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 orig_state@PState{loc=loc} = loop buf orig_state
504 case currentChar# buf of
505 '-'# | lookAhead# buf 1# `eqChar#` '}'# -> cont (stepOnBy# buf 2#)
507 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
508 nested_comment (nested_comment cont) (stepOnBy# buf 2#)
510 '\n'# -> \ s@PState{loc=loc} ->
511 let buf' = stepOn buf in
512 loop buf' s{loc = incSrcLine loc,
513 bol = currentIndex# buf',
516 -- pass the original state to lexError so that the error is
517 -- reported at the line it was originally on, not the line at
518 -- the end of the file.
519 '\NUL'# | bufferExhausted (stepOn buf) ->
520 \_ -> lexError "unterminated `{-'" buf orig_state -- -}
522 _ -> loop (stepOn buf)
524 -- When we are lexing the first token of a line, check whether we need to
525 -- insert virtual semicolons or close braces due to layout.
527 lexBOL :: (Token -> P a) -> P a
528 lexBOL cont buf s@(PState{
530 glasgow_exts = glaexts,
535 if need_close_curly then
536 --trace ("col = " ++ show (I# col) ++ ", layout: inserting '}'") $
537 cont ITvccurly buf s{atbol = 1#, context = tail ctx}
538 else if need_semi_colon then
539 --trace ("col = " ++ show (I# col) ++ ", layout: inserting ';'") $
540 cont ITsemi buf s{atbol = 0#}
542 lexToken cont glaexts buf s{atbol = 0#}
544 col = currentIndex# buf -# bol
557 Layout n -> col ==# n
560 lexToken :: (Token -> P a) -> Int# -> P a
561 lexToken cont glaexts buf =
562 -- trace "lexToken" $
563 case currentChar# buf of
565 -- special symbols ----------------------------------------------------
566 '('# | flag glaexts && lookAhead# buf 1# `eqChar#` '#'#
567 -> cont IToubxparen (setCurrentPos# buf 2#)
569 -> cont IToparen (incLexeme buf)
571 ')'# -> cont ITcparen (incLexeme buf)
572 '['# -> cont ITobrack (incLexeme buf)
573 ']'# -> cont ITcbrack (incLexeme buf)
574 ','# -> cont ITcomma (incLexeme buf)
575 ';'# -> cont ITsemi (incLexeme buf)
576 '}'# -> \ s@PState{context = ctx} ->
578 (_:ctx') -> cont ITccurly (incLexeme buf) s{context=ctx'}
579 _ -> lexError "too many '}'s" buf s
580 '|'# -> case lookAhead# buf 1# of
581 '}'# | flag glaexts -> cont ITccurlybar
582 (setCurrentPos# buf 2#)
583 _ -> lex_sym cont (incLexeme buf)
586 '#'# -> case lookAhead# buf 1# of
587 ')'# | flag glaexts -> cont ITcubxparen (setCurrentPos# buf 2#)
588 '-'# -> case lookAhead# buf 2# of
589 '}'# -> cont ITclose_prag (setCurrentPos# buf 3#)
590 _ -> lex_sym cont (incLexeme buf)
591 _ -> lex_sym cont (incLexeme buf)
593 '`'# | flag glaexts && lookAhead# buf 1# `eqChar#` '`'#
594 -> lex_cstring cont (setCurrentPos# buf 2#)
596 -> cont ITbackquote (incLexeme buf)
598 '{'# -> -- look for "{-##" special iface pragma
599 case lookAhead# buf 1# of
601 -> cont ITocurlybar (setCurrentPos# buf 2#)
602 '-'# -> case lookAhead# buf 2# of
603 '#'# -> case lookAhead# buf 3# of
606 = doDiscard 0# (stepOnBy# (stepOverLexeme buf) 4#) in
607 cont (ITpragma lexeme) buf'
608 _ -> lex_prag cont (setCurrentPos# buf 3#)
609 _ -> cont ITocurly (incLexeme buf)
610 _ -> (layoutOff `thenP_` cont ITocurly) (incLexeme buf)
612 -- strings/characters -------------------------------------------------
613 '\"'#{-"-} -> lex_string cont glaexts [] (incLexeme buf)
614 '\''# -> lex_char (char_end cont) glaexts (incLexeme buf)
616 -- strictness and cpr pragmas and __scc treated specially.
617 '_'# | flag glaexts ->
618 case lookAhead# buf 1# of
619 '_'# -> case lookAhead# buf 2# of
621 lex_demand cont (stepOnUntil (not . isSpace)
622 (stepOnBy# buf 3#)) -- past __S
624 cont ITcprinfo (stepOnBy# buf 3#) -- past __M
627 case prefixMatch (stepOnBy# buf 3#) "cc" of
628 Just buf' -> lex_scc cont (stepOverLexeme buf')
629 Nothing -> lex_id cont glaexts buf
630 _ -> lex_id cont glaexts buf
631 _ -> lex_id cont glaexts buf
633 -- Hexadecimal and octal constants
634 '0'# | (ch `eqChar#` 'x'# || ch `eqChar#` 'X'#) && is_hexdigit ch2
635 -> readNum (after_lexnum cont glaexts) buf' is_hexdigit 16 hex
636 | (ch `eqChar#` 'o'# || ch `eqChar#` 'O'#) && is_octdigit ch2
637 -> readNum (after_lexnum cont glaexts) buf' is_octdigit 8 oct_or_dec
638 where ch = lookAhead# buf 1#
639 ch2 = lookAhead# buf 2#
640 buf' = setCurrentPos# buf 2#
643 if bufferExhausted (stepOn buf) then
646 trace "lexIface: misplaced NUL?" $
647 cont (ITunknown "\NUL") (stepOn buf)
649 '?'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
650 lex_ip cont (incLexeme buf)
651 c | is_digit c -> lex_num cont glaexts 0 buf
652 | is_symbol c -> lex_sym cont buf
653 | is_upper c -> lex_con cont glaexts buf
654 | is_ident c -> lex_id cont glaexts buf
655 | otherwise -> lexError "illegal character" buf
657 -- Int# is unlifted, and therefore faster than Bool for flags.
663 -------------------------------------------------------------------------------
667 = case expandWhile# is_space buf of { buf1 ->
668 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2 ->
669 let lexeme = mkFastString (map toUpper (lexemeToString buf2)) in
670 case lookupUFM pragmaKeywordsFM lexeme of
671 Just kw -> cont kw (mergeLexemes buf buf2)
672 Nothing -> panic "lex_prag"
675 -------------------------------------------------------------------------------
678 lex_string cont glaexts s buf
679 = case currentChar# buf of
681 let buf' = incLexeme buf
682 s' = mkFastStringNarrow (map chr (reverse s))
683 in case currentChar# buf' of
684 '#'# | flag glaexts -> if all (<= 0xFF) s
685 then cont (ITprimstring s') (incLexeme buf')
686 else lexError "primitive string literal must contain only characters <= \'\\xFF\'" buf'
687 _ -> cont (ITstring s') buf'
689 -- ignore \& in a string, deal with string gaps
690 '\\'# | next_ch `eqChar#` '&'#
691 -> lex_string cont glaexts s buf'
693 -> lex_stringgap cont glaexts s (incLexeme buf)
695 where next_ch = lookAhead# buf 1#
696 buf' = setCurrentPos# buf 2#
698 _ -> lex_char (lex_next_string cont s) glaexts buf
700 lex_stringgap cont glaexts s buf
701 = let buf' = incLexeme buf in
702 case currentChar# buf of
703 '\n'# -> \st@PState{loc = loc} -> lex_stringgap cont glaexts s buf'
704 st{loc = incSrcLine loc}
705 '\\'# -> lex_string cont glaexts s buf'
706 c | is_space c -> lex_stringgap cont glaexts s buf'
707 other -> charError buf'
709 lex_next_string cont s glaexts c buf = lex_string cont glaexts (c:s) buf
711 lex_char :: (Int# -> Int -> P a) -> Int# -> P a
712 lex_char cont glaexts buf
713 = case currentChar# buf of
714 '\\'# -> lex_escape (cont glaexts) (incLexeme buf)
715 c | is_any c -> cont glaexts (I# (ord# c)) (incLexeme buf)
716 other -> charError buf
718 char_end cont glaexts c buf
719 = case currentChar# buf of
720 '\''# -> let buf' = incLexeme buf in
721 case currentChar# buf' of
723 -> cont (ITprimchar c) (incLexeme buf')
724 _ -> cont (ITchar c) buf'
728 = let buf' = incLexeme buf in
729 case currentChar# buf of
730 'a'# -> cont (ord '\a') buf'
731 'b'# -> cont (ord '\b') buf'
732 'f'# -> cont (ord '\f') buf'
733 'n'# -> cont (ord '\n') buf'
734 'r'# -> cont (ord '\r') buf'
735 't'# -> cont (ord '\t') buf'
736 'v'# -> cont (ord '\v') buf'
737 '\\'# -> cont (ord '\\') buf'
738 '"'# -> cont (ord '\"') buf'
739 '\''# -> cont (ord '\'') buf'
740 '^'# -> let c = currentChar# buf' in
741 if c `geChar#` '@'# && c `leChar#` '_'#
742 then cont (I# (ord# c -# ord# '@'#)) (incLexeme buf')
745 'x'# -> readNum (after_charnum cont) buf' is_hexdigit 16 hex
746 'o'# -> readNum (after_charnum cont) buf' is_octdigit 8 oct_or_dec
748 -> readNum (after_charnum cont) buf is_digit 10 oct_or_dec
750 _ -> case [ (c,buf2) | (p,c) <- silly_escape_chars,
751 Just buf2 <- [prefixMatch buf p] ] of
752 (c,buf2):_ -> cont (ord c) buf2
755 after_charnum cont i buf
756 = if i >= 0 && i <= 0x10FFFF
757 then cont (fromInteger i) buf
760 readNum cont buf is_digit base conv = read buf 0
762 = case currentChar# buf of { c ->
764 then read (incLexeme buf) (i*base + (toInteger (I# (conv c))))
770 || (c `geChar#` 'a'# && c `leChar#` 'f'#)
771 || (c `geChar#` 'A'# && c `leChar#` 'F'#)
773 hex c | is_digit c = ord# c -# ord# '0'#
774 | otherwise = ord# (to_lower c) -# ord# 'a'# +# 10#
775 oct_or_dec c = ord# c -# ord# '0'#
777 is_octdigit c = c `geChar#` '0'# && c `leChar#` '7'#
780 | c `geChar#` 'A'# && c `leChar#` 'Z'#
781 = chr# (ord# c -# (ord# 'A'# -# ord# 'a'#))
784 charError buf = lexError "error in character literal" buf
786 silly_escape_chars = [
823 -------------------------------------------------------------------------------
825 lex_demand cont buf =
826 case read_em [] buf of { (ls,buf') ->
827 case currentChar# buf' of
828 'b'# -> cont (ITstrict (StrictSig (mkTopDmdType ls BotRes))) (incLexeme buf')
829 'm'# -> cont (ITstrict (StrictSig (mkTopDmdType ls RetCPR))) (incLexeme buf')
830 _ -> cont (ITstrict (StrictSig (mkTopDmdType ls TopRes))) buf'
834 case currentChar# buf of
835 'L'# -> read_em (Lazy : acc) (stepOn buf)
836 'A'# -> read_em (Abs : acc) (stepOn buf)
837 'V'# -> read_em (Eval : acc) (stepOn buf)
838 'X'# -> read_em (Err : acc) (stepOn buf)
839 'B'# -> read_em (Bot : acc) (stepOn buf)
840 ')'# -> (reverse acc, stepOn buf)
841 'C'# -> do_call acc (stepOnBy# buf 2#)
842 'D'# -> do_unpack1 Defer acc (stepOnBy# buf 1#)
843 'U'# -> do_unpack1 Drop acc (stepOnBy# buf 1#)
844 'S'# -> do_unpack1 Keep acc (stepOnBy# buf 1#)
845 _ -> (reverse acc, buf)
847 do_unpack1 keepity acc buf
848 = case currentChar# buf of
849 '('# -> do_unpack2 keepity acc (stepOnBy# buf 1#)
850 _ -> read_em (Seq keepity [] : acc) buf
852 do_unpack2 keepity acc buf
853 = case read_em [] buf of
854 (stuff, rest) -> read_em (Seq keepity stuff : acc) rest
857 = case read_em [] buf of
858 ([dmd], rest) -> read_em (Call dmd : acc) rest
862 case currentChar# buf of
863 'C'# -> cont ITsccAllCafs (incLexeme buf)
864 other -> cont ITscc buf
866 -----------------------------------------------------------------------------
869 lex_num :: (Token -> P a) -> Int# -> Integer -> P a
870 lex_num cont glaexts acc buf =
871 case scanNumLit acc buf of
873 case currentChar# buf' of
874 '.'# | is_digit (lookAhead# buf' 1#) ->
875 -- this case is not optimised at all, as the
876 -- presence of floating point numbers in interface
877 -- files is not that common. (ToDo)
878 case expandWhile# is_digit (incLexeme buf') of
879 buf2 -> -- points to first non digit char
881 let l = case currentChar# buf2 of
887 = let buf3 = incLexeme buf2 in
888 case currentChar# buf3 of
889 '-'# -> expandWhile# is_digit (incLexeme buf3)
890 '+'# -> expandWhile# is_digit (incLexeme buf3)
891 x | is_digit x -> expandWhile# is_digit buf3
894 v = readRational__ (lexemeToString l)
896 in case currentChar# l of -- glasgow exts only
897 '#'# | flag glaexts -> let l' = incLexeme l in
898 case currentChar# l' of
899 '#'# -> cont (ITprimdouble v) (incLexeme l')
900 _ -> cont (ITprimfloat v) l'
901 _ -> cont (ITrational v) l
903 _ -> after_lexnum cont glaexts acc' buf'
905 after_lexnum cont glaexts i buf
906 = case currentChar# buf of
907 '#'# | flag glaexts -> cont (ITprimint i) (incLexeme buf)
908 _ -> cont (ITinteger i) buf
910 -----------------------------------------------------------------------------
911 -- C "literal literal"s (i.e. things like ``NULL'', ``stdout'' etc.)
913 -- we lexemeToFastString on the bit between the ``''s, but include the
914 -- quotes in the full lexeme.
916 lex_cstring cont buf =
917 case expandUntilMatch (stepOverLexeme buf) "\'\'" of
918 Just buf' -> cont (ITlitlit (lexemeToFastString
919 (setCurrentPos# buf' (negateInt# 2#))))
920 (mergeLexemes buf buf')
921 Nothing -> lexError "unterminated ``" buf
923 -----------------------------------------------------------------------------
924 -- identifiers, symbols etc.
927 case expandWhile# is_ident buf of
928 buf' -> cont (ITipvarid lexeme) buf'
929 where lexeme = lexemeToFastString buf'
931 lex_id cont glaexts buf =
932 let buf1 = expandWhile# is_ident buf in
935 case (if flag glaexts
936 then expandWhile# (eqChar# '#'#) buf1 -- slurp trailing hashes
937 else buf1) of { buf' ->
939 let lexeme = lexemeToFastString buf' in
941 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
942 Just kwd_token -> --trace ("hkeywd: "++_UNPK_(lexeme)) $
946 let var_token = cont (ITvarid lexeme) buf' in
948 if not (flag glaexts)
952 case lookupUFM ghcExtensionKeywordsFM lexeme of {
953 Just kwd_token -> cont kwd_token buf';
960 case expandWhile# is_symbol buf of
961 buf' -> case lookupUFM haskellKeySymsFM lexeme of {
962 Just kwd_token -> --trace ("keysym: "++unpackFS lexeme) $
963 cont kwd_token buf' ;
964 Nothing -> --trace ("sym: "++unpackFS lexeme) $
965 cont (mk_var_token lexeme) buf'
967 where lexeme = lexemeToFastString buf'
970 -- lex_con recursively collects components of a qualified identifer.
971 -- The argument buf is the StringBuffer representing the lexeme
972 -- identified so far, where the next character is upper-case.
974 lex_con cont glaexts buf =
975 -- trace ("con: "{-++unpackFS lexeme-}) $
976 let empty_buf = stepOverLexeme buf in
977 case expandWhile# is_ident empty_buf of { buf1 ->
978 case slurp_trailing_hashes buf1 glaexts of { con_buf ->
980 let all_buf = mergeLexemes buf con_buf
982 con_lexeme = lexemeToFastString con_buf
983 mod_lexeme = lexemeToFastString (decLexeme buf)
984 all_lexeme = lexemeToFastString all_buf
987 | emptyLexeme buf = cont (ITconid con_lexeme) all_buf
988 | otherwise = cont (ITqconid (mod_lexeme,con_lexeme)) all_buf
991 case currentChar# all_buf of
992 '.'# -> maybe_qualified cont glaexts all_lexeme
993 (incLexeme all_buf) just_a_conid
998 maybe_qualified cont glaexts mod buf just_a_conid =
999 -- trace ("qid: "{-++unpackFS lexeme-}) $
1000 case currentChar# buf of
1001 '['# -> -- Special case for []
1002 case lookAhead# buf 1# of
1003 ']'# -> cont (ITqconid (mod,SLIT("[]"))) (setCurrentPos# buf 2#)
1006 '('# -> -- Special case for (,,,)
1007 -- This *is* necessary to deal with e.g. "instance C PrelBase.(,,)"
1008 case lookAhead# buf 1# of
1009 '#'# | flag glaexts -> case lookAhead# buf 2# of
1010 ','# -> lex_ubx_tuple cont mod (setCurrentPos# buf 3#)
1013 ')'# -> cont (ITqconid (mod,SLIT("()"))) (setCurrentPos# buf 2#)
1014 ','# -> lex_tuple cont mod (setCurrentPos# buf 2#) just_a_conid
1017 '-'# -> case lookAhead# buf 1# of
1018 '>'# -> cont (ITqconid (mod,SLIT("(->)"))) (setCurrentPos# buf 2#)
1019 _ -> lex_id3 cont glaexts mod buf just_a_conid
1021 _ -> lex_id3 cont glaexts mod buf just_a_conid
1024 lex_id3 cont glaexts mod buf just_a_conid
1025 | is_upper (currentChar# buf) =
1026 lex_con cont glaexts buf
1028 | is_symbol (currentChar# buf) =
1030 start_new_lexeme = stepOverLexeme buf
1032 -- trace ("lex_id31 "{-++unpackFS lexeme-}) $
1033 case expandWhile# is_symbol start_new_lexeme of { buf' ->
1035 lexeme = lexemeToFastString buf'
1036 -- real lexeme is M.<sym>
1037 new_buf = mergeLexemes buf buf'
1039 cont (mk_qvar_token mod lexeme) new_buf
1040 -- wrong, but arguably morally right: M... is now a qvarsym
1045 start_new_lexeme = stepOverLexeme buf
1047 -- trace ("lex_id32 "{-++unpackFS lexeme-}) $
1048 case expandWhile# is_ident start_new_lexeme of { buf1 ->
1053 case slurp_trailing_hashes buf1 glaexts of { buf' ->
1056 lexeme = lexemeToFastString buf'
1057 new_buf = mergeLexemes buf buf'
1058 is_a_qvarid = cont (mk_qvar_token mod lexeme) new_buf
1060 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
1061 Nothing -> is_a_qvarid ;
1063 Just kwd_token | isSpecial kwd_token -- special ids (as, qualified, hiding) shouldn't be
1064 -> is_a_qvarid -- recognised as keywords here.
1066 -> just_a_conid -- avoid M.where etc.
1069 slurp_trailing_hashes buf glaexts
1070 | flag glaexts = expandWhile# (`eqChar#` '#'#) buf
1075 | is_upper f = ITconid pk_str
1076 | is_ident f = ITvarid pk_str
1077 | f `eqChar#` ':'# = ITconsym pk_str
1078 | otherwise = ITvarsym pk_str
1080 (C# f) = _HEAD_ pk_str
1081 -- tl = _TAIL_ pk_str
1083 mk_qvar_token m token =
1084 -- trace ("mk_qvar ") $
1085 case mk_var_token token of
1086 ITconid n -> ITqconid (m,n)
1087 ITvarid n -> ITqvarid (m,n)
1088 ITconsym n -> ITqconsym (m,n)
1089 ITvarsym n -> ITqvarsym (m,n)
1090 _ -> ITunknown (show token)
1093 ----------------------------------------------------------------------------
1094 Horrible stuff for dealing with M.(,,,)
1097 lex_tuple cont mod buf back_off =
1101 case currentChar# buf of
1102 ','# -> go (n+1) (stepOn buf)
1103 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Boxed n))) (stepOn buf)
1106 lex_ubx_tuple cont mod buf back_off =
1110 case currentChar# buf of
1111 ','# -> go (n+1) (stepOn buf)
1112 '#'# -> case lookAhead# buf 1# of
1113 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Unboxed n)))
1119 -----------------------------------------------------------------------------
1120 doDiscard rips along really fast, looking for a '##-}',
1121 indicating the end of the pragma we're skipping
1124 doDiscard inStr buf =
1125 case currentChar# buf of
1126 '#'# | inStr ==# 0# ->
1127 case lookAhead# buf 1# of { '#'# ->
1128 case lookAhead# buf 2# of { '-'# ->
1129 case lookAhead# buf 3# of { '}'# ->
1130 (lexemeToBuffer buf, stepOverLexeme (setCurrentPos# buf 4#));
1131 _ -> doDiscard inStr (incLexeme buf) };
1132 _ -> doDiscard inStr (incLexeme buf) };
1133 _ -> doDiscard inStr (incLexeme buf) }
1137 odd_slashes buf flg i# =
1138 case lookAhead# buf i# of
1139 '\\'# -> odd_slashes buf (not flg) (i# -# 1#)
1142 not_inStr = if inStr ==# 0# then 1# else 0#
1144 case lookAhead# buf (negateInt# 1#) of --backwards, actually
1145 '\\'# -> -- escaping something..
1146 if odd_slashes buf True (negateInt# 2#)
1147 then -- odd number of slashes, " is escaped.
1148 doDiscard inStr (incLexeme buf)
1149 else -- even number of slashes, \ is escaped.
1150 doDiscard not_inStr (incLexeme buf)
1151 _ -> doDiscard not_inStr (incLexeme buf)
1153 '\''# | inStr ==# 0# ->
1154 case lookAhead# buf 1# of { '"'# ->
1155 case lookAhead# buf 2# of { '\''# ->
1156 doDiscard inStr (setCurrentPos# buf 3#);
1157 _ -> doDiscard inStr (incLexeme buf) };
1158 _ -> doDiscard inStr (incLexeme buf) }
1160 _ -> doDiscard inStr (incLexeme buf)
1164 -----------------------------------------------------------------------------
1175 data PState = PState {
1177 glasgow_exts :: Int#,
1180 context :: [LayoutContext]
1183 type P a = StringBuffer -- Input string
1188 returnP a buf s = POk s a
1190 thenP :: P a -> (a -> P b) -> P b
1191 m `thenP` k = \ buf s ->
1193 POk s1 a -> k a buf s1
1194 PFailed err -> PFailed err
1196 thenP_ :: P a -> P b -> P b
1197 m `thenP_` k = m `thenP` \_ -> k
1199 mapP :: (a -> P b) -> [a] -> P [b]
1200 mapP f [] = returnP []
1203 mapP f as `thenP` \bs ->
1206 failP :: String -> P a
1207 failP msg buf s = PFailed (text msg)
1209 failMsgP :: Message -> P a
1210 failMsgP msg buf s = PFailed msg
1212 lexError :: String -> P a
1213 lexError str buf s@PState{ loc = loc }
1214 = failMsgP (hcat [ppr loc, text ": ", text str]) buf s
1216 getSrcLocP :: P SrcLoc
1217 getSrcLocP buf s@(PState{ loc = loc }) = POk s loc
1219 -- use a temporary SrcLoc for the duration of the argument
1220 setSrcLocP :: SrcLoc -> P a -> P a
1221 setSrcLocP new_loc p buf s =
1222 case p buf s{ loc=new_loc } of
1224 PFailed e -> PFailed e
1226 getSrcFile :: P FAST_STRING
1227 getSrcFile buf s@(PState{ loc = loc }) = POk s (srcLocFile loc)
1229 pushContext :: LayoutContext -> P ()
1230 pushContext ctxt buf s@(PState{ context = ctx }) = POk s{context = ctxt:ctx} ()
1234 This special case in layoutOn is to handle layout contexts with are
1235 indented the same or less than the current context. This is illegal
1236 according to the Haskell spec, so we have to arrange to close the
1237 current context. eg.
1242 after the first 'where', the sequence of events is:
1244 - layout system inserts a ';' (column 0)
1245 - parser begins a new context at column 0
1246 - parser shifts ';' (legal empty declaration)
1247 - parser sees 'class': parse error (we're still in the inner context)
1249 trouble is, by the time we know we need a new context, the lexer has
1250 already generated the ';'. Hacky solution is as follows: since we
1251 know the column of the next token (it's the column number of the new
1252 context), we set the ACTUAL column number of the new context to this
1253 numer plus one. Hence the next time the lexer is called, a '}' will
1254 be generated to close the new context straight away. Furthermore, we
1255 have to set the atbol flag so that the ';' that the parser shifted as
1256 part of the new context is re-generated.
1258 when the new context is *less* indented than the current one:
1260 f = f where g = g where
1263 - current context: column 12.
1264 - on seeing 'h' (column 0), the layout system inserts '}'
1265 - parser starts a new context, column 0
1266 - parser sees '}', uses it to close new context
1267 - we still need to insert another '}' followed by a ';',
1268 hence the atbol trick.
1270 There's also a special hack in here to deal with
1277 i.e. the inner context is at the same indentation level as the outer
1278 context. This is strictly illegal according to Haskell 98, but
1279 there's a lot of existing code using this style and it doesn't make
1280 any sense to disallow it, since empty 'do' lists don't make sense.
1283 layoutOn :: Bool -> P ()
1284 layoutOn strict buf s@(PState{ bol = bol, context = ctx }) =
1285 let offset = lexemeIndex buf -# bol in
1288 | if strict then prev_off >=# offset else prev_off ># offset ->
1289 --trace ("layout on, column: " ++ show (I# offset)) $
1290 POk s{ context = Layout (offset +# 1#) : ctx, atbol = 1# } ()
1292 --trace ("layout on, column: " ++ show (I# offset)) $
1293 POk s{ context = Layout offset : ctx } ()
1296 layoutOff buf s@(PState{ context = ctx }) =
1297 POk s{ context = NoLayout:ctx } ()
1300 popContext = \ buf s@(PState{ context = ctx, loc = loc }) ->
1302 (_:tl) -> POk s{ context = tl } ()
1303 [] -> PFailed (srcParseErr buf loc)
1306 Note that if the name of the file we're processing ends
1307 with `hi-boot', we accept it on faith as having the right
1308 version. This is done so that .hi-boot files that comes
1309 with hsc don't have to be updated before every release,
1310 *and* it allows us to share .hi-boot files with versions
1311 of hsc that don't have .hi version checking (e.g., ghc-2.10's)
1313 If the version number is 0, the checking is also turned off.
1314 (needed to deal with GHC.hi only!)
1316 Once we can assume we're compiling with a version of ghc that
1317 supports interface file checking, we can drop the special
1320 checkVersion :: Maybe Integer -> P ()
1321 checkVersion mb@(Just v) buf s@(PState{loc = loc})
1322 | (v==0) || (v == fromInt opt_HiVersion) || opt_NoHiCheck = POk s ()
1323 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1324 checkVersion mb@Nothing buf s@(PState{loc = loc})
1325 | "hi-boot" `isSuffixOf` (_UNPK_ (srcLocFile loc)) = POk s ()
1326 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1328 -----------------------------------------------------------------
1330 ifaceParseErr :: StringBuffer -> SrcLoc -> Message
1332 = hsep [ppr l, ptext SLIT("Interface file parse error; on input `"),
1333 text (lexemeToString s), char '\'']
1335 ifaceVersionErr hi_vers l toks
1336 = hsep [ppr l, ptext SLIT("Interface file version error;"),
1337 ptext SLIT("Expected"), int opt_HiVersion,
1338 ptext SLIT("found "), pp_version]
1342 Nothing -> ptext SLIT("pre ghc-3.02 version")
1343 Just v -> ptext SLIT("version") <+> integer v
1345 -----------------------------------------------------------------------------
1347 srcParseErr :: StringBuffer -> SrcLoc -> Message
1351 then ptext SLIT(": parse error (possibly incorrect indentation)")
1352 else hcat [ptext SLIT(": parse error on input "),
1353 char '`', text token, char '\'']
1356 token = lexemeToString s