITqconid (FastString,FastString)
520 | ITqvarsym (FastString,FastString)
521 | ITqconsym (FastString,FastString)
523 | ITdupipvarid FastString -- GHC extension: implicit param: ?x
525 | ITpragma StringBuffer
528 | ITstring FastString
530 | ITrational Rational
533 | ITprimstring FastString
535 | ITprimfloat Rational
536 | ITprimdouble Rational
538 -- MetaHaskell extension tokens
539 | ITopenExpQuote -- [| or [e|
540 | ITopenPatQuote -- [p|
541 | ITopenDecQuote -- [d|
542 | ITopenTypQuote -- [t|
544 | ITidEscape FastString -- $x
545 | ITparenEscape -- $(
548 | ITquasiQuote (FastString,FastString,SrcSpan) -- [:...|...|]
550 -- Arrow notation extension
557 | ITLarrowtail -- -<<
558 | ITRarrowtail -- >>-
560 | ITunknown String -- Used when the lexer can't make sense of it
561 | ITeof -- end of file token
563 -- Documentation annotations
564 | ITdocCommentNext String -- something beginning '-- |'
565 | ITdocCommentPrev String -- something beginning '-- ^'
566 | ITdocCommentNamed String -- something beginning '-- $'
567 | ITdocSection Int String -- a section heading
568 | ITdocOptions String -- doc options (prune, ignore-exports, etc)
569 | ITdocOptionsOld String -- doc options declared "-- # ..."-style
572 deriving Show -- debugging
575 isSpecial :: Token -> Bool
576 -- If we see M.x, where x is a keyword, but
577 -- is special, we treat is as just plain M.x,
579 isSpecial ITas = True
580 isSpecial IThiding = True
581 isSpecial ITqualified = True
582 isSpecial ITforall = True
583 isSpecial ITexport = True
584 isSpecial ITlabel = True
585 isSpecial ITdynamic = True
586 isSpecial ITsafe = True
587 isSpecial ITthreadsafe = True
588 isSpecial ITunsafe = True
589 isSpecial ITccallconv = True
590 isSpecial ITstdcallconv = True
591 isSpecial ITmdo = True
592 isSpecial ITfamily = True
593 isSpecial ITgroup = True
594 isSpecial ITby = True
595 isSpecial ITusing = True
598 -- the bitmap provided as the third component indicates whether the
599 -- corresponding extension keyword is valid under the extension options
600 -- provided to the compiler; if the extension corresponding to *any* of the
601 -- bits set in the bitmap is enabled, the keyword is valid (this setup
602 -- facilitates using a keyword in two different extensions that can be
603 -- activated independently)
605 reservedWordsFM = listToUFM $
606 map (\(x, y, z) -> (mkFastString x, (y, z)))
607 [( "_", ITunderscore, 0 ),
609 ( "case", ITcase, 0 ),
610 ( "class", ITclass, 0 ),
611 ( "data", ITdata, 0 ),
612 ( "default", ITdefault, 0 ),
613 ( "deriving", ITderiving, 0 ),
615 ( "else", ITelse, 0 ),
616 ( "hiding", IThiding, 0 ),
618 ( "import", ITimport, 0 ),
620 ( "infix", ITinfix, 0 ),
621 ( "infixl", ITinfixl, 0 ),
622 ( "infixr", ITinfixr, 0 ),
623 ( "instance", ITinstance, 0 ),
625 ( "module", ITmodule, 0 ),
626 ( "newtype", ITnewtype, 0 ),
628 ( "qualified", ITqualified, 0 ),
629 ( "then", ITthen, 0 ),
630 ( "type", ITtype, 0 ),
631 ( "where", ITwhere, 0 ),
632 ( "_scc_", ITscc, 0 ), -- ToDo: remove
634 ( "forall", ITforall, bit explicitForallBit),
635 ( "mdo", ITmdo, bit recursiveDoBit),
636 ( "family", ITfamily, bit tyFamBit),
637 ( "group", ITgroup, bit transformComprehensionsBit),
638 ( "by", ITby, bit transformComprehensionsBit),
639 ( "using", ITusing, bit transformComprehensionsBit),
641 ( "foreign", ITforeign, bit ffiBit),
642 ( "export", ITexport, bit ffiBit),
643 ( "label", ITlabel, bit ffiBit),
644 ( "dynamic", ITdynamic, bit ffiBit),
645 ( "safe", ITsafe, bit ffiBit),
646 ( "threadsafe", ITthreadsafe, bit ffiBit),
647 ( "unsafe", ITunsafe, bit ffiBit),
648 ( "stdcall", ITstdcallconv, bit ffiBit),
649 ( "ccall", ITccallconv, bit ffiBit),
650 ( "dotnet", ITdotnet, bit ffiBit),
652 ( "rec", ITrec, bit arrowsBit),
653 ( "proc", ITproc, bit arrowsBit)
656 reservedSymsFM :: UniqFM (Token, Int -> Bool)
657 reservedSymsFM = listToUFM $
658 map (\ (x,y,z) -> (mkFastString x,(y,z)))
659 [ ("..", ITdotdot, always)
660 -- (:) is a reserved op, meaning only list cons
661 ,(":", ITcolon, always)
662 ,("::", ITdcolon, always)
663 ,("=", ITequal, always)
664 ,("\\", ITlam, always)
665 ,("|", ITvbar, always)
666 ,("<-", ITlarrow, always)
667 ,("->", ITrarrow, always)
669 ,("~", ITtilde, always)
670 ,("=>", ITdarrow, always)
671 ,("-", ITminus, always)
672 ,("!", ITbang, always)
674 -- For data T (a::*) = MkT
675 ,("*", ITstar, \i -> kindSigsEnabled i || tyFamEnabled i)
676 -- For 'forall a . t'
677 ,(".", ITdot, explicitForallEnabled)
679 ,("-<", ITlarrowtail, arrowsEnabled)
680 ,(">-", ITrarrowtail, arrowsEnabled)
681 ,("-<<", ITLarrowtail, arrowsEnabled)
682 ,(">>-", ITRarrowtail, arrowsEnabled)
684 #if __GLASGOW_HASKELL__ >= 605
685 ,("∷", ITdcolon, unicodeSyntaxEnabled)
686 ,("⇒", ITdarrow, unicodeSyntaxEnabled)
687 ,("∀", ITforall, \i -> unicodeSyntaxEnabled i &&
688 explicitForallEnabled i)
689 ,("→", ITrarrow, unicodeSyntaxEnabled)
690 ,("←", ITlarrow, unicodeSyntaxEnabled)
691 ,("⋯", ITdotdot, unicodeSyntaxEnabled)
692 -- ToDo: ideally, → and ∷ should be "specials", so that they cannot
693 -- form part of a large operator. This would let us have a better
694 -- syntax for kinds: ɑ∷*→* would be a legal kind signature. (maybe).
698 -- -----------------------------------------------------------------------------
701 type Action = SrcSpan -> StringBuffer -> Int -> P (Located Token)
703 special :: Token -> Action
704 special tok span _buf len = return (L span tok)
706 token, layout_token :: Token -> Action
707 token t span buf len = return (L span t)
708 layout_token t span buf len = pushLexState layout >> return (L span t)
710 idtoken :: (StringBuffer -> Int -> Token) -> Action
711 idtoken f span buf len = return (L span $! (f buf len))
713 skip_one_varid :: (FastString -> Token) -> Action
714 skip_one_varid f span buf len
715 = return (L span $! f (lexemeToFastString (stepOn buf) (len-1)))
717 strtoken :: (String -> Token) -> Action
718 strtoken f span buf len =
719 return (L span $! (f $! lexemeToString buf len))
721 init_strtoken :: Int -> (String -> Token) -> Action
722 -- like strtoken, but drops the last N character(s)
723 init_strtoken drop f span buf len =
724 return (L span $! (f $! lexemeToString buf (len-drop)))
726 begin :: Int -> Action
727 begin code _span _str _len = do pushLexState code; lexToken
730 pop _span _buf _len = do popLexState; lexToken
732 pop_and :: Action -> Action
733 pop_and act span buf len = do popLexState; act span buf len
735 {-# INLINE nextCharIs #-}
736 nextCharIs buf p = not (atEnd buf) && p (currentChar buf)
738 notFollowedBy char _ _ _ (AI _ _ buf)
739 = nextCharIs buf (/=char)
741 notFollowedBySymbol _ _ _ (AI _ _ buf)
742 = nextCharIs buf (`notElem` "!#$%&*+./<=>?@\\^|-~")
744 -- We must reject doc comments as being ordinary comments everywhere.
745 -- In some cases the doc comment will be selected as the lexeme due to
746 -- maximal munch, but not always, because the nested comment rule is
747 -- valid in all states, but the doc-comment rules are only valid in
748 -- the non-layout states.
749 isNormalComment bits _ _ (AI _ _ buf)
750 | haddockEnabled bits = notFollowedByDocOrPragma
751 | otherwise = nextCharIs buf (/='#')
753 notFollowedByDocOrPragma
754 = not $ spaceAndP buf (`nextCharIs` (`elem` "|^*$#"))
756 spaceAndP buf p = p buf || nextCharIs buf (==' ') && p (snd (nextChar buf))
758 haddockDisabledAnd p bits _ _ (AI _ _ buf)
759 = if haddockEnabled bits then False else (p buf)
761 atEOL _ _ _ (AI _ _ buf) = atEnd buf || currentChar buf == '\n'
763 ifExtension pred bits _ _ _ = pred bits
765 multiline_doc_comment :: Action
766 multiline_doc_comment span buf _len = withLexedDocType (worker "")
768 worker commentAcc input docType oneLine = case alexGetChar input of
770 | oneLine -> docCommentEnd input commentAcc docType buf span
771 | otherwise -> case checkIfCommentLine input' of
772 Just input -> worker ('\n':commentAcc) input docType False
773 Nothing -> docCommentEnd input commentAcc docType buf span
774 Just (c, input) -> worker (c:commentAcc) input docType oneLine
775 Nothing -> docCommentEnd input commentAcc docType buf span
777 checkIfCommentLine input = check (dropNonNewlineSpace input)
779 check input = case alexGetChar input of
780 Just ('-', input) -> case alexGetChar input of
781 Just ('-', input) -> case alexGetChar input of
782 Just (c, _) | c /= '-' -> Just input
787 dropNonNewlineSpace input = case alexGetChar input of
789 | isSpace c && c /= '\n' -> dropNonNewlineSpace input'
794 nested comments require traversing by hand, they can't be parsed
795 using regular expressions.
797 nested_comment :: P (Located Token) -> Action
798 nested_comment cont span _str _len = do
802 go 0 input = do setInput input; cont
803 go n input = case alexGetChar input of
804 Nothing -> errBrace input span
805 Just ('-',input) -> case alexGetChar input of
806 Nothing -> errBrace input span
807 Just ('\125',input) -> go (n-1) input
808 Just (c,_) -> go n input
809 Just ('\123',input) -> case alexGetChar input of
810 Nothing -> errBrace input span
811 Just ('-',input) -> go (n+1) input
812 Just (c,_) -> go n input
813 Just (c,input) -> go n input
815 nested_doc_comment :: Action
816 nested_doc_comment span buf _len = withLexedDocType (go "")
818 go commentAcc input docType _ = case alexGetChar input of
819 Nothing -> errBrace input span
820 Just ('-',input) -> case alexGetChar input of
821 Nothing -> errBrace input span
822 Just ('\125',input@(AI end _ buf2)) ->
823 docCommentEnd input commentAcc docType buf span
824 Just (c,_) -> go ('-':commentAcc) input docType False
825 Just ('\123', input) -> case alexGetChar input of
826 Nothing -> errBrace input span
827 Just ('-',input) -> do
829 let cont = do input <- getInput; go commentAcc input docType False
830 nested_comment cont span buf _len
831 Just (c,_) -> go ('\123':commentAcc) input docType False
832 Just (c,input) -> go (c:commentAcc) input docType False
834 withLexedDocType lexDocComment = do
835 input@(AI _ _ buf) <- getInput
836 case prevChar buf ' ' of
837 '|' -> lexDocComment input ITdocCommentNext False
838 '^' -> lexDocComment input ITdocCommentPrev False
839 '$' -> lexDocComment input ITdocCommentNamed False
840 '*' -> lexDocSection 1 input
841 '#' -> lexDocComment input ITdocOptionsOld False
843 lexDocSection n input = case alexGetChar input of
844 Just ('*', input) -> lexDocSection (n+1) input
845 Just (c, _) -> lexDocComment input (ITdocSection n) True
846 Nothing -> do setInput input; lexToken -- eof reached, lex it normally
849 -------------------------------------------------------------------------------
850 -- This function is quite tricky. We can't just return a new token, we also
851 -- need to update the state of the parser. Why? Because the token is longer
852 -- than what was lexed by Alex, and the lexToken function doesn't know this, so
853 -- it writes the wrong token length to the parser state. This function is
854 -- called afterwards, so it can just update the state.
856 -- This is complicated by the fact that Haddock tokens can span multiple lines,
857 -- which is something that the original lexer didn't account for.
858 -- I have added last_line_len in the parser state which represents the length
859 -- of the part of the token that is on the last line. It is now used for layout
860 -- calculation in pushCurrentContext instead of last_len. last_len is, like it
861 -- was before, the full length of the token, and it is now only used for error
864 docCommentEnd :: AlexInput -> String -> (String -> Token) -> StringBuffer ->
865 SrcSpan -> P (Located Token)
866 docCommentEnd input commentAcc docType buf span = do
868 let (AI loc last_offs nextBuf) = input
869 comment = reverse commentAcc
870 span' = mkSrcSpan (srcSpanStart span) loc
871 last_len = byteDiff buf nextBuf
873 last_line_len = if (last_offs - last_len < 0)
877 span `seq` setLastToken span' last_len last_line_len
878 return (L span' (docType comment))
880 errBrace (AI end _ _) span = failLocMsgP (srcSpanStart span) end "unterminated `{-'"
882 open_brace, close_brace :: Action
883 open_brace span _str _len = do
885 setContext (NoLayout:ctx)
886 return (L span ITocurly)
887 close_brace span _str _len = do
889 return (L span ITccurly)
891 qvarid buf len = ITqvarid $! splitQualName buf len
892 qconid buf len = ITqconid $! splitQualName buf len
894 splitQualName :: StringBuffer -> Int -> (FastString,FastString)
895 -- takes a StringBuffer and a length, and returns the module name
896 -- and identifier parts of a qualified name. Splits at the *last* dot,
897 -- because of hierarchical module names.
898 splitQualName orig_buf len = split orig_buf orig_buf
901 | orig_buf `byteDiff` buf >= len = done dot_buf
902 | c == '.' = found_dot buf'
903 | otherwise = split buf' dot_buf
905 (c,buf') = nextChar buf
907 -- careful, we might get names like M....
908 -- so, if the character after the dot is not upper-case, this is
909 -- the end of the qualifier part.
910 found_dot buf -- buf points after the '.'
911 | isUpper c = split buf' buf
912 | otherwise = done buf
914 (c,buf') = nextChar buf
917 (lexemeToFastString orig_buf (qual_size - 1),
918 lexemeToFastString dot_buf (len - qual_size))
920 qual_size = orig_buf `byteDiff` dot_buf
924 case lookupUFM reservedWordsFM fs of
925 Just (keyword,0) -> do
927 return (L span keyword)
928 Just (keyword,exts) -> do
929 b <- extension (\i -> exts .&. i /= 0)
930 if b then do maybe_layout keyword
931 return (L span keyword)
932 else return (L span (ITvarid fs))
933 _other -> return (L span (ITvarid fs))
935 fs = lexemeToFastString buf len
937 conid buf len = ITconid fs
938 where fs = lexemeToFastString buf len
940 qvarsym buf len = ITqvarsym $! splitQualName buf len
941 qconsym buf len = ITqconsym $! splitQualName buf len
943 varsym = sym ITvarsym
944 consym = sym ITconsym
946 sym con span buf len =
947 case lookupUFM reservedSymsFM fs of
948 Just (keyword,exts) -> do
950 if b then return (L span keyword)
951 else return (L span $! con fs)
952 _other -> return (L span $! con fs)
954 fs = lexemeToFastString buf len
956 -- Variations on the integral numeric literal.
957 tok_integral :: (Integer -> Token)
958 -> (Integer -> Integer)
959 -- -> (StringBuffer -> StringBuffer) -> (Int -> Int)
961 -> (Integer, (Char->Int)) -> Action
962 tok_integral itint transint transbuf translen (radix,char_to_int) span buf len =
963 return $ L span $ itint $! transint $ parseUnsignedInteger
964 (offsetBytes transbuf buf) (subtract translen len) radix char_to_int
966 -- some conveniences for use with tok_integral
967 tok_num = tok_integral ITinteger
968 tok_primint = tok_integral ITprimint
971 decimal = (10,octDecDigit)
972 octal = (8,octDecDigit)
973 hexadecimal = (16,hexDigit)
975 -- readRational can understand negative rationals, exponents, everything.
976 tok_float str = ITrational $! readRational str
977 tok_primfloat str = ITprimfloat $! readRational str
978 tok_primdouble str = ITprimdouble $! readRational str
980 -- -----------------------------------------------------------------------------
983 -- we're at the first token on a line, insert layout tokens if necessary
985 do_bol span _str _len = do
989 --trace "layout: inserting '}'" $ do
991 -- do NOT pop the lex state, we might have a ';' to insert
992 return (L span ITvccurly)
994 --trace "layout: inserting ';'" $ do
996 return (L span ITsemi)
1001 -- certain keywords put us in the "layout" state, where we might
1002 -- add an opening curly brace.
1003 maybe_layout ITdo = pushLexState layout_do
1004 maybe_layout ITmdo = pushLexState layout_do
1005 maybe_layout ITof = pushLexState layout
1006 maybe_layout ITlet = pushLexState layout
1007 maybe_layout ITwhere = pushLexState layout
1008 maybe_layout ITrec = pushLexState layout
1009 maybe_layout _ = return ()
1011 -- Pushing a new implicit layout context. If the indentation of the
1012 -- next token is not greater than the previous layout context, then
1013 -- Haskell 98 says that the new layout context should be empty; that is
1014 -- the lexer must generate {}.
1016 -- We are slightly more lenient than this: when the new context is started
1017 -- by a 'do', then we allow the new context to be at the same indentation as
1018 -- the previous context. This is what the 'strict' argument is for.
1020 new_layout_context strict span _buf _len = do
1022 (AI _ offset _) <- getInput
1025 Layout prev_off : _ |
1026 (strict && prev_off >= offset ||
1027 not strict && prev_off > offset) -> do
1028 -- token is indented to the left of the previous context.
1029 -- we must generate a {} sequence now.
1030 pushLexState layout_left
1031 return (L span ITvocurly)
1033 setContext (Layout offset : ctx)
1034 return (L span ITvocurly)
1036 do_layout_left span _buf _len = do
1038 pushLexState bol -- we must be at the start of a line
1039 return (L span ITvccurly)
1041 -- -----------------------------------------------------------------------------
1044 setLine :: Int -> Action
1045 setLine code span buf len = do
1046 let line = parseUnsignedInteger buf len 10 octDecDigit
1047 setSrcLoc (mkSrcLoc (srcSpanFile span) (fromIntegral line - 1) 0)
1048 -- subtract one: the line number refers to the *following* line
1053 setFile :: Int -> Action
1054 setFile code span buf len = do
1055 let file = lexemeToFastString (stepOn buf) (len-2)
1056 setSrcLoc (mkSrcLoc file (srcSpanEndLine span) (srcSpanEndCol span))
1062 -- -----------------------------------------------------------------------------
1063 -- Options, includes and language pragmas.
1065 lex_string_prag :: (String -> Token) -> Action
1066 lex_string_prag mkTok span buf len
1067 = do input <- getInput
1071 return (L (mkSrcSpan start end) tok)
1073 = if isString input "#-}"
1074 then do setInput input
1075 return (mkTok (reverse acc))
1076 else case alexGetChar input of
1077 Just (c,i) -> go (c:acc) i
1078 Nothing -> err input
1079 isString i [] = True
1081 = case alexGetChar i of
1082 Just (c,i') | c == x -> isString i' xs
1084 err (AI end _ _) = failLocMsgP (srcSpanStart span) end "unterminated options pragma"
1087 -- -----------------------------------------------------------------------------
1090 -- This stuff is horrible. I hates it.
1092 lex_string_tok :: Action
1093 lex_string_tok span buf len = do
1094 tok <- lex_string ""
1096 return (L (mkSrcSpan (srcSpanStart span) end) tok)
1098 lex_string :: String -> P Token
1101 case alexGetChar' i of
1102 Nothing -> lit_error
1106 magicHash <- extension magicHashEnabled
1110 case alexGetChar' i of
1114 then failMsgP "primitive string literal must contain only characters <= \'\\xFF\'"
1115 else let s' = mkZFastString (reverse s) in
1116 return (ITprimstring s')
1117 -- mkZFastString is a hack to avoid encoding the
1118 -- string in UTF-8. We just want the exact bytes.
1120 return (ITstring (mkFastString (reverse s)))
1122 return (ITstring (mkFastString (reverse s)))
1125 | Just ('&',i) <- next -> do
1126 setInput i; lex_string s
1127 | Just (c,i) <- next, is_space c -> do
1128 setInput i; lex_stringgap s
1129 where next = alexGetChar' i
1135 lex_stringgap s = do
1138 '\\' -> lex_string s
1139 c | is_space c -> lex_stringgap s
1143 lex_char_tok :: Action
1144 -- Here we are basically parsing character literals, such as 'x' or '\n'
1145 -- but, when Template Haskell is on, we additionally spot
1146 -- 'x and ''T, returning ITvarQuote and ITtyQuote respectively,
1147 -- but WIHTOUT CONSUMING the x or T part (the parser does that).
1148 -- So we have to do two characters of lookahead: when we see 'x we need to
1149 -- see if there's a trailing quote
1150 lex_char_tok span buf len = do -- We've seen '
1151 i1 <- getInput -- Look ahead to first character
1152 let loc = srcSpanStart span
1153 case alexGetChar' i1 of
1154 Nothing -> lit_error
1156 Just ('\'', i2@(AI end2 _ _)) -> do -- We've seen ''
1157 th_exts <- extension thEnabled
1160 return (L (mkSrcSpan loc end2) ITtyQuote)
1163 Just ('\\', i2@(AI end2 _ _)) -> do -- We've seen 'backslash
1165 lit_ch <- lex_escape
1166 mc <- getCharOrFail -- Trailing quote
1167 if mc == '\'' then finish_char_tok loc lit_ch
1168 else do setInput i2; lit_error
1170 Just (c, i2@(AI end2 _ _))
1171 | not (isAny c) -> lit_error
1174 -- We've seen 'x, where x is a valid character
1175 -- (i.e. not newline etc) but not a quote or backslash
1176 case alexGetChar' i2 of -- Look ahead one more character
1177 Nothing -> lit_error
1178 Just ('\'', i3) -> do -- We've seen 'x'
1180 finish_char_tok loc c
1181 _other -> do -- We've seen 'x not followed by quote
1182 -- If TH is on, just parse the quote only
1183 th_exts <- extension thEnabled
1184 let (AI end _ _) = i1
1185 if th_exts then return (L (mkSrcSpan loc end) ITvarQuote)
1186 else do setInput i2; lit_error
1188 finish_char_tok :: SrcLoc -> Char -> P (Located Token)
1189 finish_char_tok loc ch -- We've already seen the closing quote
1190 -- Just need to check for trailing #
1191 = do magicHash <- extension magicHashEnabled
1192 i@(AI end _ _) <- getInput
1193 if magicHash then do
1194 case alexGetChar' i of
1195 Just ('#',i@(AI end _ _)) -> do
1197 return (L (mkSrcSpan loc end) (ITprimchar ch))
1199 return (L (mkSrcSpan loc end) (ITchar ch))
1201 return (L (mkSrcSpan loc end) (ITchar ch))
1203 lex_char :: Char -> AlexInput -> P Char
1206 '\\' -> do setInput inp; lex_escape
1207 c | isAny c -> do setInput inp; return c
1210 isAny c | c > '\xff' = isPrint c
1211 | otherwise = is_any c
1213 lex_escape :: P Char
1227 '^' -> do c <- getCharOrFail
1228 if c >= '@' && c <= '_'
1229 then return (chr (ord c - ord '@'))
1232 'x' -> readNum is_hexdigit 16 hexDigit
1233 'o' -> readNum is_octdigit 8 octDecDigit
1234 x | is_digit x -> readNum2 is_digit 10 octDecDigit (octDecDigit x)
1238 case alexGetChar' i of
1239 Nothing -> lit_error
1241 case alexGetChar' i2 of
1242 Nothing -> do setInput i2; lit_error
1244 let str = [c1,c2,c3] in
1245 case [ (c,rest) | (p,c) <- silly_escape_chars,
1246 Just rest <- [maybePrefixMatch p str] ] of
1247 (escape_char,[]):_ -> do
1250 (escape_char,_:_):_ -> do
1255 readNum :: (Char -> Bool) -> Int -> (Char -> Int) -> P Char
1256 readNum is_digit base conv = do
1260 then readNum2 is_digit base conv (conv c)
1261 else do setInput i; lit_error
1263 readNum2 is_digit base conv i = do
1266 where read i input = do
1267 case alexGetChar' input of
1268 Just (c,input') | is_digit c -> do
1269 read (i*base + conv c) input'
1271 if i >= 0 && i <= 0x10FFFF
1272 then do setInput input; return (chr i)
1275 silly_escape_chars = [
1312 -- before calling lit_error, ensure that the current input is pointing to
1313 -- the position of the error in the buffer. This is so that we can report
1314 -- a correct location to the user, but also so we can detect UTF-8 decoding
1315 -- errors if they occur.
1316 lit_error = lexError "lexical error in string/character literal"
1318 getCharOrFail :: P Char
1321 case alexGetChar' i of
1322 Nothing -> lexError "unexpected end-of-file in string/character literal"
1323 Just (c,i) -> do setInput i; return c
1325 -- -----------------------------------------------------------------------------
1328 lex_quasiquote_tok :: Action
1329 lex_quasiquote_tok span buf len = do
1330 let quoter = reverse $ takeWhile (/= '$')
1331 $ reverse $ lexemeToString buf (len - 1)
1332 quoteStart <- getSrcLoc
1333 quote <- lex_quasiquote ""
1335 return (L (mkSrcSpan (srcSpanStart span) end)
1336 (ITquasiQuote (mkFastString quoter,
1337 mkFastString (reverse quote),
1338 mkSrcSpan quoteStart end)))
1340 lex_quasiquote :: String -> P String
1341 lex_quasiquote s = do
1343 case alexGetChar' i of
1344 Nothing -> lit_error
1347 | Just ('|',i) <- next -> do
1348 setInput i; lex_quasiquote ('|' : s)
1349 | Just (']',i) <- next -> do
1350 setInput i; lex_quasiquote (']' : s)
1351 where next = alexGetChar' i
1354 | Just (']',i) <- next -> do
1355 setInput i; return s
1356 where next = alexGetChar' i
1359 setInput i; lex_quasiquote (c : s)
1361 -- -----------------------------------------------------------------------------
1364 warn :: DynFlag -> SDoc -> Action
1365 warn option warning srcspan _buf _len = do
1366 addWarning option srcspan warning
1369 -- -----------------------------------------------------------------------------
1380 SrcSpan -- The start and end of the text span related to
1381 -- the error. Might be used in environments which can
1382 -- show this span, e.g. by highlighting it.
1383 Message -- The error message
1385 data PState = PState {
1386 buffer :: StringBuffer,
1388 messages :: Messages,
1389 last_loc :: SrcSpan, -- pos of previous token
1390 last_offs :: !Int, -- offset of the previous token from the
1391 -- beginning of the current line.
1392 -- \t is equal to 8 spaces.
1393 last_len :: !Int, -- len of previous token
1394 last_line_len :: !Int,
1395 loc :: SrcLoc, -- current loc (end of prev token + 1)
1396 extsBitmap :: !Int, -- bitmap that determines permitted extensions
1397 context :: [LayoutContext],
1400 -- last_loc and last_len are used when generating error messages,
1401 -- and in pushCurrentContext only. Sigh, if only Happy passed the
1402 -- current token to happyError, we could at least get rid of last_len.
1403 -- Getting rid of last_loc would require finding another way to
1404 -- implement pushCurrentContext (which is only called from one place).
1406 newtype P a = P { unP :: PState -> ParseResult a }
1408 instance Monad P where
1414 returnP a = a `seq` (P $ \s -> POk s a)
1416 thenP :: P a -> (a -> P b) -> P b
1417 (P m) `thenP` k = P $ \ s ->
1419 POk s1 a -> (unP (k a)) s1
1420 PFailed span err -> PFailed span err
1422 failP :: String -> P a
1423 failP msg = P $ \s -> PFailed (last_loc s) (text msg)
1425 failMsgP :: String -> P a
1426 failMsgP msg = P $ \s -> PFailed (last_loc s) (text msg)
1428 failLocMsgP :: SrcLoc -> SrcLoc -> String -> P a
1429 failLocMsgP loc1 loc2 str = P $ \s -> PFailed (mkSrcSpan loc1 loc2) (text str)
1431 failSpanMsgP :: SrcSpan -> String -> P a
1432 failSpanMsgP span msg = P $ \s -> PFailed span (text msg)
1434 extension :: (Int -> Bool) -> P Bool
1435 extension p = P $ \s -> POk s (p $! extsBitmap s)
1438 getExts = P $ \s -> POk s (extsBitmap s)
1440 setSrcLoc :: SrcLoc -> P ()
1441 setSrcLoc new_loc = P $ \s -> POk s{loc=new_loc} ()
1443 getSrcLoc :: P SrcLoc
1444 getSrcLoc = P $ \s@(PState{ loc=loc }) -> POk s loc
1446 setLastToken :: SrcSpan -> Int -> Int -> P ()
1447 setLastToken loc len line_len = P $ \s -> POk s {
1450 last_line_len=line_len
1453 data AlexInput = AI SrcLoc {-#UNPACK#-}!Int StringBuffer
1455 alexInputPrevChar :: AlexInput -> Char
1456 alexInputPrevChar (AI _ _ buf) = prevChar buf '\n'
1458 alexGetChar :: AlexInput -> Maybe (Char,AlexInput)
1459 alexGetChar (AI loc ofs s)
1461 | otherwise = adj_c `seq` loc' `seq` ofs' `seq` s' `seq`
1462 --trace (show (ord c)) $
1463 Just (adj_c, (AI loc' ofs' s'))
1464 where (c,s') = nextChar s
1465 loc' = advanceSrcLoc loc c
1466 ofs' = advanceOffs c ofs
1474 other_graphic = '\x6'
1477 | c <= '\x06' = non_graphic
1479 -- Alex doesn't handle Unicode, so when Unicode
1480 -- character is encoutered we output these values
1481 -- with the actual character value hidden in the state.
1483 case generalCategory c of
1484 UppercaseLetter -> upper
1485 LowercaseLetter -> lower
1486 TitlecaseLetter -> upper
1487 ModifierLetter -> other_graphic
1488 OtherLetter -> other_graphic
1489 NonSpacingMark -> other_graphic
1490 SpacingCombiningMark -> other_graphic
1491 EnclosingMark -> other_graphic
1492 DecimalNumber -> digit
1493 LetterNumber -> other_graphic
1494 OtherNumber -> other_graphic
1495 ConnectorPunctuation -> other_graphic
1496 DashPunctuation -> other_graphic
1497 OpenPunctuation -> other_graphic
1498 ClosePunctuation -> other_graphic
1499 InitialQuote -> other_graphic
1500 FinalQuote -> other_graphic
1501 OtherPunctuation -> other_graphic
1502 MathSymbol -> symbol
1503 CurrencySymbol -> symbol
1504 ModifierSymbol -> symbol
1505 OtherSymbol -> symbol
1507 _other -> non_graphic
1509 -- This version does not squash unicode characters, it is used when
1511 alexGetChar' :: AlexInput -> Maybe (Char,AlexInput)
1512 alexGetChar' (AI loc ofs s)
1514 | otherwise = c `seq` loc' `seq` ofs' `seq` s' `seq`
1515 --trace (show (ord c)) $
1516 Just (c, (AI loc' ofs' s'))
1517 where (c,s') = nextChar s
1518 loc' = advanceSrcLoc loc c
1519 ofs' = advanceOffs c ofs
1521 advanceOffs :: Char -> Int -> Int
1522 advanceOffs '\n' offs = 0
1523 advanceOffs '\t' offs = (offs `quot` 8 + 1) * 8
1524 advanceOffs _ offs = offs + 1
1526 getInput :: P AlexInput
1527 getInput = P $ \s@PState{ loc=l, last_offs=o, buffer=b } -> POk s (AI l o b)
1529 setInput :: AlexInput -> P ()
1530 setInput (AI l o b) = P $ \s -> POk s{ loc=l, last_offs=o, buffer=b } ()
1532 pushLexState :: Int -> P ()
1533 pushLexState ls = P $ \s@PState{ lex_state=l } -> POk s{lex_state=ls:l} ()
1535 popLexState :: P Int
1536 popLexState = P $ \s@PState{ lex_state=ls:l } -> POk s{ lex_state=l } ls
1538 getLexState :: P Int
1539 getLexState = P $ \s@PState{ lex_state=ls:l } -> POk s ls
1541 -- for reasons of efficiency, flags indicating language extensions (eg,
1542 -- -fglasgow-exts or -fparr) are represented by a bitmap stored in an unboxed
1545 genericsBit, ffiBit, parrBit :: Int
1546 genericsBit = 0 -- {| and |}
1552 explicitForallBit = 7 -- the 'forall' keyword and '.' symbol
1553 bangPatBit = 8 -- Tells the parser to understand bang-patterns
1554 -- (doesn't affect the lexer)
1555 tyFamBit = 9 -- indexed type families: 'family' keyword and kind sigs
1556 haddockBit = 10 -- Lex and parse Haddock comments
1557 magicHashBit = 11 -- # in both functions and operators
1558 kindSigsBit = 12 -- Kind signatures on type variables
1559 recursiveDoBit = 13 -- mdo
1560 unicodeSyntaxBit = 14 -- the forall symbol, arrow symbols, etc
1561 unboxedTuplesBit = 15 -- (# and #)
1562 standaloneDerivingBit = 16 -- standalone instance deriving declarations
1563 transformComprehensionsBit = 17
1564 qqBit = 18 -- enable quasiquoting
1566 genericsEnabled, ffiEnabled, parrEnabled :: Int -> Bool
1568 genericsEnabled flags = testBit flags genericsBit
1569 ffiEnabled flags = testBit flags ffiBit
1570 parrEnabled flags = testBit flags parrBit
1571 arrowsEnabled flags = testBit flags arrowsBit
1572 thEnabled flags = testBit flags thBit
1573 ipEnabled flags = testBit flags ipBit
1574 explicitForallEnabled flags = testBit flags explicitForallBit
1575 bangPatEnabled flags = testBit flags bangPatBit
1576 tyFamEnabled flags = testBit flags tyFamBit
1577 haddockEnabled flags = testBit flags haddockBit
1578 magicHashEnabled flags = testBit flags magicHashBit
1579 kindSigsEnabled flags = testBit flags kindSigsBit
1580 recursiveDoEnabled flags = testBit flags recursiveDoBit
1581 unicodeSyntaxEnabled flags = testBit flags unicodeSyntaxBit
1582 unboxedTuplesEnabled flags = testBit flags unboxedTuplesBit
1583 standaloneDerivingEnabled flags = testBit flags standaloneDerivingBit
1584 transformComprehensionsEnabled flags = testBit flags transformComprehensionsBit
1585 qqEnabled flags = testBit flags qqBit
1587 -- PState for parsing options pragmas
1589 pragState :: StringBuffer -> SrcLoc -> PState
1593 messages = emptyMessages,
1594 -- XXX defaultDynFlags is not right, but we don't have a real
1596 dflags = defaultDynFlags,
1597 last_loc = mkSrcSpan loc loc,
1604 lex_state = [bol, option_prags, 0]
1608 -- create a parse state
1610 mkPState :: StringBuffer -> SrcLoc -> DynFlags -> PState
1611 mkPState buf loc flags =
1615 messages = emptyMessages,
1616 last_loc = mkSrcSpan loc loc,
1621 extsBitmap = fromIntegral bitmap,
1623 lex_state = [bol, 0]
1624 -- we begin in the layout state if toplev_layout is set
1627 bitmap = genericsBit `setBitIf` dopt Opt_Generics flags
1628 .|. ffiBit `setBitIf` dopt Opt_ForeignFunctionInterface flags
1629 .|. parrBit `setBitIf` dopt Opt_PArr flags
1630 .|. arrowsBit `setBitIf` dopt Opt_Arrows flags
1631 .|. thBit `setBitIf` dopt Opt_TemplateHaskell flags
1632 .|. qqBit `setBitIf` dopt Opt_QuasiQuotes flags
1633 .|. ipBit `setBitIf` dopt Opt_ImplicitParams flags
1634 .|. explicitForallBit `setBitIf` dopt Opt_ScopedTypeVariables flags
1635 .|. explicitForallBit `setBitIf` dopt Opt_PolymorphicComponents flags
1636 .|. explicitForallBit `setBitIf` dopt Opt_ExistentialQuantification flags
1637 .|. explicitForallBit `setBitIf` dopt Opt_Rank2Types flags
1638 .|. explicitForallBit `setBitIf` dopt Opt_RankNTypes flags
1639 .|. bangPatBit `setBitIf` dopt Opt_BangPatterns flags
1640 .|. tyFamBit `setBitIf` dopt Opt_TypeFamilies flags
1641 .|. haddockBit `setBitIf` dopt Opt_Haddock flags
1642 .|. magicHashBit `setBitIf` dopt Opt_MagicHash flags
1643 .|. kindSigsBit `setBitIf` dopt Opt_KindSignatures flags
1644 .|. recursiveDoBit `setBitIf` dopt Opt_RecursiveDo flags
1645 .|. unicodeSyntaxBit `setBitIf` dopt Opt_UnicodeSyntax flags
1646 .|. unboxedTuplesBit `setBitIf` dopt Opt_UnboxedTuples flags
1647 .|. standaloneDerivingBit `setBitIf` dopt Opt_StandaloneDeriving flags
1648 .|. transformComprehensionsBit `setBitIf` dopt Opt_TransformListComp flags
1650 setBitIf :: Int -> Bool -> Int
1651 b `setBitIf` cond | cond = bit b
1654 addWarning :: DynFlag -> SrcSpan -> SDoc -> P ()
1655 addWarning option srcspan warning
1656 = P $ \s@PState{messages=(ws,es), dflags=d} ->
1657 let warning' = mkWarnMsg srcspan alwaysQualify warning
1658 ws' = if dopt option d then ws `snocBag` warning' else ws
1659 in POk s{messages=(ws', es)} ()
1661 getMessages :: PState -> Messages
1662 getMessages PState{messages=ms} = ms
1664 getContext :: P [LayoutContext]
1665 getContext = P $ \s@PState{context=ctx} -> POk s ctx
1667 setContext :: [LayoutContext] -> P ()
1668 setContext ctx = P $ \s -> POk s{context=ctx} ()
1671 popContext = P $ \ s@(PState{ buffer = buf, context = ctx,
1672 loc = loc, last_len = len, last_loc = last_loc }) ->
1674 (_:tl) -> POk s{ context = tl } ()
1675 [] -> PFailed last_loc (srcParseErr buf len)
1677 -- Push a new layout context at the indentation of the last token read.
1678 -- This is only used at the outer level of a module when the 'module'
1679 -- keyword is missing.
1680 pushCurrentContext :: P ()
1681 pushCurrentContext = P $ \ s@PState{ last_offs=offs, last_line_len=len, context=ctx } ->
1682 POk s{context = Layout (offs-len) : ctx} ()
1683 --trace ("off: " ++ show offs ++ ", len: " ++ show len) $ POk s{context = Layout (offs-len) : ctx} ()
1685 getOffside :: P Ordering
1686 getOffside = P $ \s@PState{last_offs=offs, context=stk} ->
1687 let ord = case stk of
1688 (Layout n:_) -> compare offs n
1692 -- ---------------------------------------------------------------------------
1693 -- Construct a parse error
1696 :: StringBuffer -- current buffer (placed just after the last token)
1697 -> Int -- length of the previous token
1700 = hcat [ if null token
1701 then ptext SLIT("parse error (possibly incorrect indentation)")
1702 else hcat [ptext SLIT("parse error on input "),
1703 char '`', text token, char '\'']
1705 where token = lexemeToString (offsetBytes (-len) buf) len
1707 -- Report a parse failure, giving the span of the previous token as
1708 -- the location of the error. This is the entry point for errors
1709 -- detected during parsing.
1711 srcParseFail = P $ \PState{ buffer = buf, last_len = len,
1712 last_loc = last_loc } ->
1713 PFailed last_loc (srcParseErr buf len)
1715 -- A lexical error is reported at a particular position in the source file,
1716 -- not over a token range.
1717 lexError :: String -> P a
1720 i@(AI end _ buf) <- getInput
1721 reportLexError loc end buf str
1723 -- -----------------------------------------------------------------------------
1724 -- This is the top-level function: called from the parser each time a
1725 -- new token is to be read from the input.
1727 lexer :: (Located Token -> P a) -> P a
1729 tok@(L span tok__) <- lexToken
1730 -- trace ("token: " ++ show tok__) $ do
1733 lexToken :: P (Located Token)
1735 inp@(AI loc1 _ buf) <- getInput
1738 case alexScanUser exts inp sc of
1740 let span = mkSrcSpan loc1 loc1
1741 setLastToken span 0 0
1742 return (L span ITeof)
1743 AlexError (AI loc2 _ buf) ->
1744 reportLexError loc1 loc2 buf "lexical error"
1745 AlexSkip inp2 _ -> do
1748 AlexToken inp2@(AI end _ buf2) len t -> do
1750 let span = mkSrcSpan loc1 end
1751 let bytes = byteDiff buf buf2
1752 span `seq` setLastToken span bytes bytes
1755 reportLexError loc1 loc2 buf str
1756 | atEnd buf = failLocMsgP loc1 loc2 (str ++ " at end of input")
1759 c = fst (nextChar buf)
1761 if c == '\0' -- decoding errors are mapped to '\0', see utf8DecodeChar#
1762 then failLocMsgP loc2 loc2 (str ++ " (UTF-8 decoding error)")
1763 else failLocMsgP loc1 loc2 (str ++ " at character " ++ show c)