2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[Lexical analysis]{Lexical analysis}
6 --------------------------------------------------------
8 There's a known bug in here:
10 If an interface file ends prematurely, Lex tries to
11 do headFS of an empty FastString.
13 An example that provokes the error is
15 f _:_ _forall_ [a] <<<END OF FILE>>>
16 --------------------------------------------------------
22 ifaceParseErr, srcParseErr,
25 Token(..), lexer, ParseResult(..), PState(..),
29 P, thenP, thenP_, returnP, mapP, failP, failMsgP,
30 getSrcLocP, setSrcLocP, getSrcFile,
31 layoutOn, layoutOff, pushContext, popContext
34 #include "HsVersions.h"
36 import Char ( isSpace, toUpper )
37 import List ( isSuffixOf )
39 import IdInfo ( InlinePragInfo(..) )
40 import PrelNames ( mkTupNameStr )
41 import CmdLineOpts ( opt_HiVersion, opt_NoHiCheck )
42 import ForeignCall ( Safety(..) )
43 import Demand ( Demand(..) {- instance Read -} )
44 import UniqFM ( listToUFM, lookupUFM )
45 import BasicTypes ( Boxity(..) )
46 import SrcLoc ( SrcLoc, incSrcLine, srcLocFile, srcLocLine,
47 replaceSrcLine, mkSrcLoc )
49 import ErrUtils ( Message )
56 import Char ( chr, ord )
57 import PrelRead ( readRational__ ) -- Glasgow non-std
60 %************************************************************************
62 \subsection{Data types}
64 %************************************************************************
66 The token data type, fairly un-interesting except from one
67 constructor, @ITidinfo@, which is used to lazily lex id info (arity,
68 strictness, unfolding etc).
70 The Idea/Observation here is that the renamer needs to scan through
71 all of an interface file before it can continue. But only a fraction
72 of the information contained in the file turns out to be useful, so
73 delaying as much as possible of the scanning and parsing of an
74 interface file Makes Sense (Heap profiles of the compiler
75 show a reduction in heap usage by at least a factor of two,
78 Hence, the interface file lexer spots when value declarations are
79 being scanned and return the @ITidinfo@ and @ITtype@ constructors
80 for the type and any other id info for that binding (unfolding, strictness
81 etc). These constructors are applied to the result of lexing these sub-chunks.
83 The lexing of the type and id info is all done lazily, of course, so
84 the scanning (and subsequent parsing) will be done *only* on the ids the
85 renamer finds out that it is interested in. The rest will just be junked.
86 Laziness, you know it makes sense :-)
90 = ITas -- Haskell keywords
114 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
116 | ITforall -- GHC extension keywords
127 | ITinterface -- interface keywords
135 | ITccall (Bool,Bool,Safety) -- (is_dyn, is_casm, may_gc)
154 | ITunfold InlinePragInfo
155 | ITstrict ([Demand], Bool)
162 | ITspecialise_prag -- Pragmas
172 | ITdotdot -- reserved symbols
186 | ITbiglam -- GHC-extension symbols
188 | ITocurly -- special symbols
190 | ITocurlybar -- {|, for type applications
191 | ITccurlybar -- |}, for type applications
204 | ITvarid FAST_STRING -- identifiers
205 | ITconid FAST_STRING
206 | ITvarsym FAST_STRING
207 | ITconsym FAST_STRING
208 | ITqvarid (FAST_STRING,FAST_STRING)
209 | ITqconid (FAST_STRING,FAST_STRING)
210 | ITqvarsym (FAST_STRING,FAST_STRING)
211 | ITqconsym (FAST_STRING,FAST_STRING)
213 | ITipvarid FAST_STRING -- GHC extension: implicit param: ?x
215 | ITpragma StringBuffer
218 | ITstring FAST_STRING
220 | ITrational Rational
223 | ITprimstring FAST_STRING
225 | ITprimfloat Rational
226 | ITprimdouble Rational
227 | ITlitlit FAST_STRING
229 | ITunknown String -- Used when the lexer can't make sense of it
230 | ITeof -- end of file token
231 deriving Show -- debugging
234 -----------------------------------------------------------------------------
238 pragmaKeywordsFM = listToUFM $
239 map (\ (x,y) -> (_PK_ x,y))
240 [( "SPECIALISE", ITspecialise_prag ),
241 ( "SPECIALIZE", ITspecialise_prag ),
242 ( "SOURCE", ITsource_prag ),
243 ( "INLINE", ITinline_prag ),
244 ( "NOINLINE", ITnoinline_prag ),
245 ( "NOTINLINE", ITnoinline_prag ),
246 ( "LINE", ITline_prag ),
247 ( "RULES", ITrules_prag ),
248 ( "RULEZ", ITrules_prag ), -- american spelling :-)
249 ( "SCC", ITscc_prag ),
250 ( "DEPRECATED", ITdeprecated_prag )
253 haskellKeywordsFM = listToUFM $
254 map (\ (x,y) -> (_PK_ x,y))
255 [( "_", ITunderscore ),
258 ( "class", ITclass ),
260 ( "default", ITdefault ),
261 ( "deriving", ITderiving ),
264 ( "hiding", IThiding ),
266 ( "import", ITimport ),
268 ( "infix", ITinfix ),
269 ( "infixl", ITinfixl ),
270 ( "infixr", ITinfixr ),
271 ( "instance", ITinstance ),
273 ( "module", ITmodule ),
274 ( "newtype", ITnewtype ),
276 ( "qualified", ITqualified ),
279 ( "where", ITwhere ),
280 ( "_scc_", ITscc ) -- ToDo: remove
283 isSpecial :: Token -> Bool
284 -- If we see M.x, where x is a keyword, but
285 -- is special, we treat is as just plain M.x,
287 isSpecial ITas = True
288 isSpecial IThiding = True
289 isSpecial ITqualified = True
290 isSpecial ITforall = True
291 isSpecial ITexport = True
292 isSpecial ITlabel = True
293 isSpecial ITdynamic = True
294 isSpecial ITunsafe = True
295 isSpecial ITwith = True
296 isSpecial ITccallconv = True
297 isSpecial ITstdcallconv = True
300 -- IMPORTANT: Keep this in synch with ParseIface.y's var_fs production! (SUP)
301 ghcExtensionKeywordsFM = listToUFM $
302 map (\ (x,y) -> (_PK_ x,y))
303 [ ( "forall", ITforall ),
304 ( "foreign", ITforeign ),
305 ( "export", ITexport ),
306 ( "label", ITlabel ),
307 ( "dynamic", ITdynamic ),
308 ( "unsafe", ITunsafe ),
310 ( "stdcall", ITstdcallconv),
311 ( "ccall", ITccallconv),
312 ( "dotnet", ITdotnet),
313 ("_ccall_", ITccall (False, False, PlayRisky)),
314 ("_ccall_GC_", ITccall (False, False, PlaySafe)),
315 ("_casm_", ITccall (False, True, PlayRisky)),
316 ("_casm_GC_", ITccall (False, True, PlaySafe)),
318 -- interface keywords
319 ("__interface", ITinterface),
320 ("__export", IT__export),
321 ("__depends", ITdepends),
322 ("__forall", IT__forall),
323 ("__letrec", ITletrec),
324 ("__coerce", ITcoerce),
325 ("__inline_me", ITinlineMe),
326 ("__inline_call", ITinlineCall),
327 ("__depends", ITdepends),
328 ("__DEFAULT", ITdefaultbranch),
330 ("__integer", ITinteger_lit),
331 ("__float", ITfloat_lit),
332 ("__int64", ITint64_lit),
333 ("__word", ITword_lit),
334 ("__word64", ITword64_lit),
335 ("__rational", ITrational_lit),
336 ("__addr", ITaddr_lit),
337 ("__label", ITlabel_lit),
338 ("__litlit", ITlit_lit),
339 ("__string", ITstring_lit),
342 ("__fuall", ITfuall),
344 ("__P", ITspecialise),
347 ("__D", ITdeprecated),
348 ("__U", ITunfold NoInlinePragInfo),
350 ("__ccall", ITccall (False, False, PlayRisky)),
351 ("__ccall_GC", ITccall (False, False, PlaySafe)),
352 ("__dyn_ccall", ITccall (True, False, PlayRisky)),
353 ("__dyn_ccall_GC", ITccall (True, False, PlaySafe)),
354 ("__casm", ITccall (False, True, PlayRisky)),
355 ("__dyn_casm", ITccall (True, True, PlayRisky)),
356 ("__casm_GC", ITccall (False, True, PlaySafe)),
357 ("__dyn_casm_GC", ITccall (True, True, PlaySafe)),
363 haskellKeySymsFM = listToUFM $
364 map (\ (x,y) -> (_PK_ x,y))
377 ,(".", ITdot) -- sadly, for 'forall a . t'
381 -----------------------------------------------------------------------------
386 - (glaexts) lexing an interface file or -fglasgow-exts
387 - (bol) pointer to beginning of line (for column calculations)
388 - (buf) pointer to beginning of token
389 - (buf) pointer to current char
390 - (atbol) flag indicating whether we're at the beginning of a line
393 lexer :: (Token -> P a) -> P a
394 lexer cont buf s@(PState{
396 glasgow_exts = glaexts,
402 -- first, start a new lexeme and lose all the whitespace
404 tab line bol atbol (stepOverLexeme buf)
406 line = srcLocLine loc
408 tab y bol atbol buf = -- trace ("tab: " ++ show (I# y) ++ " : " ++ show (currentChar buf)) $
409 case currentChar# buf of
412 if bufferExhausted (stepOn buf)
413 then cont ITeof buf s'
414 else trace "lexer: misplaced NUL?" $
415 tab y bol atbol (stepOn buf)
417 '\n'# -> let buf' = stepOn buf
418 in tab (y +# 1#) (currentIndex# buf') 1# buf'
420 -- find comments. This got harder in Haskell 98.
421 '-'# -> let trundle n =
422 let next = lookAhead# buf n in
423 if next `eqChar#` '-'# then trundle (n +# 1#)
424 else if is_symbol next || n <# 2#
427 (stepOnUntilChar# (stepOnBy# buf n) '\n'#)
430 -- comments and pragmas. We deal with LINE pragmas here,
431 -- and throw out any unrecognised pragmas as comments. Any
432 -- pragmas we know about are dealt with later (after any layout
433 -- processing if necessary).
434 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
435 if lookAhead# buf 2# `eqChar#` '#'# then
436 if lookAhead# buf 3# `eqChar#` '#'# then is_a_token else
437 case expandWhile# is_space (setCurrentPos# buf 3#) of { buf1->
438 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2->
439 let lexeme = mkFastString -- ToDo: too slow
440 (map toUpper (lexemeToString buf2)) in
441 case lookupUFM pragmaKeywordsFM lexeme of
443 line_prag skip_to_end buf2 s'
444 Just other -> is_a_token
445 Nothing -> skip_to_end (stepOnBy# buf 2#) s'
448 else skip_to_end (stepOnBy# buf 2#) s'
450 skip_to_end = nested_comment (lexer cont)
452 -- special GHC extension: we grok cpp-style #line pragmas
453 '#'# | lexemeIndex buf ==# bol -> -- the '#' must be in column 0
454 case expandWhile# is_space (stepOn buf) of { buf1 ->
455 if is_digit (currentChar# buf1)
456 then line_prag next_line buf1 s'
460 next_line buf = lexer cont (stepOnUntilChar# buf '\n'#)
462 -- tabs have been expanded beforehand
463 c | is_space c -> tab y bol atbol (stepOn buf)
464 | otherwise -> is_a_token
466 where s' = s{loc = replaceSrcLine loc y,
470 is_a_token | atbol /=# 0# = lexBOL cont buf s'
471 | otherwise = lexToken cont glaexts buf s'
473 -- {-# LINE .. #-} pragmas. yeuch.
474 line_prag cont buf s@PState{loc=loc} =
475 case expandWhile# is_space buf of { buf1 ->
476 case scanNumLit 0 (stepOverLexeme buf1) of { (line,buf2) ->
477 -- subtract one: the line number refers to the *following* line.
478 let real_line = line - 1 in
479 case fromInteger real_line of { i@(I# l) ->
480 -- ToDo, if no filename then we skip the newline.... d'oh
481 case expandWhile# is_space buf2 of { buf3 ->
482 case currentChar# buf3 of
484 case untilEndOfString# (stepOn (stepOverLexeme buf3)) of { buf4 ->
486 file = lexemeToFastString buf4
487 new_buf = stepOn (stepOverLexeme buf4)
489 if nullFastString file
490 then cont new_buf s{loc = replaceSrcLine loc l}
491 else cont new_buf s{loc = mkSrcLoc file i}
493 _other -> cont (stepOverLexeme buf3) s{loc = replaceSrcLine loc l}
496 nested_comment :: P a -> P a
497 nested_comment cont buf = loop buf
500 case currentChar# buf of
501 '\NUL'# | bufferExhausted (stepOn buf) ->
502 lexError "unterminated `{-'" buf -- -}
503 '-'# | lookAhead# buf 1# `eqChar#` '}'# ->
504 cont (stepOnBy# buf 2#)
506 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
507 nested_comment (nested_comment cont) (stepOnBy# buf 2#)
509 '\n'# -> \ s@PState{loc=loc} ->
510 let buf' = stepOn buf in
511 nested_comment cont buf'
512 s{loc = incSrcLine loc, bol = currentIndex# buf',
515 _ -> nested_comment cont (stepOn buf)
517 -- When we are lexing the first token of a line, check whether we need to
518 -- insert virtual semicolons or close braces due to layout.
520 lexBOL :: (Token -> P a) -> P a
521 lexBOL cont buf s@(PState{
523 glasgow_exts = glaexts,
528 if need_close_curly then
529 --trace ("col = " ++ show (I# col) ++ ", layout: inserting '}'") $
530 cont ITvccurly buf s{atbol = 1#, context = tail ctx}
531 else if need_semi_colon then
532 --trace ("col = " ++ show (I# col) ++ ", layout: inserting ';'") $
533 cont ITsemi buf s{atbol = 0#}
535 lexToken cont glaexts buf s{atbol = 0#}
537 col = currentIndex# buf -# bol
550 Layout n -> col ==# n
553 lexToken :: (Token -> P a) -> Int# -> P a
554 lexToken cont glaexts buf =
555 -- trace "lexToken" $
556 case currentChar# buf of
558 -- special symbols ----------------------------------------------------
559 '('# | flag glaexts && lookAhead# buf 1# `eqChar#` '#'#
560 -> cont IToubxparen (setCurrentPos# buf 2#)
562 -> cont IToparen (incLexeme buf)
564 ')'# -> cont ITcparen (incLexeme buf)
565 '['# -> cont ITobrack (incLexeme buf)
566 ']'# -> cont ITcbrack (incLexeme buf)
567 ','# -> cont ITcomma (incLexeme buf)
568 ';'# -> cont ITsemi (incLexeme buf)
569 '}'# -> \ s@PState{context = ctx} ->
571 (_:ctx') -> cont ITccurly (incLexeme buf) s{context=ctx'}
572 _ -> lexError "too many '}'s" buf s
573 '|'# -> case lookAhead# buf 1# of
574 '}'# | flag glaexts -> cont ITccurlybar
575 (setCurrentPos# buf 2#)
576 _ -> lex_sym cont (incLexeme buf)
579 '#'# -> case lookAhead# buf 1# of
580 ')'# | flag glaexts -> cont ITcubxparen (setCurrentPos# buf 2#)
581 '-'# -> case lookAhead# buf 2# of
582 '}'# -> cont ITclose_prag (setCurrentPos# buf 3#)
583 _ -> lex_sym cont (incLexeme buf)
584 _ -> lex_sym cont (incLexeme buf)
586 '`'# | flag glaexts && lookAhead# buf 1# `eqChar#` '`'#
587 -> lex_cstring cont (setCurrentPos# buf 2#)
589 -> cont ITbackquote (incLexeme buf)
591 '{'# -> -- look for "{-##" special iface pragma
592 case lookAhead# buf 1# of
594 -> cont ITocurlybar (setCurrentPos# buf 2#)
595 '-'# -> case lookAhead# buf 2# of
596 '#'# -> case lookAhead# buf 3# of
599 = doDiscard 0# (stepOnBy# (stepOverLexeme buf) 4#) in
600 cont (ITpragma lexeme) buf'
601 _ -> lex_prag cont (setCurrentPos# buf 3#)
602 _ -> cont ITocurly (incLexeme buf)
603 _ -> (layoutOff `thenP_` cont ITocurly) (incLexeme buf)
605 -- strings/characters -------------------------------------------------
606 '\"'#{-"-} -> lex_string cont glaexts [] (incLexeme buf)
607 '\''# -> lex_char (char_end cont) glaexts (incLexeme buf)
609 -- strictness and cpr pragmas and __scc treated specially.
610 '_'# | flag glaexts ->
611 case lookAhead# buf 1# of
612 '_'# -> case lookAhead# buf 2# of
614 lex_demand cont (stepOnUntil (not . isSpace)
615 (stepOnBy# buf 3#)) -- past __S
617 cont ITcprinfo (stepOnBy# buf 3#) -- past __M
620 case prefixMatch (stepOnBy# buf 3#) "cc" of
621 Just buf' -> lex_scc cont (stepOverLexeme buf')
622 Nothing -> lex_id cont glaexts buf
623 _ -> lex_id cont glaexts buf
624 _ -> lex_id cont glaexts buf
626 -- Hexadecimal and octal constants
627 '0'# | (ch `eqChar#` 'x'# || ch `eqChar#` 'X'#) && is_hexdigit ch2
628 -> readNum (after_lexnum cont glaexts) buf' is_hexdigit 16 hex
629 | (ch `eqChar#` 'o'# || ch `eqChar#` 'O'#) && is_octdigit ch2
630 -> readNum (after_lexnum cont glaexts) buf' is_octdigit 8 oct_or_dec
631 where ch = lookAhead# buf 1#
632 ch2 = lookAhead# buf 2#
633 buf' = setCurrentPos# buf 2#
636 if bufferExhausted (stepOn buf) then
639 trace "lexIface: misplaced NUL?" $
640 cont (ITunknown "\NUL") (stepOn buf)
642 '?'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
643 lex_ip cont (incLexeme buf)
644 c | is_digit c -> lex_num cont glaexts 0 buf
645 | is_symbol c -> lex_sym cont buf
646 | is_upper c -> lex_con cont glaexts buf
647 | is_ident c -> lex_id cont glaexts buf
648 | otherwise -> lexError "illegal character" buf
650 -- Int# is unlifted, and therefore faster than Bool for flags.
656 -------------------------------------------------------------------------------
660 = case expandWhile# is_space buf of { buf1 ->
661 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2 ->
662 let lexeme = mkFastString (map toUpper (lexemeToString buf2)) in
663 case lookupUFM pragmaKeywordsFM lexeme of
664 Just kw -> cont kw (mergeLexemes buf buf2)
665 Nothing -> panic "lex_prag"
668 -------------------------------------------------------------------------------
671 lex_string cont glaexts s buf
672 = case currentChar# buf of
674 let buf' = incLexeme buf
675 s' = mkFastStringNarrow (map chr (reverse s))
676 in case currentChar# buf' of
677 '#'# | flag glaexts -> if all (<= 0xFF) s
678 then cont (ITprimstring s') (incLexeme buf')
679 else lexError "primitive string literal must contain only characters <= \'\\xFF\'" buf'
680 _ -> cont (ITstring s') buf'
682 -- ignore \& in a string, deal with string gaps
683 '\\'# | next_ch `eqChar#` '&'#
684 -> lex_string cont glaexts s buf'
686 -> lex_stringgap cont glaexts s (incLexeme buf)
688 where next_ch = lookAhead# buf 1#
689 buf' = setCurrentPos# buf 2#
691 _ -> lex_char (lex_next_string cont s) glaexts buf
693 lex_stringgap cont glaexts s buf
694 = let buf' = incLexeme buf in
695 case currentChar# buf of
696 '\n'# -> \st@PState{loc = loc} -> lex_stringgap cont glaexts s buf'
697 st{loc = incSrcLine loc}
698 '\\'# -> lex_string cont glaexts s buf'
699 c | is_space c -> lex_stringgap cont glaexts s buf'
700 other -> charError buf'
702 lex_next_string cont s glaexts c buf = lex_string cont glaexts (c:s) buf
704 lex_char :: (Int# -> Int -> P a) -> Int# -> P a
705 lex_char cont glaexts buf
706 = case currentChar# buf of
707 '\\'# -> lex_escape (cont glaexts) (incLexeme buf)
708 c | is_any c -> cont glaexts (I# (ord# c)) (incLexeme buf)
709 other -> charError buf
711 char_end cont glaexts c buf
712 = case currentChar# buf of
713 '\''# -> let buf' = incLexeme buf in
714 case currentChar# buf' of
716 -> cont (ITprimchar c) (incLexeme buf')
717 _ -> cont (ITchar c) buf'
721 = let buf' = incLexeme buf in
722 case currentChar# buf of
723 'a'# -> cont (ord '\a') buf'
724 'b'# -> cont (ord '\b') buf'
725 'f'# -> cont (ord '\f') buf'
726 'n'# -> cont (ord '\n') buf'
727 'r'# -> cont (ord '\r') buf'
728 't'# -> cont (ord '\t') buf'
729 'v'# -> cont (ord '\v') buf'
730 '\\'# -> cont (ord '\\') buf'
731 '"'# -> cont (ord '\"') buf'
732 '\''# -> cont (ord '\'') buf'
733 '^'# -> let c = currentChar# buf' in
734 if c `geChar#` '@'# && c `leChar#` '_'#
735 then cont (I# (ord# c -# ord# '@'#)) (incLexeme buf')
738 'x'# -> readNum (after_charnum cont) buf' is_hexdigit 16 hex
739 'o'# -> readNum (after_charnum cont) buf' is_octdigit 8 oct_or_dec
741 -> readNum (after_charnum cont) buf is_digit 10 oct_or_dec
743 _ -> case [ (c,buf2) | (p,c) <- silly_escape_chars,
744 Just buf2 <- [prefixMatch buf p] ] of
745 (c,buf2):_ -> cont (ord c) buf2
748 after_charnum cont i buf
749 = if i >= 0 && i <= 0x10FFFF
750 then cont (fromInteger i) buf
753 readNum cont buf is_digit base conv = read buf 0
755 = case currentChar# buf of { c ->
757 then read (incLexeme buf) (i*base + (toInteger (I# (conv c))))
763 || (c `geChar#` 'a'# && c `leChar#` 'f'#)
764 || (c `geChar#` 'A'# && c `leChar#` 'F'#)
766 hex c | is_digit c = ord# c -# ord# '0'#
767 | otherwise = ord# (to_lower c) -# ord# 'a'# +# 10#
768 oct_or_dec c = ord# c -# ord# '0'#
770 is_octdigit c = c `geChar#` '0'# && c `leChar#` '7'#
773 | c `geChar#` 'A'# && c `leChar#` 'Z'#
774 = chr# (ord# c -# (ord# 'A'# -# ord# 'a'#))
777 charError buf = lexError "error in character literal" buf
779 silly_escape_chars = [
816 -------------------------------------------------------------------------------
818 lex_demand cont buf =
819 case read_em [] buf of { (ls,buf') ->
820 case currentChar# buf' of
821 'B'# -> cont (ITstrict (ls, True )) (incLexeme buf')
822 _ -> cont (ITstrict (ls, False)) buf'
825 -- code snatched from Demand.lhs
827 case currentChar# buf of
828 'L'# -> read_em (WwLazy False : acc) (stepOn buf)
829 'A'# -> read_em (WwLazy True : acc) (stepOn buf)
830 'S'# -> read_em (WwStrict : acc) (stepOn buf)
831 'P'# -> read_em (WwPrim : acc) (stepOn buf)
832 'E'# -> read_em (WwEnum : acc) (stepOn buf)
833 ')'# -> (reverse acc, stepOn buf)
834 'U'# -> do_unpack True acc (stepOnBy# buf 2#)
835 'u'# -> do_unpack False acc (stepOnBy# buf 2#)
836 _ -> (reverse acc, buf)
838 do_unpack wrapper_unpacks acc buf
839 = case read_em [] buf of
840 (stuff, rest) -> read_em (WwUnpack wrapper_unpacks stuff : acc) rest
845 case currentChar# buf of
846 'C'# -> cont ITsccAllCafs (incLexeme buf)
847 other -> cont ITscc buf
849 -----------------------------------------------------------------------------
852 lex_num :: (Token -> P a) -> Int# -> Integer -> P a
853 lex_num cont glaexts acc buf =
854 case scanNumLit acc buf of
856 case currentChar# buf' of
857 '.'# | is_digit (lookAhead# buf' 1#) ->
858 -- this case is not optimised at all, as the
859 -- presence of floating point numbers in interface
860 -- files is not that common. (ToDo)
861 case expandWhile# is_digit (incLexeme buf') of
862 buf2 -> -- points to first non digit char
864 let l = case currentChar# buf2 of
870 = let buf3 = incLexeme buf2 in
871 case currentChar# buf3 of
872 '-'# -> expandWhile# is_digit (incLexeme buf3)
873 '+'# -> expandWhile# is_digit (incLexeme buf3)
874 x | is_digit x -> expandWhile# is_digit buf3
877 v = readRational__ (lexemeToString l)
879 in case currentChar# l of -- glasgow exts only
880 '#'# | flag glaexts -> let l' = incLexeme l in
881 case currentChar# l' of
882 '#'# -> cont (ITprimdouble v) (incLexeme l')
883 _ -> cont (ITprimfloat v) l'
884 _ -> cont (ITrational v) l
886 _ -> after_lexnum cont glaexts acc' buf'
888 after_lexnum cont glaexts i buf
889 = case currentChar# buf of
890 '#'# | flag glaexts -> cont (ITprimint i) (incLexeme buf)
891 _ -> cont (ITinteger i) buf
893 -----------------------------------------------------------------------------
894 -- C "literal literal"s (i.e. things like ``NULL'', ``stdout'' etc.)
896 -- we lexemeToFastString on the bit between the ``''s, but include the
897 -- quotes in the full lexeme.
899 lex_cstring cont buf =
900 case expandUntilMatch (stepOverLexeme buf) "\'\'" of
901 Just buf' -> cont (ITlitlit (lexemeToFastString
902 (setCurrentPos# buf' (negateInt# 2#))))
903 (mergeLexemes buf buf')
904 Nothing -> lexError "unterminated ``" buf
906 -----------------------------------------------------------------------------
907 -- identifiers, symbols etc.
910 case expandWhile# is_ident buf of
911 buf' -> cont (ITipvarid lexeme) buf'
912 where lexeme = lexemeToFastString buf'
914 lex_id cont glaexts buf =
915 let buf1 = expandWhile# is_ident buf in
918 case (if flag glaexts
919 then expandWhile# (eqChar# '#'#) buf1 -- slurp trailing hashes
920 else buf1) of { buf' ->
922 let lexeme = lexemeToFastString buf' in
924 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
925 Just kwd_token -> --trace ("hkeywd: "++_UNPK_(lexeme)) $
929 let var_token = cont (ITvarid lexeme) buf' in
931 if not (flag glaexts)
935 case lookupUFM ghcExtensionKeywordsFM lexeme of {
936 Just kwd_token -> cont kwd_token buf';
943 case expandWhile# is_symbol buf of
944 buf' -> case lookupUFM haskellKeySymsFM lexeme of {
945 Just kwd_token -> --trace ("keysym: "++unpackFS lexeme) $
946 cont kwd_token buf' ;
947 Nothing -> --trace ("sym: "++unpackFS lexeme) $
948 cont (mk_var_token lexeme) buf'
950 where lexeme = lexemeToFastString buf'
953 -- lex_con recursively collects components of a qualified identifer.
954 -- The argument buf is the StringBuffer representing the lexeme
955 -- identified so far, where the next character is upper-case.
957 lex_con cont glaexts buf =
958 -- trace ("con: "{-++unpackFS lexeme-}) $
959 let empty_buf = stepOverLexeme buf in
960 case expandWhile# is_ident empty_buf of { buf1 ->
961 case slurp_trailing_hashes buf1 glaexts of { con_buf ->
963 let all_buf = mergeLexemes buf con_buf
965 con_lexeme = lexemeToFastString con_buf
966 mod_lexeme = lexemeToFastString (decLexeme buf)
967 all_lexeme = lexemeToFastString all_buf
970 | emptyLexeme buf = cont (ITconid con_lexeme) all_buf
971 | otherwise = cont (ITqconid (mod_lexeme,con_lexeme)) all_buf
974 case currentChar# all_buf of
975 '.'# -> maybe_qualified cont glaexts all_lexeme
976 (incLexeme all_buf) just_a_conid
981 maybe_qualified cont glaexts mod buf just_a_conid =
982 -- trace ("qid: "{-++unpackFS lexeme-}) $
983 case currentChar# buf of
984 '['# -> -- Special case for []
985 case lookAhead# buf 1# of
986 ']'# -> cont (ITqconid (mod,SLIT("[]"))) (setCurrentPos# buf 2#)
989 '('# -> -- Special case for (,,,)
990 -- This *is* necessary to deal with e.g. "instance C PrelBase.(,,)"
991 case lookAhead# buf 1# of
992 '#'# | flag glaexts -> case lookAhead# buf 2# of
993 ','# -> lex_ubx_tuple cont mod (setCurrentPos# buf 3#)
996 ')'# -> cont (ITqconid (mod,SLIT("()"))) (setCurrentPos# buf 2#)
997 ','# -> lex_tuple cont mod (setCurrentPos# buf 2#) just_a_conid
1000 '-'# -> case lookAhead# buf 1# of
1001 '>'# -> cont (ITqconid (mod,SLIT("(->)"))) (setCurrentPos# buf 2#)
1002 _ -> lex_id3 cont glaexts mod buf just_a_conid
1004 _ -> lex_id3 cont glaexts mod buf just_a_conid
1007 lex_id3 cont glaexts mod buf just_a_conid
1008 | is_upper (currentChar# buf) =
1009 lex_con cont glaexts buf
1011 | is_symbol (currentChar# buf) =
1013 start_new_lexeme = stepOverLexeme buf
1015 -- trace ("lex_id31 "{-++unpackFS lexeme-}) $
1016 case expandWhile# is_symbol start_new_lexeme of { buf' ->
1018 lexeme = lexemeToFastString buf'
1019 -- real lexeme is M.<sym>
1020 new_buf = mergeLexemes buf buf'
1022 cont (mk_qvar_token mod lexeme) new_buf
1023 -- wrong, but arguably morally right: M... is now a qvarsym
1028 start_new_lexeme = stepOverLexeme buf
1030 -- trace ("lex_id32 "{-++unpackFS lexeme-}) $
1031 case expandWhile# is_ident start_new_lexeme of { buf1 ->
1036 case slurp_trailing_hashes buf1 glaexts of { buf' ->
1039 lexeme = lexemeToFastString buf'
1040 new_buf = mergeLexemes buf buf'
1041 is_a_qvarid = cont (mk_qvar_token mod lexeme) new_buf
1043 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
1044 Nothing -> is_a_qvarid ;
1046 Just kwd_token | isSpecial kwd_token -- special ids (as, qualified, hiding) shouldn't be
1047 -> is_a_qvarid -- recognised as keywords here.
1049 -> just_a_conid -- avoid M.where etc.
1052 slurp_trailing_hashes buf glaexts
1053 | flag glaexts = expandWhile# (`eqChar#` '#'#) buf
1058 | is_upper f = ITconid pk_str
1059 | is_ident f = ITvarid pk_str
1060 | f `eqChar#` ':'# = ITconsym pk_str
1061 | otherwise = ITvarsym pk_str
1063 (C# f) = _HEAD_ pk_str
1064 -- tl = _TAIL_ pk_str
1066 mk_qvar_token m token =
1067 -- trace ("mk_qvar ") $
1068 case mk_var_token token of
1069 ITconid n -> ITqconid (m,n)
1070 ITvarid n -> ITqvarid (m,n)
1071 ITconsym n -> ITqconsym (m,n)
1072 ITvarsym n -> ITqvarsym (m,n)
1073 _ -> ITunknown (show token)
1076 ----------------------------------------------------------------------------
1077 Horrible stuff for dealing with M.(,,,)
1080 lex_tuple cont mod buf back_off =
1084 case currentChar# buf of
1085 ','# -> go (n+1) (stepOn buf)
1086 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Boxed n))) (stepOn buf)
1089 lex_ubx_tuple cont mod buf back_off =
1093 case currentChar# buf of
1094 ','# -> go (n+1) (stepOn buf)
1095 '#'# -> case lookAhead# buf 1# of
1096 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Unboxed n)))
1102 -----------------------------------------------------------------------------
1103 doDiscard rips along really fast, looking for a '##-}',
1104 indicating the end of the pragma we're skipping
1107 doDiscard inStr buf =
1108 case currentChar# buf of
1109 '#'# | inStr ==# 0# ->
1110 case lookAhead# buf 1# of { '#'# ->
1111 case lookAhead# buf 2# of { '-'# ->
1112 case lookAhead# buf 3# of { '}'# ->
1113 (lexemeToBuffer buf, stepOverLexeme (setCurrentPos# buf 4#));
1114 _ -> doDiscard inStr (incLexeme buf) };
1115 _ -> doDiscard inStr (incLexeme buf) };
1116 _ -> doDiscard inStr (incLexeme buf) }
1120 odd_slashes buf flg i# =
1121 case lookAhead# buf i# of
1122 '\\'# -> odd_slashes buf (not flg) (i# -# 1#)
1125 not_inStr = if inStr ==# 0# then 1# else 0#
1127 case lookAhead# buf (negateInt# 1#) of --backwards, actually
1128 '\\'# -> -- escaping something..
1129 if odd_slashes buf True (negateInt# 2#)
1130 then -- odd number of slashes, " is escaped.
1131 doDiscard inStr (incLexeme buf)
1132 else -- even number of slashes, \ is escaped.
1133 doDiscard not_inStr (incLexeme buf)
1134 _ -> doDiscard not_inStr (incLexeme buf)
1136 '\''# | inStr ==# 0# ->
1137 case lookAhead# buf 1# of { '"'# ->
1138 case lookAhead# buf 2# of { '\''# ->
1139 doDiscard inStr (setCurrentPos# buf 3#);
1140 _ -> doDiscard inStr (incLexeme buf) };
1141 _ -> doDiscard inStr (incLexeme buf) }
1143 _ -> doDiscard inStr (incLexeme buf)
1147 -----------------------------------------------------------------------------
1158 data PState = PState {
1160 glasgow_exts :: Int#,
1163 context :: [LayoutContext]
1166 type P a = StringBuffer -- Input string
1171 returnP a buf s = POk s a
1173 thenP :: P a -> (a -> P b) -> P b
1174 m `thenP` k = \ buf s ->
1176 POk s1 a -> k a buf s1
1177 PFailed err -> PFailed err
1179 thenP_ :: P a -> P b -> P b
1180 m `thenP_` k = m `thenP` \_ -> k
1182 mapP :: (a -> P b) -> [a] -> P [b]
1183 mapP f [] = returnP []
1186 mapP f as `thenP` \bs ->
1189 failP :: String -> P a
1190 failP msg buf s = PFailed (text msg)
1192 failMsgP :: Message -> P a
1193 failMsgP msg buf s = PFailed msg
1195 lexError :: String -> P a
1196 lexError str buf s@PState{ loc = loc }
1197 = failMsgP (hcat [ppr loc, text ": ", text str]) buf s
1199 getSrcLocP :: P SrcLoc
1200 getSrcLocP buf s@(PState{ loc = loc }) = POk s loc
1202 -- use a temporary SrcLoc for the duration of the argument
1203 setSrcLocP :: SrcLoc -> P a -> P a
1204 setSrcLocP new_loc p buf s =
1205 case p buf s{ loc=new_loc } of
1207 PFailed e -> PFailed e
1209 getSrcFile :: P FAST_STRING
1210 getSrcFile buf s@(PState{ loc = loc }) = POk s (srcLocFile loc)
1212 pushContext :: LayoutContext -> P ()
1213 pushContext ctxt buf s@(PState{ context = ctx }) = POk s{context = ctxt:ctx} ()
1217 This special case in layoutOn is to handle layout contexts with are
1218 indented the same or less than the current context. This is illegal
1219 according to the Haskell spec, so we have to arrange to close the
1220 current context. eg.
1225 after the first 'where', the sequence of events is:
1227 - layout system inserts a ';' (column 0)
1228 - parser begins a new context at column 0
1229 - parser shifts ';' (legal empty declaration)
1230 - parser sees 'class': parse error (we're still in the inner context)
1232 trouble is, by the time we know we need a new context, the lexer has
1233 already generated the ';'. Hacky solution is as follows: since we
1234 know the column of the next token (it's the column number of the new
1235 context), we set the ACTUAL column number of the new context to this
1236 numer plus one. Hence the next time the lexer is called, a '}' will
1237 be generated to close the new context straight away. Furthermore, we
1238 have to set the atbol flag so that the ';' that the parser shifted as
1239 part of the new context is re-generated.
1241 when the new context is *less* indented than the current one:
1243 f = f where g = g where
1246 - current context: column 12.
1247 - on seeing 'h' (column 0), the layout system inserts '}'
1248 - parser starts a new context, column 0
1249 - parser sees '}', uses it to close new context
1250 - we still need to insert another '}' followed by a ';',
1251 hence the atbol trick.
1253 There's also a special hack in here to deal with
1260 i.e. the inner context is at the same indentation level as the outer
1261 context. This is strictly illegal according to Haskell 98, but
1262 there's a lot of existing code using this style and it doesn't make
1263 any sense to disallow it, since empty 'do' lists don't make sense.
1266 layoutOn :: Bool -> P ()
1267 layoutOn strict buf s@(PState{ bol = bol, context = ctx }) =
1268 let offset = lexemeIndex buf -# bol in
1271 | if strict then prev_off >=# offset else prev_off ># offset ->
1272 --trace ("layout on, column: " ++ show (I# offset)) $
1273 POk s{ context = Layout (offset +# 1#) : ctx, atbol = 1# } ()
1275 --trace ("layout on, column: " ++ show (I# offset)) $
1276 POk s{ context = Layout offset : ctx } ()
1279 layoutOff buf s@(PState{ context = ctx }) =
1280 POk s{ context = NoLayout:ctx } ()
1283 popContext = \ buf s@(PState{ context = ctx, loc = loc }) ->
1285 (_:tl) -> POk s{ context = tl } ()
1286 [] -> PFailed (srcParseErr buf loc)
1289 Note that if the name of the file we're processing ends
1290 with `hi-boot', we accept it on faith as having the right
1291 version. This is done so that .hi-boot files that comes
1292 with hsc don't have to be updated before every release,
1293 *and* it allows us to share .hi-boot files with versions
1294 of hsc that don't have .hi version checking (e.g., ghc-2.10's)
1296 If the version number is 0, the checking is also turned off.
1297 (needed to deal with GHC.hi only!)
1299 Once we can assume we're compiling with a version of ghc that
1300 supports interface file checking, we can drop the special
1303 checkVersion :: Maybe Integer -> P ()
1304 checkVersion mb@(Just v) buf s@(PState{loc = loc})
1305 | (v==0) || (v == fromInt opt_HiVersion) || opt_NoHiCheck = POk s ()
1306 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1307 checkVersion mb@Nothing buf s@(PState{loc = loc})
1308 | "hi-boot" `isSuffixOf` (_UNPK_ (srcLocFile loc)) = POk s ()
1309 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1311 -----------------------------------------------------------------
1313 ifaceParseErr :: StringBuffer -> SrcLoc -> Message
1315 = hsep [ppr l, ptext SLIT("Interface file parse error; on input `"),
1316 text (lexemeToString s), char '\'']
1318 ifaceVersionErr hi_vers l toks
1319 = hsep [ppr l, ptext SLIT("Interface file version error;"),
1320 ptext SLIT("Expected"), int opt_HiVersion,
1321 ptext SLIT("found "), pp_version]
1325 Nothing -> ptext SLIT("pre ghc-3.02 version")
1326 Just v -> ptext SLIT("version") <+> integer v
1328 -----------------------------------------------------------------------------
1330 srcParseErr :: StringBuffer -> SrcLoc -> Message
1334 then ptext SLIT(": parse error (possibly incorrect indentation)")
1335 else hcat [ptext SLIT(": parse error on input "),
1336 char '`', text token, char '\'']
1339 token = lexemeToString s