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 NewDemand ( StrictSig(..), Demand(..), Keepity(..),
44 DmdResult(..), mkTopDmdType )
45 import UniqFM ( listToUFM, lookupUFM )
46 import BasicTypes ( Boxity(..) )
47 import SrcLoc ( SrcLoc, incSrcLine, srcLocFile, srcLocLine,
48 replaceSrcLine, mkSrcLoc )
50 import ErrUtils ( Message )
57 import Char ( chr, ord )
58 import PrelRead ( readRational__ ) -- Glasgow non-std
61 %************************************************************************
63 \subsection{Data types}
65 %************************************************************************
67 The token data type, fairly un-interesting except from one
68 constructor, @ITidinfo@, which is used to lazily lex id info (arity,
69 strictness, unfolding etc).
71 The Idea/Observation here is that the renamer needs to scan through
72 all of an interface file before it can continue. But only a fraction
73 of the information contained in the file turns out to be useful, so
74 delaying as much as possible of the scanning and parsing of an
75 interface file Makes Sense (Heap profiles of the compiler
76 show a reduction in heap usage by at least a factor of two,
79 Hence, the interface file lexer spots when value declarations are
80 being scanned and return the @ITidinfo@ and @ITtype@ constructors
81 for the type and any other id info for that binding (unfolding, strictness
82 etc). These constructors are applied to the result of lexing these sub-chunks.
84 The lexing of the type and id info is all done lazily, of course, so
85 the scanning (and subsequent parsing) will be done *only* on the ids the
86 renamer finds out that it is interested in. The rest will just be junked.
87 Laziness, you know it makes sense :-)
91 = ITas -- Haskell keywords
115 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
117 | ITforall -- GHC extension keywords
128 | ITinterface -- interface keywords
136 | ITccall (Bool,Bool,Safety) -- (is_dyn, is_casm, may_gc)
155 | ITunfold InlinePragInfo
163 | ITspecialise_prag -- Pragmas
173 | ITdotdot -- reserved symbols
187 | ITbiglam -- GHC-extension symbols
189 | ITocurly -- special symbols
191 | ITocurlybar -- {|, for type applications
192 | ITccurlybar -- |}, for type applications
205 | ITvarid FAST_STRING -- identifiers
206 | ITconid FAST_STRING
207 | ITvarsym FAST_STRING
208 | ITconsym FAST_STRING
209 | ITqvarid (FAST_STRING,FAST_STRING)
210 | ITqconid (FAST_STRING,FAST_STRING)
211 | ITqvarsym (FAST_STRING,FAST_STRING)
212 | ITqconsym (FAST_STRING,FAST_STRING)
214 | ITipvarid FAST_STRING -- GHC extension: implicit param: ?x
216 | ITpragma StringBuffer
219 | ITstring FAST_STRING
221 | ITrational Rational
224 | ITprimstring FAST_STRING
226 | ITprimfloat Rational
227 | ITprimdouble Rational
228 | ITlitlit FAST_STRING
230 | ITunknown String -- Used when the lexer can't make sense of it
231 | ITeof -- end of file token
232 deriving Show -- debugging
235 -----------------------------------------------------------------------------
239 pragmaKeywordsFM = listToUFM $
240 map (\ (x,y) -> (_PK_ x,y))
241 [( "SPECIALISE", ITspecialise_prag ),
242 ( "SPECIALIZE", ITspecialise_prag ),
243 ( "SOURCE", ITsource_prag ),
244 ( "INLINE", ITinline_prag ),
245 ( "NOINLINE", ITnoinline_prag ),
246 ( "NOTINLINE", ITnoinline_prag ),
247 ( "LINE", ITline_prag ),
248 ( "RULES", ITrules_prag ),
249 ( "RULEZ", ITrules_prag ), -- american spelling :-)
250 ( "SCC", ITscc_prag ),
251 ( "DEPRECATED", ITdeprecated_prag )
254 haskellKeywordsFM = listToUFM $
255 map (\ (x,y) -> (_PK_ x,y))
256 [( "_", ITunderscore ),
259 ( "class", ITclass ),
261 ( "default", ITdefault ),
262 ( "deriving", ITderiving ),
265 ( "hiding", IThiding ),
267 ( "import", ITimport ),
269 ( "infix", ITinfix ),
270 ( "infixl", ITinfixl ),
271 ( "infixr", ITinfixr ),
272 ( "instance", ITinstance ),
274 ( "module", ITmodule ),
275 ( "newtype", ITnewtype ),
277 ( "qualified", ITqualified ),
280 ( "where", ITwhere ),
281 ( "_scc_", ITscc ) -- ToDo: remove
284 isSpecial :: Token -> Bool
285 -- If we see M.x, where x is a keyword, but
286 -- is special, we treat is as just plain M.x,
288 isSpecial ITas = True
289 isSpecial IThiding = True
290 isSpecial ITqualified = True
291 isSpecial ITforall = True
292 isSpecial ITexport = True
293 isSpecial ITlabel = True
294 isSpecial ITdynamic = True
295 isSpecial ITunsafe = True
296 isSpecial ITwith = True
297 isSpecial ITccallconv = True
298 isSpecial ITstdcallconv = True
301 -- IMPORTANT: Keep this in synch with ParseIface.y's var_fs production! (SUP)
302 ghcExtensionKeywordsFM = listToUFM $
303 map (\ (x,y) -> (_PK_ x,y))
304 [ ( "forall", ITforall ),
305 ( "foreign", ITforeign ),
306 ( "export", ITexport ),
307 ( "label", ITlabel ),
308 ( "dynamic", ITdynamic ),
309 ( "unsafe", ITunsafe ),
311 ( "stdcall", ITstdcallconv),
312 ( "ccall", ITccallconv),
313 ( "dotnet", ITdotnet),
314 ("_ccall_", ITccall (False, False, PlayRisky)),
315 ("_ccall_GC_", ITccall (False, False, PlaySafe)),
316 ("_casm_", ITccall (False, True, PlayRisky)),
317 ("_casm_GC_", ITccall (False, True, PlaySafe)),
319 -- interface keywords
320 ("__interface", ITinterface),
321 ("__export", IT__export),
322 ("__depends", ITdepends),
323 ("__forall", IT__forall),
324 ("__letrec", ITletrec),
325 ("__coerce", ITcoerce),
326 ("__inline_me", ITinlineMe),
327 ("__inline_call", ITinlineCall),
328 ("__depends", ITdepends),
329 ("__DEFAULT", ITdefaultbranch),
331 ("__integer", ITinteger_lit),
332 ("__float", ITfloat_lit),
333 ("__int64", ITint64_lit),
334 ("__word", ITword_lit),
335 ("__word64", ITword64_lit),
336 ("__rational", ITrational_lit),
337 ("__addr", ITaddr_lit),
338 ("__label", ITlabel_lit),
339 ("__litlit", ITlit_lit),
340 ("__string", ITstring_lit),
343 ("__fuall", ITfuall),
345 ("__P", ITspecialise),
348 ("__D", ITdeprecated),
349 ("__U", ITunfold NoInlinePragInfo),
351 ("__ccall", ITccall (False, False, PlayRisky)),
352 ("__ccall_GC", ITccall (False, False, PlaySafe)),
353 ("__dyn_ccall", ITccall (True, False, PlayRisky)),
354 ("__dyn_ccall_GC", ITccall (True, False, PlaySafe)),
355 ("__casm", ITccall (False, True, PlayRisky)),
356 ("__dyn_casm", ITccall (True, True, PlayRisky)),
357 ("__casm_GC", ITccall (False, True, PlaySafe)),
358 ("__dyn_casm_GC", ITccall (True, True, PlaySafe)),
364 haskellKeySymsFM = listToUFM $
365 map (\ (x,y) -> (_PK_ x,y))
378 ,(".", ITdot) -- sadly, for 'forall a . t'
382 -----------------------------------------------------------------------------
387 - (glaexts) lexing an interface file or -fglasgow-exts
388 - (bol) pointer to beginning of line (for column calculations)
389 - (buf) pointer to beginning of token
390 - (buf) pointer to current char
391 - (atbol) flag indicating whether we're at the beginning of a line
394 lexer :: (Token -> P a) -> P a
395 lexer cont buf s@(PState{
397 glasgow_exts = glaexts,
403 -- first, start a new lexeme and lose all the whitespace
405 tab line bol atbol (stepOverLexeme buf)
407 line = srcLocLine loc
409 tab y bol atbol buf = -- trace ("tab: " ++ show (I# y) ++ " : " ++ show (currentChar buf)) $
410 case currentChar# buf of
413 if bufferExhausted (stepOn buf)
414 then cont ITeof buf s'
415 else trace "lexer: misplaced NUL?" $
416 tab y bol atbol (stepOn buf)
418 '\n'# -> let buf' = stepOn buf
419 in tab (y +# 1#) (currentIndex# buf') 1# buf'
421 -- find comments. This got harder in Haskell 98.
422 '-'# -> let trundle n =
423 let next = lookAhead# buf n in
424 if next `eqChar#` '-'# then trundle (n +# 1#)
425 else if is_symbol next || n <# 2#
428 (stepOnUntilChar# (stepOnBy# buf n) '\n'#)
431 -- comments and pragmas. We deal with LINE pragmas here,
432 -- and throw out any unrecognised pragmas as comments. Any
433 -- pragmas we know about are dealt with later (after any layout
434 -- processing if necessary).
435 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
436 if lookAhead# buf 2# `eqChar#` '#'# then
437 if lookAhead# buf 3# `eqChar#` '#'# then is_a_token else
438 case expandWhile# is_space (setCurrentPos# buf 3#) of { buf1->
439 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2->
440 let lexeme = mkFastString -- ToDo: too slow
441 (map toUpper (lexemeToString buf2)) in
442 case lookupUFM pragmaKeywordsFM lexeme of
443 -- ignore RULES pragmas when -fglasgow-exts is off
444 Just ITrules_prag | not (flag glaexts) ->
445 skip_to_end (stepOnBy# buf 2#) s'
447 line_prag skip_to_end buf2 s'
448 Just other -> is_a_token
449 Nothing -> skip_to_end (stepOnBy# buf 2#) s'
452 else skip_to_end (stepOnBy# buf 2#) s'
454 skip_to_end = nested_comment (lexer cont)
456 -- special GHC extension: we grok cpp-style #line pragmas
457 '#'# | lexemeIndex buf ==# bol -> -- the '#' must be in column 0
458 case expandWhile# is_space (stepOn buf) of { buf1 ->
459 if is_digit (currentChar# buf1)
460 then line_prag next_line buf1 s'
464 next_line buf = lexer cont (stepOnUntilChar# buf '\n'#)
466 -- tabs have been expanded beforehand
467 c | is_space c -> tab y bol atbol (stepOn buf)
468 | otherwise -> is_a_token
470 where s' = s{loc = replaceSrcLine loc y,
474 is_a_token | atbol /=# 0# = lexBOL cont buf s'
475 | otherwise = lexToken cont glaexts buf s'
477 -- {-# LINE .. #-} pragmas. yeuch.
478 line_prag cont buf s@PState{loc=loc} =
479 case expandWhile# is_space buf of { buf1 ->
480 case scanNumLit 0 (stepOverLexeme buf1) of { (line,buf2) ->
481 -- subtract one: the line number refers to the *following* line.
482 let real_line = line - 1 in
483 case fromInteger real_line of { i@(I# l) ->
484 -- ToDo, if no filename then we skip the newline.... d'oh
485 case expandWhile# is_space buf2 of { buf3 ->
486 case currentChar# buf3 of
488 case untilEndOfString# (stepOn (stepOverLexeme buf3)) of { buf4 ->
490 file = lexemeToFastString buf4
491 new_buf = stepOn (stepOverLexeme buf4)
493 if nullFastString file
494 then cont new_buf s{loc = replaceSrcLine loc l}
495 else cont new_buf s{loc = mkSrcLoc file i}
497 _other -> cont (stepOverLexeme buf3) s{loc = replaceSrcLine loc l}
500 nested_comment :: P a -> P a
501 nested_comment cont buf = loop buf
504 case currentChar# buf of
505 '\NUL'# | bufferExhausted (stepOn buf) ->
506 lexError "unterminated `{-'" buf -- -}
507 '-'# | lookAhead# buf 1# `eqChar#` '}'# ->
508 cont (stepOnBy# buf 2#)
510 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
511 nested_comment (nested_comment cont) (stepOnBy# buf 2#)
513 '\n'# -> \ s@PState{loc=loc} ->
514 let buf' = stepOn buf in
515 nested_comment cont buf'
516 s{loc = incSrcLine loc, bol = currentIndex# buf',
519 _ -> nested_comment cont (stepOn buf)
521 -- When we are lexing the first token of a line, check whether we need to
522 -- insert virtual semicolons or close braces due to layout.
524 lexBOL :: (Token -> P a) -> P a
525 lexBOL cont buf s@(PState{
527 glasgow_exts = glaexts,
532 if need_close_curly then
533 --trace ("col = " ++ show (I# col) ++ ", layout: inserting '}'") $
534 cont ITvccurly buf s{atbol = 1#, context = tail ctx}
535 else if need_semi_colon then
536 --trace ("col = " ++ show (I# col) ++ ", layout: inserting ';'") $
537 cont ITsemi buf s{atbol = 0#}
539 lexToken cont glaexts buf s{atbol = 0#}
541 col = currentIndex# buf -# bol
554 Layout n -> col ==# n
557 lexToken :: (Token -> P a) -> Int# -> P a
558 lexToken cont glaexts buf =
559 -- trace "lexToken" $
560 case currentChar# buf of
562 -- special symbols ----------------------------------------------------
563 '('# | flag glaexts && lookAhead# buf 1# `eqChar#` '#'#
564 -> cont IToubxparen (setCurrentPos# buf 2#)
566 -> cont IToparen (incLexeme buf)
568 ')'# -> cont ITcparen (incLexeme buf)
569 '['# -> cont ITobrack (incLexeme buf)
570 ']'# -> cont ITcbrack (incLexeme buf)
571 ','# -> cont ITcomma (incLexeme buf)
572 ';'# -> cont ITsemi (incLexeme buf)
573 '}'# -> \ s@PState{context = ctx} ->
575 (_:ctx') -> cont ITccurly (incLexeme buf) s{context=ctx'}
576 _ -> lexError "too many '}'s" buf s
577 '|'# -> case lookAhead# buf 1# of
578 '}'# | flag glaexts -> cont ITccurlybar
579 (setCurrentPos# buf 2#)
580 _ -> lex_sym cont (incLexeme buf)
583 '#'# -> case lookAhead# buf 1# of
584 ')'# | flag glaexts -> cont ITcubxparen (setCurrentPos# buf 2#)
585 '-'# -> case lookAhead# buf 2# of
586 '}'# -> cont ITclose_prag (setCurrentPos# buf 3#)
587 _ -> lex_sym cont (incLexeme buf)
588 _ -> lex_sym cont (incLexeme buf)
590 '`'# | flag glaexts && lookAhead# buf 1# `eqChar#` '`'#
591 -> lex_cstring cont (setCurrentPos# buf 2#)
593 -> cont ITbackquote (incLexeme buf)
595 '{'# -> -- look for "{-##" special iface pragma
596 case lookAhead# buf 1# of
598 -> cont ITocurlybar (setCurrentPos# buf 2#)
599 '-'# -> case lookAhead# buf 2# of
600 '#'# -> case lookAhead# buf 3# of
603 = doDiscard 0# (stepOnBy# (stepOverLexeme buf) 4#) in
604 cont (ITpragma lexeme) buf'
605 _ -> lex_prag cont (setCurrentPos# buf 3#)
606 _ -> cont ITocurly (incLexeme buf)
607 _ -> (layoutOff `thenP_` cont ITocurly) (incLexeme buf)
609 -- strings/characters -------------------------------------------------
610 '\"'#{-"-} -> lex_string cont glaexts [] (incLexeme buf)
611 '\''# -> lex_char (char_end cont) glaexts (incLexeme buf)
613 -- strictness and cpr pragmas and __scc treated specially.
614 '_'# | flag glaexts ->
615 case lookAhead# buf 1# of
616 '_'# -> case lookAhead# buf 2# of
618 lex_demand cont (stepOnUntil (not . isSpace)
619 (stepOnBy# buf 3#)) -- past __S
621 cont ITcprinfo (stepOnBy# buf 3#) -- past __M
624 case prefixMatch (stepOnBy# buf 3#) "cc" of
625 Just buf' -> lex_scc cont (stepOverLexeme buf')
626 Nothing -> lex_id cont glaexts buf
627 _ -> lex_id cont glaexts buf
628 _ -> lex_id cont glaexts buf
630 -- Hexadecimal and octal constants
631 '0'# | (ch `eqChar#` 'x'# || ch `eqChar#` 'X'#) && is_hexdigit ch2
632 -> readNum (after_lexnum cont glaexts) buf' is_hexdigit 16 hex
633 | (ch `eqChar#` 'o'# || ch `eqChar#` 'O'#) && is_octdigit ch2
634 -> readNum (after_lexnum cont glaexts) buf' is_octdigit 8 oct_or_dec
635 where ch = lookAhead# buf 1#
636 ch2 = lookAhead# buf 2#
637 buf' = setCurrentPos# buf 2#
640 if bufferExhausted (stepOn buf) then
643 trace "lexIface: misplaced NUL?" $
644 cont (ITunknown "\NUL") (stepOn buf)
646 '?'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
647 lex_ip cont (incLexeme buf)
648 c | is_digit c -> lex_num cont glaexts 0 buf
649 | is_symbol c -> lex_sym cont buf
650 | is_upper c -> lex_con cont glaexts buf
651 | is_ident c -> lex_id cont glaexts buf
652 | otherwise -> lexError "illegal character" buf
654 -- Int# is unlifted, and therefore faster than Bool for flags.
660 -------------------------------------------------------------------------------
664 = case expandWhile# is_space buf of { buf1 ->
665 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2 ->
666 let lexeme = mkFastString (map toUpper (lexemeToString buf2)) in
667 case lookupUFM pragmaKeywordsFM lexeme of
668 Just kw -> cont kw (mergeLexemes buf buf2)
669 Nothing -> panic "lex_prag"
672 -------------------------------------------------------------------------------
675 lex_string cont glaexts s buf
676 = case currentChar# buf of
678 let buf' = incLexeme buf
679 s' = mkFastStringNarrow (map chr (reverse s))
680 in case currentChar# buf' of
681 '#'# | flag glaexts -> if all (<= 0xFF) s
682 then cont (ITprimstring s') (incLexeme buf')
683 else lexError "primitive string literal must contain only characters <= \'\\xFF\'" buf'
684 _ -> cont (ITstring s') buf'
686 -- ignore \& in a string, deal with string gaps
687 '\\'# | next_ch `eqChar#` '&'#
688 -> lex_string cont glaexts s buf'
690 -> lex_stringgap cont glaexts s (incLexeme buf)
692 where next_ch = lookAhead# buf 1#
693 buf' = setCurrentPos# buf 2#
695 _ -> lex_char (lex_next_string cont s) glaexts buf
697 lex_stringgap cont glaexts s buf
698 = let buf' = incLexeme buf in
699 case currentChar# buf of
700 '\n'# -> \st@PState{loc = loc} -> lex_stringgap cont glaexts s buf'
701 st{loc = incSrcLine loc}
702 '\\'# -> lex_string cont glaexts s buf'
703 c | is_space c -> lex_stringgap cont glaexts s buf'
704 other -> charError buf'
706 lex_next_string cont s glaexts c buf = lex_string cont glaexts (c:s) buf
708 lex_char :: (Int# -> Int -> P a) -> Int# -> P a
709 lex_char cont glaexts buf
710 = case currentChar# buf of
711 '\\'# -> lex_escape (cont glaexts) (incLexeme buf)
712 c | is_any c -> cont glaexts (I# (ord# c)) (incLexeme buf)
713 other -> charError buf
715 char_end cont glaexts c buf
716 = case currentChar# buf of
717 '\''# -> let buf' = incLexeme buf in
718 case currentChar# buf' of
720 -> cont (ITprimchar c) (incLexeme buf')
721 _ -> cont (ITchar c) buf'
725 = let buf' = incLexeme buf in
726 case currentChar# buf of
727 'a'# -> cont (ord '\a') buf'
728 'b'# -> cont (ord '\b') buf'
729 'f'# -> cont (ord '\f') buf'
730 'n'# -> cont (ord '\n') buf'
731 'r'# -> cont (ord '\r') buf'
732 't'# -> cont (ord '\t') buf'
733 'v'# -> cont (ord '\v') buf'
734 '\\'# -> cont (ord '\\') buf'
735 '"'# -> cont (ord '\"') buf'
736 '\''# -> cont (ord '\'') buf'
737 '^'# -> let c = currentChar# buf' in
738 if c `geChar#` '@'# && c `leChar#` '_'#
739 then cont (I# (ord# c -# ord# '@'#)) (incLexeme buf')
742 'x'# -> readNum (after_charnum cont) buf' is_hexdigit 16 hex
743 'o'# -> readNum (after_charnum cont) buf' is_octdigit 8 oct_or_dec
745 -> readNum (after_charnum cont) buf is_digit 10 oct_or_dec
747 _ -> case [ (c,buf2) | (p,c) <- silly_escape_chars,
748 Just buf2 <- [prefixMatch buf p] ] of
749 (c,buf2):_ -> cont (ord c) buf2
752 after_charnum cont i buf
753 = if i >= 0 && i <= 0x10FFFF
754 then cont (fromInteger i) buf
757 readNum cont buf is_digit base conv = read buf 0
759 = case currentChar# buf of { c ->
761 then read (incLexeme buf) (i*base + (toInteger (I# (conv c))))
767 || (c `geChar#` 'a'# && c `leChar#` 'f'#)
768 || (c `geChar#` 'A'# && c `leChar#` 'F'#)
770 hex c | is_digit c = ord# c -# ord# '0'#
771 | otherwise = ord# (to_lower c) -# ord# 'a'# +# 10#
772 oct_or_dec c = ord# c -# ord# '0'#
774 is_octdigit c = c `geChar#` '0'# && c `leChar#` '7'#
777 | c `geChar#` 'A'# && c `leChar#` 'Z'#
778 = chr# (ord# c -# (ord# 'A'# -# ord# 'a'#))
781 charError buf = lexError "error in character literal" buf
783 silly_escape_chars = [
820 -------------------------------------------------------------------------------
822 lex_demand cont buf =
823 case read_em [] buf of { (ls,buf') ->
824 case currentChar# buf' of
825 'b'# -> cont (ITstrict (StrictSig (mkTopDmdType ls BotRes))) (incLexeme buf')
826 'm'# -> cont (ITstrict (StrictSig (mkTopDmdType ls RetCPR))) (incLexeme buf')
827 _ -> cont (ITstrict (StrictSig (mkTopDmdType ls TopRes))) buf'
831 case currentChar# buf of
832 'L'# -> read_em (Lazy : acc) (stepOn buf)
833 'A'# -> read_em (Abs : acc) (stepOn buf)
834 'V'# -> read_em (Eval : acc) (stepOn buf)
835 'X'# -> read_em (Err : acc) (stepOn buf)
836 'B'# -> read_em (Bot : acc) (stepOn buf)
837 ')'# -> (reverse acc, stepOn buf)
838 'C'# -> do_call acc (stepOnBy# buf 2#)
839 'D'# -> do_unpack1 Defer acc (stepOnBy# buf 1#)
840 'U'# -> do_unpack1 Drop acc (stepOnBy# buf 1#)
841 'S'# -> do_unpack1 Keep acc (stepOnBy# buf 1#)
842 _ -> (reverse acc, buf)
844 do_unpack1 keepity acc buf
845 = case currentChar# buf of
846 '('# -> do_unpack2 keepity acc (stepOnBy# buf 1#)
847 _ -> read_em (Seq keepity [] : acc) buf
849 do_unpack2 keepity acc buf
850 = case read_em [] buf of
851 (stuff, rest) -> read_em (Seq keepity stuff : acc) rest
854 = case read_em [] buf of
855 ([dmd], rest) -> read_em (Call dmd : acc) rest
859 case currentChar# buf of
860 'C'# -> cont ITsccAllCafs (incLexeme buf)
861 other -> cont ITscc buf
863 -----------------------------------------------------------------------------
866 lex_num :: (Token -> P a) -> Int# -> Integer -> P a
867 lex_num cont glaexts acc buf =
868 case scanNumLit acc buf of
870 case currentChar# buf' of
871 '.'# | is_digit (lookAhead# buf' 1#) ->
872 -- this case is not optimised at all, as the
873 -- presence of floating point numbers in interface
874 -- files is not that common. (ToDo)
875 case expandWhile# is_digit (incLexeme buf') of
876 buf2 -> -- points to first non digit char
878 let l = case currentChar# buf2 of
884 = let buf3 = incLexeme buf2 in
885 case currentChar# buf3 of
886 '-'# -> expandWhile# is_digit (incLexeme buf3)
887 '+'# -> expandWhile# is_digit (incLexeme buf3)
888 x | is_digit x -> expandWhile# is_digit buf3
891 v = readRational__ (lexemeToString l)
893 in case currentChar# l of -- glasgow exts only
894 '#'# | flag glaexts -> let l' = incLexeme l in
895 case currentChar# l' of
896 '#'# -> cont (ITprimdouble v) (incLexeme l')
897 _ -> cont (ITprimfloat v) l'
898 _ -> cont (ITrational v) l
900 _ -> after_lexnum cont glaexts acc' buf'
902 after_lexnum cont glaexts i buf
903 = case currentChar# buf of
904 '#'# | flag glaexts -> cont (ITprimint i) (incLexeme buf)
905 _ -> cont (ITinteger i) buf
907 -----------------------------------------------------------------------------
908 -- C "literal literal"s (i.e. things like ``NULL'', ``stdout'' etc.)
910 -- we lexemeToFastString on the bit between the ``''s, but include the
911 -- quotes in the full lexeme.
913 lex_cstring cont buf =
914 case expandUntilMatch (stepOverLexeme buf) "\'\'" of
915 Just buf' -> cont (ITlitlit (lexemeToFastString
916 (setCurrentPos# buf' (negateInt# 2#))))
917 (mergeLexemes buf buf')
918 Nothing -> lexError "unterminated ``" buf
920 -----------------------------------------------------------------------------
921 -- identifiers, symbols etc.
924 case expandWhile# is_ident buf of
925 buf' -> cont (ITipvarid lexeme) buf'
926 where lexeme = lexemeToFastString buf'
928 lex_id cont glaexts buf =
929 let buf1 = expandWhile# is_ident buf in
932 case (if flag glaexts
933 then expandWhile# (eqChar# '#'#) buf1 -- slurp trailing hashes
934 else buf1) of { buf' ->
936 let lexeme = lexemeToFastString buf' in
938 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
939 Just kwd_token -> --trace ("hkeywd: "++_UNPK_(lexeme)) $
943 let var_token = cont (ITvarid lexeme) buf' in
945 if not (flag glaexts)
949 case lookupUFM ghcExtensionKeywordsFM lexeme of {
950 Just kwd_token -> cont kwd_token buf';
957 case expandWhile# is_symbol buf of
958 buf' -> case lookupUFM haskellKeySymsFM lexeme of {
959 Just kwd_token -> --trace ("keysym: "++unpackFS lexeme) $
960 cont kwd_token buf' ;
961 Nothing -> --trace ("sym: "++unpackFS lexeme) $
962 cont (mk_var_token lexeme) buf'
964 where lexeme = lexemeToFastString buf'
967 -- lex_con recursively collects components of a qualified identifer.
968 -- The argument buf is the StringBuffer representing the lexeme
969 -- identified so far, where the next character is upper-case.
971 lex_con cont glaexts buf =
972 -- trace ("con: "{-++unpackFS lexeme-}) $
973 let empty_buf = stepOverLexeme buf in
974 case expandWhile# is_ident empty_buf of { buf1 ->
975 case slurp_trailing_hashes buf1 glaexts of { con_buf ->
977 let all_buf = mergeLexemes buf con_buf
979 con_lexeme = lexemeToFastString con_buf
980 mod_lexeme = lexemeToFastString (decLexeme buf)
981 all_lexeme = lexemeToFastString all_buf
984 | emptyLexeme buf = cont (ITconid con_lexeme) all_buf
985 | otherwise = cont (ITqconid (mod_lexeme,con_lexeme)) all_buf
988 case currentChar# all_buf of
989 '.'# -> maybe_qualified cont glaexts all_lexeme
990 (incLexeme all_buf) just_a_conid
995 maybe_qualified cont glaexts mod buf just_a_conid =
996 -- trace ("qid: "{-++unpackFS lexeme-}) $
997 case currentChar# buf of
998 '['# -> -- Special case for []
999 case lookAhead# buf 1# of
1000 ']'# -> cont (ITqconid (mod,SLIT("[]"))) (setCurrentPos# buf 2#)
1003 '('# -> -- Special case for (,,,)
1004 -- This *is* necessary to deal with e.g. "instance C PrelBase.(,,)"
1005 case lookAhead# buf 1# of
1006 '#'# | flag glaexts -> case lookAhead# buf 2# of
1007 ','# -> lex_ubx_tuple cont mod (setCurrentPos# buf 3#)
1010 ')'# -> cont (ITqconid (mod,SLIT("()"))) (setCurrentPos# buf 2#)
1011 ','# -> lex_tuple cont mod (setCurrentPos# buf 2#) just_a_conid
1014 '-'# -> case lookAhead# buf 1# of
1015 '>'# -> cont (ITqconid (mod,SLIT("(->)"))) (setCurrentPos# buf 2#)
1016 _ -> lex_id3 cont glaexts mod buf just_a_conid
1018 _ -> lex_id3 cont glaexts mod buf just_a_conid
1021 lex_id3 cont glaexts mod buf just_a_conid
1022 | is_upper (currentChar# buf) =
1023 lex_con cont glaexts buf
1025 | is_symbol (currentChar# buf) =
1027 start_new_lexeme = stepOverLexeme buf
1029 -- trace ("lex_id31 "{-++unpackFS lexeme-}) $
1030 case expandWhile# is_symbol start_new_lexeme of { buf' ->
1032 lexeme = lexemeToFastString buf'
1033 -- real lexeme is M.<sym>
1034 new_buf = mergeLexemes buf buf'
1036 cont (mk_qvar_token mod lexeme) new_buf
1037 -- wrong, but arguably morally right: M... is now a qvarsym
1042 start_new_lexeme = stepOverLexeme buf
1044 -- trace ("lex_id32 "{-++unpackFS lexeme-}) $
1045 case expandWhile# is_ident start_new_lexeme of { buf1 ->
1050 case slurp_trailing_hashes buf1 glaexts of { buf' ->
1053 lexeme = lexemeToFastString buf'
1054 new_buf = mergeLexemes buf buf'
1055 is_a_qvarid = cont (mk_qvar_token mod lexeme) new_buf
1057 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
1058 Nothing -> is_a_qvarid ;
1060 Just kwd_token | isSpecial kwd_token -- special ids (as, qualified, hiding) shouldn't be
1061 -> is_a_qvarid -- recognised as keywords here.
1063 -> just_a_conid -- avoid M.where etc.
1066 slurp_trailing_hashes buf glaexts
1067 | flag glaexts = expandWhile# (`eqChar#` '#'#) buf
1072 | is_upper f = ITconid pk_str
1073 | is_ident f = ITvarid pk_str
1074 | f `eqChar#` ':'# = ITconsym pk_str
1075 | otherwise = ITvarsym pk_str
1077 (C# f) = _HEAD_ pk_str
1078 -- tl = _TAIL_ pk_str
1080 mk_qvar_token m token =
1081 -- trace ("mk_qvar ") $
1082 case mk_var_token token of
1083 ITconid n -> ITqconid (m,n)
1084 ITvarid n -> ITqvarid (m,n)
1085 ITconsym n -> ITqconsym (m,n)
1086 ITvarsym n -> ITqvarsym (m,n)
1087 _ -> ITunknown (show token)
1090 ----------------------------------------------------------------------------
1091 Horrible stuff for dealing with M.(,,,)
1094 lex_tuple cont mod buf back_off =
1098 case currentChar# buf of
1099 ','# -> go (n+1) (stepOn buf)
1100 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Boxed n))) (stepOn buf)
1103 lex_ubx_tuple cont mod buf back_off =
1107 case currentChar# buf of
1108 ','# -> go (n+1) (stepOn buf)
1109 '#'# -> case lookAhead# buf 1# of
1110 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Unboxed n)))
1116 -----------------------------------------------------------------------------
1117 doDiscard rips along really fast, looking for a '##-}',
1118 indicating the end of the pragma we're skipping
1121 doDiscard inStr buf =
1122 case currentChar# buf of
1123 '#'# | inStr ==# 0# ->
1124 case lookAhead# buf 1# of { '#'# ->
1125 case lookAhead# buf 2# of { '-'# ->
1126 case lookAhead# buf 3# of { '}'# ->
1127 (lexemeToBuffer buf, stepOverLexeme (setCurrentPos# buf 4#));
1128 _ -> doDiscard inStr (incLexeme buf) };
1129 _ -> doDiscard inStr (incLexeme buf) };
1130 _ -> doDiscard inStr (incLexeme buf) }
1134 odd_slashes buf flg i# =
1135 case lookAhead# buf i# of
1136 '\\'# -> odd_slashes buf (not flg) (i# -# 1#)
1139 not_inStr = if inStr ==# 0# then 1# else 0#
1141 case lookAhead# buf (negateInt# 1#) of --backwards, actually
1142 '\\'# -> -- escaping something..
1143 if odd_slashes buf True (negateInt# 2#)
1144 then -- odd number of slashes, " is escaped.
1145 doDiscard inStr (incLexeme buf)
1146 else -- even number of slashes, \ is escaped.
1147 doDiscard not_inStr (incLexeme buf)
1148 _ -> doDiscard not_inStr (incLexeme buf)
1150 '\''# | inStr ==# 0# ->
1151 case lookAhead# buf 1# of { '"'# ->
1152 case lookAhead# buf 2# of { '\''# ->
1153 doDiscard inStr (setCurrentPos# buf 3#);
1154 _ -> doDiscard inStr (incLexeme buf) };
1155 _ -> doDiscard inStr (incLexeme buf) }
1157 _ -> doDiscard inStr (incLexeme buf)
1161 -----------------------------------------------------------------------------
1172 data PState = PState {
1174 glasgow_exts :: Int#,
1177 context :: [LayoutContext]
1180 type P a = StringBuffer -- Input string
1185 returnP a buf s = POk s a
1187 thenP :: P a -> (a -> P b) -> P b
1188 m `thenP` k = \ buf s ->
1190 POk s1 a -> k a buf s1
1191 PFailed err -> PFailed err
1193 thenP_ :: P a -> P b -> P b
1194 m `thenP_` k = m `thenP` \_ -> k
1196 mapP :: (a -> P b) -> [a] -> P [b]
1197 mapP f [] = returnP []
1200 mapP f as `thenP` \bs ->
1203 failP :: String -> P a
1204 failP msg buf s = PFailed (text msg)
1206 failMsgP :: Message -> P a
1207 failMsgP msg buf s = PFailed msg
1209 lexError :: String -> P a
1210 lexError str buf s@PState{ loc = loc }
1211 = failMsgP (hcat [ppr loc, text ": ", text str]) buf s
1213 getSrcLocP :: P SrcLoc
1214 getSrcLocP buf s@(PState{ loc = loc }) = POk s loc
1216 -- use a temporary SrcLoc for the duration of the argument
1217 setSrcLocP :: SrcLoc -> P a -> P a
1218 setSrcLocP new_loc p buf s =
1219 case p buf s{ loc=new_loc } of
1221 PFailed e -> PFailed e
1223 getSrcFile :: P FAST_STRING
1224 getSrcFile buf s@(PState{ loc = loc }) = POk s (srcLocFile loc)
1226 pushContext :: LayoutContext -> P ()
1227 pushContext ctxt buf s@(PState{ context = ctx }) = POk s{context = ctxt:ctx} ()
1231 This special case in layoutOn is to handle layout contexts with are
1232 indented the same or less than the current context. This is illegal
1233 according to the Haskell spec, so we have to arrange to close the
1234 current context. eg.
1239 after the first 'where', the sequence of events is:
1241 - layout system inserts a ';' (column 0)
1242 - parser begins a new context at column 0
1243 - parser shifts ';' (legal empty declaration)
1244 - parser sees 'class': parse error (we're still in the inner context)
1246 trouble is, by the time we know we need a new context, the lexer has
1247 already generated the ';'. Hacky solution is as follows: since we
1248 know the column of the next token (it's the column number of the new
1249 context), we set the ACTUAL column number of the new context to this
1250 numer plus one. Hence the next time the lexer is called, a '}' will
1251 be generated to close the new context straight away. Furthermore, we
1252 have to set the atbol flag so that the ';' that the parser shifted as
1253 part of the new context is re-generated.
1255 when the new context is *less* indented than the current one:
1257 f = f where g = g where
1260 - current context: column 12.
1261 - on seeing 'h' (column 0), the layout system inserts '}'
1262 - parser starts a new context, column 0
1263 - parser sees '}', uses it to close new context
1264 - we still need to insert another '}' followed by a ';',
1265 hence the atbol trick.
1267 There's also a special hack in here to deal with
1274 i.e. the inner context is at the same indentation level as the outer
1275 context. This is strictly illegal according to Haskell 98, but
1276 there's a lot of existing code using this style and it doesn't make
1277 any sense to disallow it, since empty 'do' lists don't make sense.
1280 layoutOn :: Bool -> P ()
1281 layoutOn strict buf s@(PState{ bol = bol, context = ctx }) =
1282 let offset = lexemeIndex buf -# bol in
1285 | if strict then prev_off >=# offset else prev_off ># offset ->
1286 --trace ("layout on, column: " ++ show (I# offset)) $
1287 POk s{ context = Layout (offset +# 1#) : ctx, atbol = 1# } ()
1289 --trace ("layout on, column: " ++ show (I# offset)) $
1290 POk s{ context = Layout offset : ctx } ()
1293 layoutOff buf s@(PState{ context = ctx }) =
1294 POk s{ context = NoLayout:ctx } ()
1297 popContext = \ buf s@(PState{ context = ctx, loc = loc }) ->
1299 (_:tl) -> POk s{ context = tl } ()
1300 [] -> PFailed (srcParseErr buf loc)
1303 Note that if the name of the file we're processing ends
1304 with `hi-boot', we accept it on faith as having the right
1305 version. This is done so that .hi-boot files that comes
1306 with hsc don't have to be updated before every release,
1307 *and* it allows us to share .hi-boot files with versions
1308 of hsc that don't have .hi version checking (e.g., ghc-2.10's)
1310 If the version number is 0, the checking is also turned off.
1311 (needed to deal with GHC.hi only!)
1313 Once we can assume we're compiling with a version of ghc that
1314 supports interface file checking, we can drop the special
1317 checkVersion :: Maybe Integer -> P ()
1318 checkVersion mb@(Just v) buf s@(PState{loc = loc})
1319 | (v==0) || (v == fromInt opt_HiVersion) || opt_NoHiCheck = POk s ()
1320 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1321 checkVersion mb@Nothing buf s@(PState{loc = loc})
1322 | "hi-boot" `isSuffixOf` (_UNPK_ (srcLocFile loc)) = POk s ()
1323 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1325 -----------------------------------------------------------------
1327 ifaceParseErr :: StringBuffer -> SrcLoc -> Message
1329 = hsep [ppr l, ptext SLIT("Interface file parse error; on input `"),
1330 text (lexemeToString s), char '\'']
1332 ifaceVersionErr hi_vers l toks
1333 = hsep [ppr l, ptext SLIT("Interface file version error;"),
1334 ptext SLIT("Expected"), int opt_HiVersion,
1335 ptext SLIT("found "), pp_version]
1339 Nothing -> ptext SLIT("pre ghc-3.02 version")
1340 Just v -> ptext SLIT("version") <+> integer v
1342 -----------------------------------------------------------------------------
1344 srcParseErr :: StringBuffer -> SrcLoc -> Message
1348 then ptext SLIT(": parse error (possibly incorrect indentation)")
1349 else hcat [ptext SLIT(": parse error on input "),
1350 char '`', text token, char '\'']
1353 token = lexemeToString s