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 Token(..), lexer, pragState, mkPState, PState(..),
26 P(..), ParseResult(..), getSrcLoc,
27 failLocMsgP, failSpanMsgP, srcParseFail,
29 popContext, pushCurrentContext, setLastToken, setSrcLoc,
30 getLexState, popLexState, pushLexState,
31 extension, glaExtsEnabled, bangPatEnabled
34 #include "HsVersions.h"
46 import Util ( maybePrefixMatch, readRational )
50 import Data.Char ( chr, isSpace )
54 #if __GLASGOW_HASKELL__ >= 605
55 import Data.Char ( GeneralCategory(..), generalCategory, isPrint, isUpper )
57 import Compat.Unicode ( GeneralCategory(..), generalCategory, isPrint, isUpper )
61 $unispace = \x05 -- Trick Alex into handling Unicode. See alexGetChar.
62 $whitechar = [\ \n\r\f\v\xa0 $unispace]
63 $white_no_nl = $whitechar # \n
67 $unidigit = \x03 -- Trick Alex into handling Unicode. See alexGetChar.
68 $decdigit = $ascdigit -- for now, should really be $digit (ToDo)
69 $digit = [$ascdigit $unidigit]
71 $special = [\(\)\,\;\[\]\`\{\}]
72 $ascsymbol = [\!\#\$\%\&\*\+\.\/\<\=\>\?\@\\\^\|\-\~ \xa1-\xbf \xd7 \xf7]
73 $unisymbol = \x04 -- Trick Alex into handling Unicode. See alexGetChar.
74 $symbol = [$ascsymbol $unisymbol] # [$special \_\:\"\']
76 $unilarge = \x01 -- Trick Alex into handling Unicode. See alexGetChar.
77 $asclarge = [A-Z \xc0-\xd6 \xd8-\xde]
78 $large = [$asclarge $unilarge]
80 $unismall = \x02 -- Trick Alex into handling Unicode. See alexGetChar.
81 $ascsmall = [a-z \xdf-\xf6 \xf8-\xff]
82 $small = [$ascsmall $unismall \_]
84 $unigraphic = \x06 -- Trick Alex into handling Unicode. See alexGetChar.
85 $graphic = [$small $large $symbol $digit $special $unigraphic \:\"\']
88 $hexit = [$decdigit A-F a-f]
89 $symchar = [$symbol \:]
91 $idchar = [$small $large $digit \']
93 $docsym = [\| \^ \* \$]
95 @varid = $small $idchar*
96 @conid = $large $idchar*
98 @varsym = $symbol $symchar*
99 @consym = \: $symchar*
101 @decimal = $decdigit+
103 @hexadecimal = $hexit+
104 @exponent = [eE] [\-\+]? @decimal
106 -- we support the hierarchical module name extension:
109 @floating_point = @decimal \. @decimal @exponent? | @decimal @exponent
113 -- everywhere: skip whitespace and comments
115 $tab+ { warn Opt_WarnTabs (text "Tab character") }
117 -- Everywhere: deal with nested comments. We explicitly rule out
118 -- pragmas, "{-#", so that we don't accidentally treat them as comments.
119 -- (this can happen even though pragmas will normally take precedence due to
120 -- longest-match, because pragmas aren't valid in every state, but comments
121 -- are). We also rule out nested Haddock comments, if the -haddock flag is
124 "{-" / { isNormalComment } { nested_comment lexToken }
126 -- Single-line comments are a bit tricky. Haskell 98 says that two or
127 -- more dashes followed by a symbol should be parsed as a varsym, so we
128 -- have to exclude those.
130 -- Since Haddock comments aren't valid in every state, we need to rule them
133 -- The following two rules match comments that begin with two dashes, but
134 -- continue with a different character. The rules test that this character
135 -- is not a symbol (in which case we'd have a varsym), and that it's not a
136 -- space followed by a Haddock comment symbol (docsym) (in which case we'd
137 -- have a Haddock comment). The rules then munch the rest of the line.
140 "--" [^$symbol : \ ] .* ;
142 -- Next, match Haddock comments if no -haddock flag
144 "-- " $docsym .* / { ifExtension (not . haddockEnabled) } ;
146 -- Now, when we've matched comments that begin with 2 dashes and continue
147 -- with a different character, we need to match comments that begin with three
148 -- or more dashes (which clearly can't be Haddock comments). We only need to
149 -- make sure that the first non-dash character isn't a symbol, and munch the
152 "---"\-* [^$symbol :] .* ;
154 -- Since the previous rules all match dashes followed by at least one
155 -- character, we also need to match a whole line filled with just dashes.
157 "--"\-* / { atEOL } ;
159 -- We need this rule since none of the other single line comment rules
160 -- actually match this case.
164 -- 'bol' state: beginning of a line. Slurp up all the whitespace (including
165 -- blank lines) until we find a non-whitespace character, then do layout
168 -- One slight wibble here: what if the line begins with {-#? In
169 -- theory, we have to lex the pragma to see if it's one we recognise,
170 -- and if it is, then we backtrack and do_bol, otherwise we treat it
171 -- as a nested comment. We don't bother with this: if the line begins
172 -- with {-#, then we'll assume it's a pragma we know about and go for do_bol.
175 ^\# (line)? { begin line_prag1 }
176 ^\# pragma .* \n ; -- GCC 3.3 CPP generated, apparently
177 ^\# \! .* \n ; -- #!, for scripts
181 -- after a layout keyword (let, where, do, of), we begin a new layout
182 -- context if the curly brace is missing.
183 -- Careful! This stuff is quite delicate.
184 <layout, layout_do> {
185 \{ / { notFollowedBy '-' } { pop_and open_brace }
186 -- we might encounter {-# here, but {- has been handled already
188 ^\# (line)? { begin line_prag1 }
191 -- do is treated in a subtly different way, see new_layout_context
192 <layout> () { new_layout_context True }
193 <layout_do> () { new_layout_context False }
195 -- after a new layout context which was found to be to the left of the
196 -- previous context, we have generated a '{' token, and we now need to
197 -- generate a matching '}' token.
198 <layout_left> () { do_layout_left }
200 <0,option_prags,glaexts> \n { begin bol }
202 "{-#" $whitechar* (line|LINE) { begin line_prag2 }
204 -- single-line line pragmas, of the form
205 -- # <line> "<file>" <extra-stuff> \n
206 <line_prag1> $decdigit+ { setLine line_prag1a }
207 <line_prag1a> \" [$graphic \ ]* \" { setFile line_prag1b }
208 <line_prag1b> .* { pop }
210 -- Haskell-style line pragmas, of the form
211 -- {-# LINE <line> "<file>" #-}
212 <line_prag2> $decdigit+ { setLine line_prag2a }
213 <line_prag2a> \" [$graphic \ ]* \" { setFile line_prag2b }
214 <line_prag2b> "#-}"|"-}" { pop }
215 -- NOTE: accept -} at the end of a LINE pragma, for compatibility
216 -- with older versions of GHC which generated these.
218 -- We only want RULES pragmas to be picked up when -fglasgow-exts
219 -- is on, because the contents of the pragma is always written using
220 -- glasgow-exts syntax (using forall etc.), so if glasgow exts are not
221 -- enabled, we're sure to get a parse error.
222 -- (ToDo: we should really emit a warning when ignoring pragmas)
224 "{-#" $whitechar* (RULES|rules) { token ITrules_prag }
226 <0,option_prags,glaexts> {
227 "{-#" $whitechar* (INLINE|inline) { token (ITinline_prag True) }
228 "{-#" $whitechar* (NO(T?)INLINE|no(t?)inline)
229 { token (ITinline_prag False) }
230 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
231 { token ITspec_prag }
232 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
233 $whitechar* (INLINE|inline) { token (ITspec_inline_prag True) }
234 "{-#" $whitechar* (SPECIALI[SZ]E|speciali[sz]e)
235 $whitechar* (NO(T?)INLINE|no(t?)inline)
236 { token (ITspec_inline_prag False) }
237 "{-#" $whitechar* (SOURCE|source) { token ITsource_prag }
238 "{-#" $whitechar* (DEPRECATED|deprecated)
239 { token ITdeprecated_prag }
240 "{-#" $whitechar* (SCC|scc) { token ITscc_prag }
241 "{-#" $whitechar* (GENERATED|generated)
242 { token ITgenerated_prag }
243 "{-#" $whitechar* (CORE|core) { token ITcore_prag }
244 "{-#" $whitechar* (UNPACK|unpack) { token ITunpack_prag }
246 "{-#" $whitechar* (DOCOPTIONS|docoptions)
247 / { ifExtension haddockEnabled } { lex_string_prag ITdocOptions }
249 "{-#" { nested_comment lexToken }
251 -- ToDo: should only be valid inside a pragma:
252 "#-}" { token ITclose_prag}
256 "{-#" $whitechar* (OPTIONS|options) { lex_string_prag IToptions_prag }
257 "{-#" $whitechar* (OPTIONS_GHC|options_ghc)
258 { lex_string_prag IToptions_prag }
259 "{-#" $whitechar* (LANGUAGE|language) { token ITlanguage_prag }
260 "{-#" $whitechar* (INCLUDE|include) { lex_string_prag ITinclude_prag }
263 <0,option_prags,glaexts> {
264 -- This is to catch things like {-# OPTIONS OPTIONS_HUGS ...
265 "{-#" $whitechar* $idchar+ { nested_comment lexToken }
268 -- '0' state: ordinary lexemes
269 -- 'glaexts' state: glasgow extensions (postfix '#', etc.)
274 "-- " $docsym / { ifExtension haddockEnabled } { multiline_doc_comment }
275 "{-" \ ? $docsym / { ifExtension haddockEnabled } { nested_doc_comment }
281 "[:" / { ifExtension parrEnabled } { token ITopabrack }
282 ":]" / { ifExtension parrEnabled } { token ITcpabrack }
286 "[|" / { ifExtension thEnabled } { token ITopenExpQuote }
287 "[e|" / { ifExtension thEnabled } { token ITopenExpQuote }
288 "[p|" / { ifExtension thEnabled } { token ITopenPatQuote }
289 "[d|" / { ifExtension thEnabled } { layout_token ITopenDecQuote }
290 "[t|" / { ifExtension thEnabled } { token ITopenTypQuote }
291 "|]" / { ifExtension thEnabled } { token ITcloseQuote }
292 \$ @varid / { ifExtension thEnabled } { skip_one_varid ITidEscape }
293 "$(" / { ifExtension thEnabled } { token ITparenEscape }
297 "(|" / { ifExtension arrowsEnabled `alexAndPred` notFollowedBySymbol }
298 { special IToparenbar }
299 "|)" / { ifExtension arrowsEnabled } { special ITcparenbar }
303 \? @varid / { ifExtension ipEnabled } { skip_one_varid ITdupipvarid }
307 "(#" / { notFollowedBySymbol } { token IToubxparen }
308 "#)" { token ITcubxparen }
309 "{|" { token ITocurlybar }
310 "|}" { token ITccurlybar }
313 <0,option_prags,glaexts> {
314 \( { special IToparen }
315 \) { special ITcparen }
316 \[ { special ITobrack }
317 \] { special ITcbrack }
318 \, { special ITcomma }
319 \; { special ITsemi }
320 \` { special ITbackquote }
326 <0,option_prags,glaexts> {
327 @qual @varid { check_qvarid }
328 @qual @conid { idtoken qconid }
330 @conid { idtoken conid }
333 -- after an illegal qvarid, such as 'M.let',
334 -- we back up and try again in the bad_qvarid state:
336 @conid { pop_and (idtoken conid) }
337 @qual @conid { pop_and (idtoken qconid) }
341 @qual @varid "#"+ { idtoken qvarid }
342 @qual @conid "#"+ { idtoken qconid }
343 @varid "#"+ { varid }
344 @conid "#"+ { idtoken conid }
350 @qual @varsym { idtoken qvarsym }
351 @qual @consym { idtoken qconsym }
357 @decimal { tok_decimal }
358 0[oO] @octal { tok_octal }
359 0[xX] @hexadecimal { tok_hexadecimal }
363 @decimal \# { prim_decimal }
364 0[oO] @octal \# { prim_octal }
365 0[xX] @hexadecimal \# { prim_hexadecimal }
368 <0,glaexts> @floating_point { strtoken tok_float }
369 <glaexts> @floating_point \# { init_strtoken 1 prim_float }
370 <glaexts> @floating_point \# \# { init_strtoken 2 prim_double }
372 -- Strings and chars are lexed by hand-written code. The reason is
373 -- that even if we recognise the string or char here in the regex
374 -- lexer, we would still have to parse the string afterward in order
375 -- to convert it to a String.
378 \" { lex_string_tok }
382 -- -----------------------------------------------------------------------------
386 = ITas -- Haskell keywords
411 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
413 | ITforall -- GHC extension keywords
428 | ITinline_prag Bool -- True <=> INLINE, False <=> NOINLINE
429 | ITspec_prag -- SPECIALISE
430 | ITspec_inline_prag Bool -- SPECIALISE INLINE (or NOINLINE)
437 | ITcore_prag -- hdaume: core annotations
440 | IToptions_prag String
441 | ITinclude_prag String
444 | ITdotdot -- reserved symbols
460 | ITbiglam -- GHC-extension symbols
462 | ITocurly -- special symbols
464 | ITocurlybar -- {|, for type applications
465 | ITccurlybar -- |}, for type applications
469 | ITopabrack -- [:, for parallel arrays with -fparr
470 | ITcpabrack -- :], for parallel arrays with -fparr
481 | ITvarid FastString -- identifiers
483 | ITvarsym FastString
484 | ITconsym FastString
485 | ITqvarid (FastString,FastString)
486 | ITqconid (FastString,FastString)
487 | ITqvarsym (FastString,FastString)
488 | ITqconsym (FastString,FastString)
490 | ITdupipvarid FastString -- GHC extension: implicit param: ?x
492 | ITpragma StringBuffer
495 | ITstring FastString
497 | ITrational Rational
500 | ITprimstring FastString
502 | ITprimfloat Rational
503 | ITprimdouble Rational
505 -- MetaHaskell extension tokens
506 | ITopenExpQuote -- [| or [e|
507 | ITopenPatQuote -- [p|
508 | ITopenDecQuote -- [d|
509 | ITopenTypQuote -- [t|
511 | ITidEscape FastString -- $x
512 | ITparenEscape -- $(
516 -- Arrow notation extension
523 | ITLarrowtail -- -<<
524 | ITRarrowtail -- >>-
526 | ITunknown String -- Used when the lexer can't make sense of it
527 | ITeof -- end of file token
529 -- Documentation annotations
530 | ITdocCommentNext String -- something beginning '-- |'
531 | ITdocCommentPrev String -- something beginning '-- ^'
532 | ITdocCommentNamed String -- something beginning '-- $'
533 | ITdocSection Int String -- a section heading
534 | ITdocOptions String -- doc options (prune, ignore-exports, etc)
537 deriving Show -- debugging
540 isSpecial :: Token -> Bool
541 -- If we see M.x, where x is a keyword, but
542 -- is special, we treat is as just plain M.x,
544 isSpecial ITas = True
545 isSpecial IThiding = True
546 isSpecial ITderive = True
547 isSpecial ITqualified = True
548 isSpecial ITforall = True
549 isSpecial ITexport = True
550 isSpecial ITlabel = True
551 isSpecial ITdynamic = True
552 isSpecial ITsafe = True
553 isSpecial ITthreadsafe = True
554 isSpecial ITunsafe = True
555 isSpecial ITccallconv = True
556 isSpecial ITstdcallconv = True
557 isSpecial ITmdo = True
558 isSpecial ITfamily = True
561 -- the bitmap provided as the third component indicates whether the
562 -- corresponding extension keyword is valid under the extension options
563 -- provided to the compiler; if the extension corresponding to *any* of the
564 -- bits set in the bitmap is enabled, the keyword is valid (this setup
565 -- facilitates using a keyword in two different extensions that can be
566 -- activated independently)
568 reservedWordsFM = listToUFM $
569 map (\(x, y, z) -> (mkFastString x, (y, z)))
570 [( "_", ITunderscore, 0 ),
572 ( "case", ITcase, 0 ),
573 ( "class", ITclass, 0 ),
574 ( "data", ITdata, 0 ),
575 ( "default", ITdefault, 0 ),
576 ( "deriving", ITderiving, 0 ),
577 ( "derive", ITderive, 0 ),
579 ( "else", ITelse, 0 ),
580 ( "hiding", IThiding, 0 ),
582 ( "import", ITimport, 0 ),
584 ( "infix", ITinfix, 0 ),
585 ( "infixl", ITinfixl, 0 ),
586 ( "infixr", ITinfixr, 0 ),
587 ( "instance", ITinstance, 0 ),
589 ( "module", ITmodule, 0 ),
590 ( "newtype", ITnewtype, 0 ),
592 ( "qualified", ITqualified, 0 ),
593 ( "then", ITthen, 0 ),
594 ( "type", ITtype, 0 ),
595 ( "where", ITwhere, 0 ),
596 ( "_scc_", ITscc, 0 ), -- ToDo: remove
598 ( "forall", ITforall, bit tvBit),
599 ( "mdo", ITmdo, bit glaExtsBit),
600 ( "family", ITfamily, bit idxTysBit),
602 ( "foreign", ITforeign, bit ffiBit),
603 ( "export", ITexport, bit ffiBit),
604 ( "label", ITlabel, bit ffiBit),
605 ( "dynamic", ITdynamic, bit ffiBit),
606 ( "safe", ITsafe, bit ffiBit),
607 ( "threadsafe", ITthreadsafe, bit ffiBit),
608 ( "unsafe", ITunsafe, bit ffiBit),
609 ( "stdcall", ITstdcallconv, bit ffiBit),
610 ( "ccall", ITccallconv, bit ffiBit),
611 ( "dotnet", ITdotnet, bit ffiBit),
613 ( "rec", ITrec, bit arrowsBit),
614 ( "proc", ITproc, bit arrowsBit)
617 reservedSymsFM = listToUFM $
618 map (\ (x,y,z) -> (mkFastString x,(y,z)))
619 [ ("..", ITdotdot, 0)
620 ,(":", ITcolon, 0) -- (:) is a reserved op,
621 -- meaning only list cons
634 ,("*", ITstar, bit glaExtsBit .|.
635 bit idxTysBit) -- For data T (a::*) = MkT
636 ,(".", ITdot, bit tvBit) -- For 'forall a . t'
638 ,("-<", ITlarrowtail, bit arrowsBit)
639 ,(">-", ITrarrowtail, bit arrowsBit)
640 ,("-<<", ITLarrowtail, bit arrowsBit)
641 ,(">>-", ITRarrowtail, bit arrowsBit)
643 #if __GLASGOW_HASKELL__ >= 605
644 ,("∷", ITdcolon, bit glaExtsBit)
645 ,("⇒", ITdarrow, bit glaExtsBit)
646 ,("∀", ITforall, bit glaExtsBit)
647 ,("→", ITrarrow, bit glaExtsBit)
648 ,("←", ITlarrow, bit glaExtsBit)
649 ,("⋯", ITdotdot, bit glaExtsBit)
650 -- ToDo: ideally, → and ∷ should be "specials", so that they cannot
651 -- form part of a large operator. This would let us have a better
652 -- syntax for kinds: ɑ∷*→* would be a legal kind signature. (maybe).
656 -- -----------------------------------------------------------------------------
659 type Action = SrcSpan -> StringBuffer -> Int -> P (Located Token)
661 special :: Token -> Action
662 special tok span _buf len = return (L span tok)
664 token, layout_token :: Token -> Action
665 token t span buf len = return (L span t)
666 layout_token t span buf len = pushLexState layout >> return (L span t)
668 idtoken :: (StringBuffer -> Int -> Token) -> Action
669 idtoken f span buf len = return (L span $! (f buf len))
671 skip_one_varid :: (FastString -> Token) -> Action
672 skip_one_varid f span buf len
673 = return (L span $! f (lexemeToFastString (stepOn buf) (len-1)))
675 strtoken :: (String -> Token) -> Action
676 strtoken f span buf len =
677 return (L span $! (f $! lexemeToString buf len))
679 init_strtoken :: Int -> (String -> Token) -> Action
680 -- like strtoken, but drops the last N character(s)
681 init_strtoken drop f span buf len =
682 return (L span $! (f $! lexemeToString buf (len-drop)))
684 begin :: Int -> Action
685 begin code _span _str _len = do pushLexState code; lexToken
688 pop _span _buf _len = do popLexState; lexToken
690 pop_and :: Action -> Action
691 pop_and act span buf len = do popLexState; act span buf len
693 {-# INLINE nextCharIs #-}
694 nextCharIs buf p = not (atEnd buf) && p (currentChar buf)
696 notFollowedBy char _ _ _ (AI _ _ buf)
697 = nextCharIs buf (/=char)
699 notFollowedBySymbol _ _ _ (AI _ _ buf)
700 = nextCharIs buf (`notElem` "!#$%&*+./<=>?@\\^|-~")
702 -- We must reject doc comments as being ordinary comments everywhere.
703 -- In some cases the doc comment will be selected as the lexeme due to
704 -- maximal munch, but not always, because the nested comment rule is
705 -- valid in all states, but the doc-comment rules are only valid in
706 -- the non-layout states.
707 isNormalComment bits _ _ (AI _ _ buf)
708 | haddockEnabled bits = notFollowedByDocOrPragma
709 | otherwise = nextCharIs buf (/='#')
711 notFollowedByDocOrPragma
712 = not $ spaceAndP buf (`nextCharIs` (`elem` "|^*$#"))
714 spaceAndP buf p = p buf || nextCharIs buf (==' ') && p (snd (nextChar buf))
716 haddockDisabledAnd p bits _ _ (AI _ _ buf)
717 = if haddockEnabled bits then False else (p buf)
719 atEOL _ _ _ (AI _ _ buf) = atEnd buf || currentChar buf == '\n'
721 ifExtension pred bits _ _ _ = pred bits
723 multiline_doc_comment :: Action
724 multiline_doc_comment span buf _len = withLexedDocType (worker "")
726 worker commentAcc input docType oneLine = case alexGetChar input of
728 | oneLine -> docCommentEnd input commentAcc docType buf span
729 | otherwise -> case checkIfCommentLine input' of
730 Just input -> worker ('\n':commentAcc) input docType False
731 Nothing -> docCommentEnd input commentAcc docType buf span
732 Just (c, input) -> worker (c:commentAcc) input docType oneLine
733 Nothing -> docCommentEnd input commentAcc docType buf span
735 checkIfCommentLine input = check (dropNonNewlineSpace input)
737 check input = case alexGetChar input of
738 Just ('-', input) -> case alexGetChar input of
739 Just ('-', input) -> case alexGetChar input of
740 Just (c, _) | c /= '-' -> Just input
745 dropNonNewlineSpace input = case alexGetChar input of
747 | isSpace c && c /= '\n' -> dropNonNewlineSpace input'
752 nested comments require traversing by hand, they can't be parsed
753 using regular expressions.
755 nested_comment :: P (Located Token) -> Action
756 nested_comment cont span _str _len = do
760 go 0 input = do setInput input; cont
761 go n input = case alexGetChar input of
762 Nothing -> errBrace input span
763 Just ('-',input) -> case alexGetChar input of
764 Nothing -> errBrace input span
765 Just ('\125',input) -> go (n-1) input
766 Just (c,_) -> go n input
767 Just ('\123',input) -> case alexGetChar input of
768 Nothing -> errBrace input span
769 Just ('-',input) -> go (n+1) input
770 Just (c,_) -> go n input
771 Just (c,input) -> go n input
773 nested_doc_comment :: Action
774 nested_doc_comment span buf _len = withLexedDocType (go "")
776 go commentAcc input docType _ = case alexGetChar input of
777 Nothing -> errBrace input span
778 Just ('-',input) -> case alexGetChar input of
779 Nothing -> errBrace input span
780 Just ('\125',input@(AI end _ buf2)) ->
781 docCommentEnd input commentAcc docType buf span
782 Just (c,_) -> go ('-':commentAcc) input docType False
783 Just ('\123', input) -> case alexGetChar input of
784 Nothing -> errBrace input span
785 Just ('-',input) -> do
787 let cont = do input <- getInput; go commentAcc input docType False
788 nested_comment cont span buf _len
789 Just (c,_) -> go ('\123':commentAcc) input docType False
790 Just (c,input) -> go (c:commentAcc) input docType False
792 withLexedDocType lexDocComment = do
793 input@(AI _ _ buf) <- getInput
794 case prevChar buf ' ' of
795 '|' -> lexDocComment input ITdocCommentNext False
796 '^' -> lexDocComment input ITdocCommentPrev False
797 '$' -> lexDocComment input ITdocCommentNamed False
798 '*' -> lexDocSection 1 input
800 lexDocSection n input = case alexGetChar input of
801 Just ('*', input) -> lexDocSection (n+1) input
802 Just (c, _) -> lexDocComment input (ITdocSection n) True
803 Nothing -> do setInput input; lexToken -- eof reached, lex it normally
806 -------------------------------------------------------------------------------
807 -- This function is quite tricky. We can't just return a new token, we also
808 -- need to update the state of the parser. Why? Because the token is longer
809 -- than what was lexed by Alex, and the lexToken function doesn't know this, so
810 -- it writes the wrong token length to the parser state. This function is
811 -- called afterwards, so it can just update the state.
813 -- This is complicated by the fact that Haddock tokens can span multiple lines,
814 -- which is something that the original lexer didn't account for.
815 -- I have added last_line_len in the parser state which represents the length
816 -- of the part of the token that is on the last line. It is now used for layout
817 -- calculation in pushCurrentContext instead of last_len. last_len is, like it
818 -- was before, the full length of the token, and it is now only used for error
821 docCommentEnd :: AlexInput -> String -> (String -> Token) -> StringBuffer ->
822 SrcSpan -> P (Located Token)
823 docCommentEnd input commentAcc docType buf span = do
825 let (AI loc last_offs nextBuf) = input
826 comment = reverse commentAcc
827 span' = mkSrcSpan (srcSpanStart span) loc
828 last_len = byteDiff buf nextBuf
830 last_line_len = if (last_offs - last_len < 0)
834 span `seq` setLastToken span' last_len last_line_len
835 return (L span' (docType comment))
837 errBrace (AI end _ _) span = failLocMsgP (srcSpanStart span) end "unterminated `{-'"
839 open_brace, close_brace :: Action
840 open_brace span _str _len = do
842 setContext (NoLayout:ctx)
843 return (L span ITocurly)
844 close_brace span _str _len = do
846 return (L span ITccurly)
848 -- We have to be careful not to count M.<varid> as a qualified name
849 -- when <varid> is a keyword. We hack around this by catching
850 -- the offending tokens afterward, and re-lexing in a different state.
851 check_qvarid span buf len = do
852 case lookupUFM reservedWordsFM var of
854 | not (isSpecial keyword) ->
858 b <- extension (\i -> exts .&. i /= 0)
861 _other -> return token
863 (mod,var) = splitQualName buf len
864 token = L span (ITqvarid (mod,var))
867 (AI _ offs _) <- getInput
868 setInput (AI (srcSpanStart span) (offs-len) buf)
869 pushLexState bad_qvarid
872 qvarid buf len = ITqvarid $! splitQualName buf len
873 qconid buf len = ITqconid $! splitQualName buf len
875 splitQualName :: StringBuffer -> Int -> (FastString,FastString)
876 -- takes a StringBuffer and a length, and returns the module name
877 -- and identifier parts of a qualified name. Splits at the *last* dot,
878 -- because of hierarchical module names.
879 splitQualName orig_buf len = split orig_buf orig_buf
882 | orig_buf `byteDiff` buf >= len = done dot_buf
883 | c == '.' = found_dot buf'
884 | otherwise = split buf' dot_buf
886 (c,buf') = nextChar buf
888 -- careful, we might get names like M....
889 -- so, if the character after the dot is not upper-case, this is
890 -- the end of the qualifier part.
891 found_dot buf -- buf points after the '.'
892 | isUpper c = split buf' buf
893 | otherwise = done buf
895 (c,buf') = nextChar buf
898 (lexemeToFastString orig_buf (qual_size - 1),
899 lexemeToFastString dot_buf (len - qual_size))
901 qual_size = orig_buf `byteDiff` dot_buf
904 case lookupUFM reservedWordsFM fs of
905 Just (keyword,0) -> do
907 return (L span keyword)
908 Just (keyword,exts) -> do
909 b <- extension (\i -> exts .&. i /= 0)
910 if b then do maybe_layout keyword
911 return (L span keyword)
912 else return (L span (ITvarid fs))
913 _other -> return (L span (ITvarid fs))
915 fs = lexemeToFastString buf len
917 conid buf len = ITconid fs
918 where fs = lexemeToFastString buf len
920 qvarsym buf len = ITqvarsym $! splitQualName buf len
921 qconsym buf len = ITqconsym $! splitQualName buf len
923 varsym = sym ITvarsym
924 consym = sym ITconsym
926 sym con span buf len =
927 case lookupUFM reservedSymsFM fs of
928 Just (keyword,0) -> return (L span keyword)
929 Just (keyword,exts) -> do
930 b <- extension (\i -> exts .&. i /= 0)
931 if b then return (L span keyword)
932 else return (L span $! con fs)
933 _other -> return (L span $! con fs)
935 fs = lexemeToFastString buf len
937 tok_decimal span buf len
938 = return (L span (ITinteger $! parseInteger buf len 10 octDecDigit))
940 tok_octal span buf len
941 = return (L span (ITinteger $! parseInteger (offsetBytes 2 buf) (len-2) 8 octDecDigit))
943 tok_hexadecimal span buf len
944 = return (L span (ITinteger $! parseInteger (offsetBytes 2 buf) (len-2) 16 hexDigit))
946 prim_decimal span buf len
947 = return (L span (ITprimint $! parseInteger buf (len-1) 10 octDecDigit))
949 prim_octal span buf len
950 = return (L span (ITprimint $! parseInteger (offsetBytes 2 buf) (len-3) 8 octDecDigit))
952 prim_hexadecimal span buf len
953 = return (L span (ITprimint $! parseInteger (offsetBytes 2 buf) (len-3) 16 hexDigit))
955 tok_float str = ITrational $! readRational str
956 prim_float str = ITprimfloat $! readRational str
957 prim_double str = ITprimdouble $! readRational str
959 -- -----------------------------------------------------------------------------
962 -- we're at the first token on a line, insert layout tokens if necessary
964 do_bol span _str _len = do
968 --trace "layout: inserting '}'" $ do
970 -- do NOT pop the lex state, we might have a ';' to insert
971 return (L span ITvccurly)
973 --trace "layout: inserting ';'" $ do
975 return (L span ITsemi)
980 -- certain keywords put us in the "layout" state, where we might
981 -- add an opening curly brace.
982 maybe_layout ITdo = pushLexState layout_do
983 maybe_layout ITmdo = pushLexState layout_do
984 maybe_layout ITof = pushLexState layout
985 maybe_layout ITlet = pushLexState layout
986 maybe_layout ITwhere = pushLexState layout
987 maybe_layout ITrec = pushLexState layout
988 maybe_layout _ = return ()
990 -- Pushing a new implicit layout context. If the indentation of the
991 -- next token is not greater than the previous layout context, then
992 -- Haskell 98 says that the new layout context should be empty; that is
993 -- the lexer must generate {}.
995 -- We are slightly more lenient than this: when the new context is started
996 -- by a 'do', then we allow the new context to be at the same indentation as
997 -- the previous context. This is what the 'strict' argument is for.
999 new_layout_context strict span _buf _len = do
1001 (AI _ offset _) <- getInput
1004 Layout prev_off : _ |
1005 (strict && prev_off >= offset ||
1006 not strict && prev_off > offset) -> do
1007 -- token is indented to the left of the previous context.
1008 -- we must generate a {} sequence now.
1009 pushLexState layout_left
1010 return (L span ITvocurly)
1012 setContext (Layout offset : ctx)
1013 return (L span ITvocurly)
1015 do_layout_left span _buf _len = do
1017 pushLexState bol -- we must be at the start of a line
1018 return (L span ITvccurly)
1020 -- -----------------------------------------------------------------------------
1023 setLine :: Int -> Action
1024 setLine code span buf len = do
1025 let line = parseInteger buf len 10 octDecDigit
1026 setSrcLoc (mkSrcLoc (srcSpanFile span) (fromIntegral line - 1) 0)
1027 -- subtract one: the line number refers to the *following* line
1032 setFile :: Int -> Action
1033 setFile code span buf len = do
1034 let file = lexemeToFastString (stepOn buf) (len-2)
1035 setSrcLoc (mkSrcLoc file (srcSpanEndLine span) (srcSpanEndCol span))
1041 -- -----------------------------------------------------------------------------
1042 -- Options, includes and language pragmas.
1044 lex_string_prag :: (String -> Token) -> Action
1045 lex_string_prag mkTok span buf len
1046 = do input <- getInput
1050 return (L (mkSrcSpan start end) tok)
1052 = if isString input "#-}"
1053 then do setInput input
1054 return (mkTok (reverse acc))
1055 else case alexGetChar input of
1056 Just (c,i) -> go (c:acc) i
1057 Nothing -> err input
1058 isString i [] = True
1060 = case alexGetChar i of
1061 Just (c,i') | c == x -> isString i' xs
1063 err (AI end _ _) = failLocMsgP (srcSpanStart span) end "unterminated options pragma"
1066 -- -----------------------------------------------------------------------------
1069 -- This stuff is horrible. I hates it.
1071 lex_string_tok :: Action
1072 lex_string_tok span buf len = do
1073 tok <- lex_string ""
1075 return (L (mkSrcSpan (srcSpanStart span) end) tok)
1077 lex_string :: String -> P Token
1080 case alexGetChar' i of
1081 Nothing -> lit_error
1085 glaexts <- extension glaExtsEnabled
1089 case alexGetChar' i of
1093 then failMsgP "primitive string literal must contain only characters <= \'\\xFF\'"
1094 else let s' = mkZFastString (reverse s) in
1095 return (ITprimstring s')
1096 -- mkZFastString is a hack to avoid encoding the
1097 -- string in UTF-8. We just want the exact bytes.
1099 return (ITstring (mkFastString (reverse s)))
1101 return (ITstring (mkFastString (reverse s)))
1104 | Just ('&',i) <- next -> do
1105 setInput i; lex_string s
1106 | Just (c,i) <- next, is_space c -> do
1107 setInput i; lex_stringgap s
1108 where next = alexGetChar' i
1114 lex_stringgap s = do
1117 '\\' -> lex_string s
1118 c | is_space c -> lex_stringgap s
1122 lex_char_tok :: Action
1123 -- Here we are basically parsing character literals, such as 'x' or '\n'
1124 -- but, when Template Haskell is on, we additionally spot
1125 -- 'x and ''T, returning ITvarQuote and ITtyQuote respectively,
1126 -- but WIHTOUT CONSUMING the x or T part (the parser does that).
1127 -- So we have to do two characters of lookahead: when we see 'x we need to
1128 -- see if there's a trailing quote
1129 lex_char_tok span buf len = do -- We've seen '
1130 i1 <- getInput -- Look ahead to first character
1131 let loc = srcSpanStart span
1132 case alexGetChar' i1 of
1133 Nothing -> lit_error
1135 Just ('\'', i2@(AI end2 _ _)) -> do -- We've seen ''
1136 th_exts <- extension thEnabled
1139 return (L (mkSrcSpan loc end2) ITtyQuote)
1142 Just ('\\', i2@(AI end2 _ _)) -> do -- We've seen 'backslash
1144 lit_ch <- lex_escape
1145 mc <- getCharOrFail -- Trailing quote
1146 if mc == '\'' then finish_char_tok loc lit_ch
1147 else do setInput i2; lit_error
1149 Just (c, i2@(AI end2 _ _))
1150 | not (isAny c) -> lit_error
1153 -- We've seen 'x, where x is a valid character
1154 -- (i.e. not newline etc) but not a quote or backslash
1155 case alexGetChar' i2 of -- Look ahead one more character
1156 Nothing -> lit_error
1157 Just ('\'', i3) -> do -- We've seen 'x'
1159 finish_char_tok loc c
1160 _other -> do -- We've seen 'x not followed by quote
1161 -- If TH is on, just parse the quote only
1162 th_exts <- extension thEnabled
1163 let (AI end _ _) = i1
1164 if th_exts then return (L (mkSrcSpan loc end) ITvarQuote)
1165 else do setInput i2; lit_error
1167 finish_char_tok :: SrcLoc -> Char -> P (Located Token)
1168 finish_char_tok loc ch -- We've already seen the closing quote
1169 -- Just need to check for trailing #
1170 = do glaexts <- extension glaExtsEnabled
1171 i@(AI end _ _) <- getInput
1173 case alexGetChar' i of
1174 Just ('#',i@(AI end _ _)) -> do
1176 return (L (mkSrcSpan loc end) (ITprimchar ch))
1178 return (L (mkSrcSpan loc end) (ITchar ch))
1180 return (L (mkSrcSpan loc end) (ITchar ch))
1182 lex_char :: Char -> AlexInput -> P Char
1185 '\\' -> do setInput inp; lex_escape
1186 c | isAny c -> do setInput inp; return c
1189 isAny c | c > '\xff' = isPrint c
1190 | otherwise = is_any c
1192 lex_escape :: P Char
1206 '^' -> do c <- getCharOrFail
1207 if c >= '@' && c <= '_'
1208 then return (chr (ord c - ord '@'))
1211 'x' -> readNum is_hexdigit 16 hexDigit
1212 'o' -> readNum is_octdigit 8 octDecDigit
1213 x | is_digit x -> readNum2 is_digit 10 octDecDigit (octDecDigit x)
1217 case alexGetChar' i of
1218 Nothing -> lit_error
1220 case alexGetChar' i2 of
1221 Nothing -> do setInput i2; lit_error
1223 let str = [c1,c2,c3] in
1224 case [ (c,rest) | (p,c) <- silly_escape_chars,
1225 Just rest <- [maybePrefixMatch p str] ] of
1226 (escape_char,[]):_ -> do
1229 (escape_char,_:_):_ -> do
1234 readNum :: (Char -> Bool) -> Int -> (Char -> Int) -> P Char
1235 readNum is_digit base conv = do
1239 then readNum2 is_digit base conv (conv c)
1240 else do setInput i; lit_error
1242 readNum2 is_digit base conv i = do
1245 where read i input = do
1246 case alexGetChar' input of
1247 Just (c,input') | is_digit c -> do
1248 read (i*base + conv c) input'
1250 if i >= 0 && i <= 0x10FFFF
1251 then do setInput input; return (chr i)
1254 silly_escape_chars = [
1291 -- before calling lit_error, ensure that the current input is pointing to
1292 -- the position of the error in the buffer. This is so that we can report
1293 -- a correct location to the user, but also so we can detect UTF-8 decoding
1294 -- errors if they occur.
1295 lit_error = lexError "lexical error in string/character literal"
1297 getCharOrFail :: P Char
1300 case alexGetChar' i of
1301 Nothing -> lexError "unexpected end-of-file in string/character literal"
1302 Just (c,i) -> do setInput i; return c
1304 -- -----------------------------------------------------------------------------
1307 warn :: DynFlag -> SDoc -> Action
1308 warn option warning span _buf _len = do
1309 addWarning option (mkWarnMsg span alwaysQualify warning)
1312 -- -----------------------------------------------------------------------------
1323 SrcSpan -- The start and end of the text span related to
1324 -- the error. Might be used in environments which can
1325 -- show this span, e.g. by highlighting it.
1326 Message -- The error message
1328 data PState = PState {
1329 buffer :: StringBuffer,
1331 messages :: Messages,
1332 last_loc :: SrcSpan, -- pos of previous token
1333 last_offs :: !Int, -- offset of the previous token from the
1334 -- beginning of the current line.
1335 -- \t is equal to 8 spaces.
1336 last_len :: !Int, -- len of previous token
1337 last_line_len :: !Int,
1338 loc :: SrcLoc, -- current loc (end of prev token + 1)
1339 extsBitmap :: !Int, -- bitmap that determines permitted extensions
1340 context :: [LayoutContext],
1343 -- last_loc and last_len are used when generating error messages,
1344 -- and in pushCurrentContext only. Sigh, if only Happy passed the
1345 -- current token to happyError, we could at least get rid of last_len.
1346 -- Getting rid of last_loc would require finding another way to
1347 -- implement pushCurrentContext (which is only called from one place).
1349 newtype P a = P { unP :: PState -> ParseResult a }
1351 instance Monad P where
1357 returnP a = P $ \s -> POk s a
1359 thenP :: P a -> (a -> P b) -> P b
1360 (P m) `thenP` k = P $ \ s ->
1362 POk s1 a -> (unP (k a)) s1
1363 PFailed span err -> PFailed span err
1365 failP :: String -> P a
1366 failP msg = P $ \s -> PFailed (last_loc s) (text msg)
1368 failMsgP :: String -> P a
1369 failMsgP msg = P $ \s -> PFailed (last_loc s) (text msg)
1371 failLocMsgP :: SrcLoc -> SrcLoc -> String -> P a
1372 failLocMsgP loc1 loc2 str = P $ \s -> PFailed (mkSrcSpan loc1 loc2) (text str)
1374 failSpanMsgP :: SrcSpan -> String -> P a
1375 failSpanMsgP span msg = P $ \s -> PFailed span (text msg)
1377 extension :: (Int -> Bool) -> P Bool
1378 extension p = P $ \s -> POk s (p $! extsBitmap s)
1381 getExts = P $ \s -> POk s (extsBitmap s)
1383 setSrcLoc :: SrcLoc -> P ()
1384 setSrcLoc new_loc = P $ \s -> POk s{loc=new_loc} ()
1386 getSrcLoc :: P SrcLoc
1387 getSrcLoc = P $ \s@(PState{ loc=loc }) -> POk s loc
1389 setLastToken :: SrcSpan -> Int -> Int -> P ()
1390 setLastToken loc len line_len = P $ \s -> POk s {
1393 last_line_len=line_len
1396 data AlexInput = AI SrcLoc {-#UNPACK#-}!Int StringBuffer
1398 alexInputPrevChar :: AlexInput -> Char
1399 alexInputPrevChar (AI _ _ buf) = prevChar buf '\n'
1401 alexGetChar :: AlexInput -> Maybe (Char,AlexInput)
1402 alexGetChar (AI loc ofs s)
1404 | otherwise = adj_c `seq` loc' `seq` ofs' `seq` s' `seq`
1405 --trace (show (ord c)) $
1406 Just (adj_c, (AI loc' ofs' s'))
1407 where (c,s') = nextChar s
1408 loc' = advanceSrcLoc loc c
1409 ofs' = advanceOffs c ofs
1417 other_graphic = '\x6'
1420 | c <= '\x06' = non_graphic
1422 -- Alex doesn't handle Unicode, so when Unicode
1423 -- character is encoutered we output these values
1424 -- with the actual character value hidden in the state.
1426 case generalCategory c of
1427 UppercaseLetter -> upper
1428 LowercaseLetter -> lower
1429 TitlecaseLetter -> upper
1430 ModifierLetter -> other_graphic
1431 OtherLetter -> other_graphic
1432 NonSpacingMark -> other_graphic
1433 SpacingCombiningMark -> other_graphic
1434 EnclosingMark -> other_graphic
1435 DecimalNumber -> digit
1436 LetterNumber -> other_graphic
1437 OtherNumber -> other_graphic
1438 ConnectorPunctuation -> other_graphic
1439 DashPunctuation -> other_graphic
1440 OpenPunctuation -> other_graphic
1441 ClosePunctuation -> other_graphic
1442 InitialQuote -> other_graphic
1443 FinalQuote -> other_graphic
1444 OtherPunctuation -> other_graphic
1445 MathSymbol -> symbol
1446 CurrencySymbol -> symbol
1447 ModifierSymbol -> symbol
1448 OtherSymbol -> symbol
1450 _other -> non_graphic
1452 -- This version does not squash unicode characters, it is used when
1454 alexGetChar' :: AlexInput -> Maybe (Char,AlexInput)
1455 alexGetChar' (AI loc ofs s)
1457 | otherwise = c `seq` loc' `seq` ofs' `seq` s' `seq`
1458 --trace (show (ord c)) $
1459 Just (c, (AI loc' ofs' s'))
1460 where (c,s') = nextChar s
1461 loc' = advanceSrcLoc loc c
1462 ofs' = advanceOffs c ofs
1464 advanceOffs :: Char -> Int -> Int
1465 advanceOffs '\n' offs = 0
1466 advanceOffs '\t' offs = (offs `quot` 8 + 1) * 8
1467 advanceOffs _ offs = offs + 1
1469 getInput :: P AlexInput
1470 getInput = P $ \s@PState{ loc=l, last_offs=o, buffer=b } -> POk s (AI l o b)
1472 setInput :: AlexInput -> P ()
1473 setInput (AI l o b) = P $ \s -> POk s{ loc=l, last_offs=o, buffer=b } ()
1475 pushLexState :: Int -> P ()
1476 pushLexState ls = P $ \s@PState{ lex_state=l } -> POk s{lex_state=ls:l} ()
1478 popLexState :: P Int
1479 popLexState = P $ \s@PState{ lex_state=ls:l } -> POk s{ lex_state=l } ls
1481 getLexState :: P Int
1482 getLexState = P $ \s@PState{ lex_state=ls:l } -> POk s ls
1484 -- for reasons of efficiency, flags indicating language extensions (eg,
1485 -- -fglasgow-exts or -fparr) are represented by a bitmap stored in an unboxed
1488 glaExtsBit, ffiBit, parrBit :: Int
1495 tvBit = 7 -- Scoped type variables enables 'forall' keyword
1496 bangPatBit = 8 -- Tells the parser to understand bang-patterns
1497 -- (doesn't affect the lexer)
1498 idxTysBit = 9 -- indexed type families: 'family' keyword and kind sigs
1499 haddockBit = 10 -- Lex and parse Haddock comments
1501 glaExtsEnabled, ffiEnabled, parrEnabled :: Int -> Bool
1502 glaExtsEnabled flags = testBit flags glaExtsBit
1503 ffiEnabled flags = testBit flags ffiBit
1504 parrEnabled flags = testBit flags parrBit
1505 arrowsEnabled flags = testBit flags arrowsBit
1506 thEnabled flags = testBit flags thBit
1507 ipEnabled flags = testBit flags ipBit
1508 tvEnabled flags = testBit flags tvBit
1509 bangPatEnabled flags = testBit flags bangPatBit
1510 idxTysEnabled flags = testBit flags idxTysBit
1511 haddockEnabled flags = testBit flags haddockBit
1513 -- PState for parsing options pragmas
1515 pragState :: StringBuffer -> SrcLoc -> PState
1519 messages = emptyMessages,
1520 -- XXX defaultDynFlags is not right, but we don't have a real
1522 dflags = defaultDynFlags,
1523 last_loc = mkSrcSpan loc loc,
1530 lex_state = [bol, option_prags, 0]
1534 -- create a parse state
1536 mkPState :: StringBuffer -> SrcLoc -> DynFlags -> PState
1537 mkPState buf loc flags =
1541 messages = emptyMessages,
1542 last_loc = mkSrcSpan loc loc,
1547 extsBitmap = fromIntegral bitmap,
1549 lex_state = [bol, if glaExtsEnabled bitmap then glaexts else 0]
1550 -- we begin in the layout state if toplev_layout is set
1553 bitmap = glaExtsBit `setBitIf` dopt Opt_GlasgowExts flags
1554 .|. ffiBit `setBitIf` dopt Opt_FFI flags
1555 .|. parrBit `setBitIf` dopt Opt_PArr flags
1556 .|. arrowsBit `setBitIf` dopt Opt_Arrows flags
1557 .|. thBit `setBitIf` dopt Opt_TH flags
1558 .|. ipBit `setBitIf` dopt Opt_ImplicitParams flags
1559 .|. tvBit `setBitIf` dopt Opt_ScopedTypeVariables flags
1560 .|. bangPatBit `setBitIf` dopt Opt_BangPatterns flags
1561 .|. idxTysBit `setBitIf` dopt Opt_IndexedTypes flags
1562 .|. haddockBit `setBitIf` dopt Opt_Haddock flags
1564 setBitIf :: Int -> Bool -> Int
1565 b `setBitIf` cond | cond = bit b
1568 addWarning :: DynFlag -> WarnMsg -> P ()
1570 = P $ \s@PState{messages=(ws,es), dflags=d} ->
1571 let ws' = if dopt option d then ws `snocBag` w else ws
1572 in POk s{messages=(ws', es)} ()
1574 getMessages :: PState -> Messages
1575 getMessages PState{messages=ms} = ms
1577 getContext :: P [LayoutContext]
1578 getContext = P $ \s@PState{context=ctx} -> POk s ctx
1580 setContext :: [LayoutContext] -> P ()
1581 setContext ctx = P $ \s -> POk s{context=ctx} ()
1584 popContext = P $ \ s@(PState{ buffer = buf, context = ctx,
1585 loc = loc, last_len = len, last_loc = last_loc }) ->
1587 (_:tl) -> POk s{ context = tl } ()
1588 [] -> PFailed last_loc (srcParseErr buf len)
1590 -- Push a new layout context at the indentation of the last token read.
1591 -- This is only used at the outer level of a module when the 'module'
1592 -- keyword is missing.
1593 pushCurrentContext :: P ()
1594 pushCurrentContext = P $ \ s@PState{ last_offs=offs, last_line_len=len, context=ctx } ->
1595 POk s{context = Layout (offs-len) : ctx} ()
1596 --trace ("off: " ++ show offs ++ ", len: " ++ show len) $ POk s{context = Layout (offs-len) : ctx} ()
1598 getOffside :: P Ordering
1599 getOffside = P $ \s@PState{last_offs=offs, context=stk} ->
1600 let ord = case stk of
1601 (Layout n:_) -> compare offs n
1605 -- ---------------------------------------------------------------------------
1606 -- Construct a parse error
1609 :: StringBuffer -- current buffer (placed just after the last token)
1610 -> Int -- length of the previous token
1613 = hcat [ if null token
1614 then ptext SLIT("parse error (possibly incorrect indentation)")
1615 else hcat [ptext SLIT("parse error on input "),
1616 char '`', text token, char '\'']
1618 where token = lexemeToString (offsetBytes (-len) buf) len
1620 -- Report a parse failure, giving the span of the previous token as
1621 -- the location of the error. This is the entry point for errors
1622 -- detected during parsing.
1624 srcParseFail = P $ \PState{ buffer = buf, last_len = len,
1625 last_loc = last_loc } ->
1626 PFailed last_loc (srcParseErr buf len)
1628 -- A lexical error is reported at a particular position in the source file,
1629 -- not over a token range.
1630 lexError :: String -> P a
1633 i@(AI end _ buf) <- getInput
1634 reportLexError loc end buf str
1636 -- -----------------------------------------------------------------------------
1637 -- This is the top-level function: called from the parser each time a
1638 -- new token is to be read from the input.
1640 lexer :: (Located Token -> P a) -> P a
1642 tok@(L span tok__) <- lexToken
1643 -- trace ("token: " ++ show tok__) $ do
1646 lexToken :: P (Located Token)
1648 inp@(AI loc1 _ buf) <- getInput
1651 case alexScanUser exts inp sc of
1652 AlexEOF -> do let span = mkSrcSpan loc1 loc1
1653 setLastToken span 0 0
1654 return (L span ITeof)
1655 AlexError (AI loc2 _ buf) -> do
1656 reportLexError loc1 loc2 buf "lexical error"
1657 AlexSkip inp2 _ -> do
1660 AlexToken inp2@(AI end _ buf2) len t -> do
1662 let span = mkSrcSpan loc1 end
1663 let bytes = byteDiff buf buf2
1664 span `seq` setLastToken span bytes bytes
1667 reportLexError loc1 loc2 buf str
1668 | atEnd buf = failLocMsgP loc1 loc2 (str ++ " at end of input")
1671 c = fst (nextChar buf)
1673 if c == '\0' -- decoding errors are mapped to '\0', see utf8DecodeChar#
1674 then failLocMsgP loc2 loc2 (str ++ " (UTF-8 decoding error)")
1675 else failLocMsgP loc1 loc2 (str ++ " at character " ++ show c)