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 = skipNestedComment (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 skipNestedComment :: P a -> P a
501 skipNestedComment cont buf state = skipNestedComment' (loc state) cont buf state
503 skipNestedComment' :: SrcLoc -> P a -> P a
504 skipNestedComment' orig_loc cont buf = loop buf
507 case currentChar# buf of
508 '-'# | lookAhead# buf 1# `eqChar#` '}'# -> cont (stepOnBy# buf 2#)
510 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
512 (skipNestedComment' orig_loc cont)
515 '\n'# -> \ s@PState{loc=loc} ->
516 let buf' = stepOn buf in
517 loop buf' s{loc = incSrcLine loc,
518 bol = currentIndex# buf',
521 -- pass the original SrcLoc to lexError so that the error is
522 -- reported at the line it was originally on, not the line at
523 -- the end of the file.
524 '\NUL'# | bufferExhausted (stepOn buf) ->
525 \s -> lexError "unterminated `{-'" buf s{loc=orig_loc} -- -}
527 _ -> loop (stepOn buf)
529 -- When we are lexing the first token of a line, check whether we need to
530 -- insert virtual semicolons or close braces due to layout.
532 lexBOL :: (Token -> P a) -> P a
533 lexBOL cont buf s@(PState{
535 glasgow_exts = glaexts,
540 if need_close_curly then
541 --trace ("col = " ++ show (I# col) ++ ", layout: inserting '}'") $
542 cont ITvccurly buf s{atbol = 1#, context = tail ctx}
543 else if need_semi_colon then
544 --trace ("col = " ++ show (I# col) ++ ", layout: inserting ';'") $
545 cont ITsemi buf s{atbol = 0#}
547 lexToken cont glaexts buf s{atbol = 0#}
549 col = currentIndex# buf -# bol
562 Layout n -> col ==# n
565 lexToken :: (Token -> P a) -> Int# -> P a
566 lexToken cont glaexts buf =
567 -- trace "lexToken" $
568 case currentChar# buf of
570 -- special symbols ----------------------------------------------------
571 '('# | flag glaexts && lookAhead# buf 1# `eqChar#` '#'#
572 -> cont IToubxparen (setCurrentPos# buf 2#)
574 -> cont IToparen (incLexeme buf)
576 ')'# -> cont ITcparen (incLexeme buf)
577 '['# -> cont ITobrack (incLexeme buf)
578 ']'# -> cont ITcbrack (incLexeme buf)
579 ','# -> cont ITcomma (incLexeme buf)
580 ';'# -> cont ITsemi (incLexeme buf)
581 '}'# -> \ s@PState{context = ctx} ->
583 (_:ctx') -> cont ITccurly (incLexeme buf) s{context=ctx'}
584 _ -> lexError "too many '}'s" buf s
585 '|'# -> case lookAhead# buf 1# of
586 '}'# | flag glaexts -> cont ITccurlybar
587 (setCurrentPos# buf 2#)
588 _ -> lex_sym cont (incLexeme buf)
591 '#'# -> case lookAhead# buf 1# of
592 ')'# | flag glaexts -> cont ITcubxparen (setCurrentPos# buf 2#)
593 '-'# -> case lookAhead# buf 2# of
594 '}'# -> cont ITclose_prag (setCurrentPos# buf 3#)
595 _ -> lex_sym cont (incLexeme buf)
596 _ -> lex_sym cont (incLexeme buf)
598 '`'# | flag glaexts && lookAhead# buf 1# `eqChar#` '`'#
599 -> lex_cstring cont (setCurrentPos# buf 2#)
601 -> cont ITbackquote (incLexeme buf)
603 '{'# -> -- look for "{-##" special iface pragma
604 case lookAhead# buf 1# of
606 -> cont ITocurlybar (setCurrentPos# buf 2#)
607 '-'# -> case lookAhead# buf 2# of
608 '#'# -> case lookAhead# buf 3# of
611 = doDiscard 0# (stepOnBy# (stepOverLexeme buf) 4#) in
612 cont (ITpragma lexeme) buf'
613 _ -> lex_prag cont (setCurrentPos# buf 3#)
614 _ -> cont ITocurly (incLexeme buf)
615 _ -> (layoutOff `thenP_` cont ITocurly) (incLexeme buf)
617 -- strings/characters -------------------------------------------------
618 '\"'#{-"-} -> lex_string cont glaexts [] (incLexeme buf)
619 '\''# -> lex_char (char_end cont) glaexts (incLexeme buf)
621 -- strictness and cpr pragmas and __scc treated specially.
622 '_'# | flag glaexts ->
623 case lookAhead# buf 1# of
624 '_'# -> case lookAhead# buf 2# of
626 lex_demand cont (stepOnUntil (not . isSpace)
627 (stepOnBy# buf 3#)) -- past __S
629 cont ITcprinfo (stepOnBy# buf 3#) -- past __M
632 case prefixMatch (stepOnBy# buf 3#) "cc" of
633 Just buf' -> lex_scc cont (stepOverLexeme buf')
634 Nothing -> lex_id cont glaexts buf
635 _ -> lex_id cont glaexts buf
636 _ -> lex_id cont glaexts buf
638 -- Hexadecimal and octal constants
639 '0'# | (ch `eqChar#` 'x'# || ch `eqChar#` 'X'#) && is_hexdigit ch2
640 -> readNum (after_lexnum cont glaexts) buf' is_hexdigit 16 hex
641 | (ch `eqChar#` 'o'# || ch `eqChar#` 'O'#) && is_octdigit ch2
642 -> readNum (after_lexnum cont glaexts) buf' is_octdigit 8 oct_or_dec
643 where ch = lookAhead# buf 1#
644 ch2 = lookAhead# buf 2#
645 buf' = setCurrentPos# buf 2#
648 if bufferExhausted (stepOn buf) then
651 trace "lexIface: misplaced NUL?" $
652 cont (ITunknown "\NUL") (stepOn buf)
654 '?'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
655 lex_ip cont (incLexeme buf)
656 c | is_digit c -> lex_num cont glaexts 0 buf
657 | is_symbol c -> lex_sym cont buf
658 | is_upper c -> lex_con cont glaexts buf
659 | is_ident c -> lex_id cont glaexts buf
660 | otherwise -> lexError "illegal character" buf
662 -- Int# is unlifted, and therefore faster than Bool for flags.
668 -------------------------------------------------------------------------------
672 = case expandWhile# is_space buf of { buf1 ->
673 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2 ->
674 let lexeme = mkFastString (map toUpper (lexemeToString buf2)) in
675 case lookupUFM pragmaKeywordsFM lexeme of
676 Just kw -> cont kw (mergeLexemes buf buf2)
677 Nothing -> panic "lex_prag"
680 -------------------------------------------------------------------------------
683 lex_string cont glaexts s buf
684 = case currentChar# buf of
686 let buf' = incLexeme buf
687 s' = mkFastStringNarrow (map chr (reverse s))
688 in case currentChar# buf' of
689 '#'# | flag glaexts -> if all (<= 0xFF) s
690 then cont (ITprimstring s') (incLexeme buf')
691 else lexError "primitive string literal must contain only characters <= \'\\xFF\'" buf'
692 _ -> cont (ITstring s') buf'
694 -- ignore \& in a string, deal with string gaps
695 '\\'# | next_ch `eqChar#` '&'#
696 -> lex_string cont glaexts s buf'
698 -> lex_stringgap cont glaexts s (incLexeme buf)
700 where next_ch = lookAhead# buf 1#
701 buf' = setCurrentPos# buf 2#
703 _ -> lex_char (lex_next_string cont s) glaexts buf
705 lex_stringgap cont glaexts s buf
706 = let buf' = incLexeme buf in
707 case currentChar# buf of
708 '\n'# -> \st@PState{loc = loc} -> lex_stringgap cont glaexts s buf'
709 st{loc = incSrcLine loc}
710 '\\'# -> lex_string cont glaexts s buf'
711 c | is_space c -> lex_stringgap cont glaexts s buf'
712 other -> charError buf'
714 lex_next_string cont s glaexts c buf = lex_string cont glaexts (c:s) buf
716 lex_char :: (Int# -> Int -> P a) -> Int# -> P a
717 lex_char cont glaexts buf
718 = case currentChar# buf of
719 '\\'# -> lex_escape (cont glaexts) (incLexeme buf)
720 c | is_any c -> cont glaexts (I# (ord# c)) (incLexeme buf)
721 other -> charError buf
723 char_end cont glaexts c buf
724 = case currentChar# buf of
725 '\''# -> let buf' = incLexeme buf in
726 case currentChar# buf' of
728 -> cont (ITprimchar c) (incLexeme buf')
729 _ -> cont (ITchar c) buf'
733 = let buf' = incLexeme buf in
734 case currentChar# buf of
735 'a'# -> cont (ord '\a') buf'
736 'b'# -> cont (ord '\b') buf'
737 'f'# -> cont (ord '\f') buf'
738 'n'# -> cont (ord '\n') buf'
739 'r'# -> cont (ord '\r') buf'
740 't'# -> cont (ord '\t') buf'
741 'v'# -> cont (ord '\v') buf'
742 '\\'# -> cont (ord '\\') buf'
743 '"'# -> cont (ord '\"') buf'
744 '\''# -> cont (ord '\'') buf'
745 '^'# -> let c = currentChar# buf' in
746 if c `geChar#` '@'# && c `leChar#` '_'#
747 then cont (I# (ord# c -# ord# '@'#)) (incLexeme buf')
750 'x'# -> readNum (after_charnum cont) buf' is_hexdigit 16 hex
751 'o'# -> readNum (after_charnum cont) buf' is_octdigit 8 oct_or_dec
753 -> readNum (after_charnum cont) buf is_digit 10 oct_or_dec
755 _ -> case [ (c,buf2) | (p,c) <- silly_escape_chars,
756 Just buf2 <- [prefixMatch buf p] ] of
757 (c,buf2):_ -> cont (ord c) buf2
760 after_charnum cont i buf
761 = if i >= 0 && i <= 0x10FFFF
762 then cont (fromInteger i) buf
765 readNum cont buf is_digit base conv = read buf 0
767 = case currentChar# buf of { c ->
769 then read (incLexeme buf) (i*base + (toInteger (I# (conv c))))
775 || (c `geChar#` 'a'# && c `leChar#` 'f'#)
776 || (c `geChar#` 'A'# && c `leChar#` 'F'#)
778 hex c | is_digit c = ord# c -# ord# '0'#
779 | otherwise = ord# (to_lower c) -# ord# 'a'# +# 10#
780 oct_or_dec c = ord# c -# ord# '0'#
782 is_octdigit c = c `geChar#` '0'# && c `leChar#` '7'#
785 | c `geChar#` 'A'# && c `leChar#` 'Z'#
786 = chr# (ord# c -# (ord# 'A'# -# ord# 'a'#))
789 charError buf = lexError "error in character literal" buf
791 silly_escape_chars = [
828 -------------------------------------------------------------------------------
830 lex_demand cont buf =
831 case read_em [] buf of { (ls,buf') ->
832 case currentChar# buf' of
833 'b'# -> cont (ITstrict (StrictSig (mkTopDmdType ls BotRes))) (incLexeme buf')
834 'm'# -> cont (ITstrict (StrictSig (mkTopDmdType ls RetCPR))) (incLexeme buf')
835 _ -> cont (ITstrict (StrictSig (mkTopDmdType ls TopRes))) buf'
839 case currentChar# buf of
840 'L'# -> read_em (Lazy : acc) (stepOn buf)
841 'A'# -> read_em (Abs : acc) (stepOn buf)
842 'V'# -> read_em (Eval : acc) (stepOn buf)
843 'X'# -> read_em (Err : acc) (stepOn buf)
844 'B'# -> read_em (Bot : acc) (stepOn buf)
845 ')'# -> (reverse acc, stepOn buf)
846 'C'# -> do_call acc (stepOnBy# buf 2#)
847 'D'# -> do_unpack1 Defer acc (stepOnBy# buf 1#)
848 'U'# -> do_unpack1 Drop acc (stepOnBy# buf 1#)
849 'S'# -> do_unpack1 Keep acc (stepOnBy# buf 1#)
850 _ -> (reverse acc, buf)
852 do_unpack1 keepity acc buf
853 = case currentChar# buf of
854 '('# -> do_unpack2 keepity acc (stepOnBy# buf 1#)
855 _ -> read_em (Seq keepity [] : acc) buf
857 do_unpack2 keepity acc buf
858 = case read_em [] buf of
859 (stuff, rest) -> read_em (Seq keepity stuff : acc) rest
862 = case read_em [] buf of
863 ([dmd], rest) -> read_em (Call dmd : acc) rest
867 case currentChar# buf of
868 'C'# -> cont ITsccAllCafs (incLexeme buf)
869 other -> cont ITscc buf
871 -----------------------------------------------------------------------------
874 lex_num :: (Token -> P a) -> Int# -> Integer -> P a
875 lex_num cont glaexts acc buf =
876 case scanNumLit acc buf of
878 case currentChar# buf' of
879 '.'# | is_digit (lookAhead# buf' 1#) ->
880 -- this case is not optimised at all, as the
881 -- presence of floating point numbers in interface
882 -- files is not that common. (ToDo)
883 case expandWhile# is_digit (incLexeme buf') of
884 buf2 -> -- points to first non digit char
886 let l = case currentChar# buf2 of
892 = let buf3 = incLexeme buf2 in
893 case currentChar# buf3 of
894 '-'# -> expandWhile# is_digit (incLexeme buf3)
895 '+'# -> expandWhile# is_digit (incLexeme buf3)
896 x | is_digit x -> expandWhile# is_digit buf3
899 v = readRational__ (lexemeToString l)
901 in case currentChar# l of -- glasgow exts only
902 '#'# | flag glaexts -> let l' = incLexeme l in
903 case currentChar# l' of
904 '#'# -> cont (ITprimdouble v) (incLexeme l')
905 _ -> cont (ITprimfloat v) l'
906 _ -> cont (ITrational v) l
908 _ -> after_lexnum cont glaexts acc' buf'
910 after_lexnum cont glaexts i buf
911 = case currentChar# buf of
912 '#'# | flag glaexts -> cont (ITprimint i) (incLexeme buf)
913 _ -> cont (ITinteger i) buf
915 -----------------------------------------------------------------------------
916 -- C "literal literal"s (i.e. things like ``NULL'', ``stdout'' etc.)
918 -- we lexemeToFastString on the bit between the ``''s, but include the
919 -- quotes in the full lexeme.
921 lex_cstring cont buf =
922 case expandUntilMatch (stepOverLexeme buf) "\'\'" of
923 Just buf' -> cont (ITlitlit (lexemeToFastString
924 (setCurrentPos# buf' (negateInt# 2#))))
925 (mergeLexemes buf buf')
926 Nothing -> lexError "unterminated ``" buf
928 -----------------------------------------------------------------------------
929 -- identifiers, symbols etc.
932 case expandWhile# is_ident buf of
933 buf' -> cont (ITipvarid lexeme) buf'
934 where lexeme = lexemeToFastString buf'
936 lex_id cont glaexts buf =
937 let buf1 = expandWhile# is_ident buf in
940 case (if flag glaexts
941 then expandWhile# (eqChar# '#'#) buf1 -- slurp trailing hashes
942 else buf1) of { buf' ->
944 let lexeme = lexemeToFastString buf' in
946 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
947 Just kwd_token -> --trace ("hkeywd: "++_UNPK_(lexeme)) $
951 let var_token = cont (ITvarid lexeme) buf' in
953 if not (flag glaexts)
957 case lookupUFM ghcExtensionKeywordsFM lexeme of {
958 Just kwd_token -> cont kwd_token buf';
965 case expandWhile# is_symbol buf of
966 buf' -> case lookupUFM haskellKeySymsFM lexeme of {
967 Just kwd_token -> --trace ("keysym: "++unpackFS lexeme) $
968 cont kwd_token buf' ;
969 Nothing -> --trace ("sym: "++unpackFS lexeme) $
970 cont (mk_var_token lexeme) buf'
972 where lexeme = lexemeToFastString buf'
975 -- lex_con recursively collects components of a qualified identifer.
976 -- The argument buf is the StringBuffer representing the lexeme
977 -- identified so far, where the next character is upper-case.
979 lex_con cont glaexts buf =
980 -- trace ("con: "{-++unpackFS lexeme-}) $
981 let empty_buf = stepOverLexeme buf in
982 case expandWhile# is_ident empty_buf of { buf1 ->
983 case slurp_trailing_hashes buf1 glaexts of { con_buf ->
985 let all_buf = mergeLexemes buf con_buf
987 con_lexeme = lexemeToFastString con_buf
988 mod_lexeme = lexemeToFastString (decLexeme buf)
989 all_lexeme = lexemeToFastString all_buf
992 | emptyLexeme buf = cont (ITconid con_lexeme) all_buf
993 | otherwise = cont (ITqconid (mod_lexeme,con_lexeme)) all_buf
996 case currentChar# all_buf of
997 '.'# -> maybe_qualified cont glaexts all_lexeme
998 (incLexeme all_buf) just_a_conid
1003 maybe_qualified cont glaexts mod buf just_a_conid =
1004 -- trace ("qid: "{-++unpackFS lexeme-}) $
1005 case currentChar# buf of
1006 '['# -> -- Special case for []
1007 case lookAhead# buf 1# of
1008 ']'# -> cont (ITqconid (mod,SLIT("[]"))) (setCurrentPos# buf 2#)
1011 '('# -> -- Special case for (,,,)
1012 -- This *is* necessary to deal with e.g. "instance C PrelBase.(,,)"
1013 case lookAhead# buf 1# of
1014 '#'# | flag glaexts -> case lookAhead# buf 2# of
1015 ','# -> lex_ubx_tuple cont mod (setCurrentPos# buf 3#)
1018 ')'# -> cont (ITqconid (mod,SLIT("()"))) (setCurrentPos# buf 2#)
1019 ','# -> lex_tuple cont mod (setCurrentPos# buf 2#) just_a_conid
1022 '-'# -> case lookAhead# buf 1# of
1023 '>'# -> cont (ITqconid (mod,SLIT("(->)"))) (setCurrentPos# buf 2#)
1024 _ -> lex_id3 cont glaexts mod buf just_a_conid
1026 _ -> lex_id3 cont glaexts mod buf just_a_conid
1029 lex_id3 cont glaexts mod buf just_a_conid
1030 | is_upper (currentChar# buf) =
1031 lex_con cont glaexts buf
1033 | is_symbol (currentChar# buf) =
1035 start_new_lexeme = stepOverLexeme buf
1037 -- trace ("lex_id31 "{-++unpackFS lexeme-}) $
1038 case expandWhile# is_symbol start_new_lexeme of { buf' ->
1040 lexeme = lexemeToFastString buf'
1041 -- real lexeme is M.<sym>
1042 new_buf = mergeLexemes buf buf'
1044 cont (mk_qvar_token mod lexeme) new_buf
1045 -- wrong, but arguably morally right: M... is now a qvarsym
1050 start_new_lexeme = stepOverLexeme buf
1052 -- trace ("lex_id32 "{-++unpackFS lexeme-}) $
1053 case expandWhile# is_ident start_new_lexeme of { buf1 ->
1058 case slurp_trailing_hashes buf1 glaexts of { buf' ->
1061 lexeme = lexemeToFastString buf'
1062 new_buf = mergeLexemes buf buf'
1063 is_a_qvarid = cont (mk_qvar_token mod lexeme) new_buf
1065 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
1066 Nothing -> is_a_qvarid ;
1068 Just kwd_token | isSpecial kwd_token -- special ids (as, qualified, hiding) shouldn't be
1069 -> is_a_qvarid -- recognised as keywords here.
1071 -> just_a_conid -- avoid M.where etc.
1074 slurp_trailing_hashes buf glaexts
1075 | flag glaexts = expandWhile# (`eqChar#` '#'#) buf
1080 | is_upper f = ITconid pk_str
1081 | is_ident f = ITvarid pk_str
1082 | f `eqChar#` ':'# = ITconsym pk_str
1083 | otherwise = ITvarsym pk_str
1085 (C# f) = _HEAD_ pk_str
1086 -- tl = _TAIL_ pk_str
1088 mk_qvar_token m token =
1089 -- trace ("mk_qvar ") $
1090 case mk_var_token token of
1091 ITconid n -> ITqconid (m,n)
1092 ITvarid n -> ITqvarid (m,n)
1093 ITconsym n -> ITqconsym (m,n)
1094 ITvarsym n -> ITqvarsym (m,n)
1095 _ -> ITunknown (show token)
1098 ----------------------------------------------------------------------------
1099 Horrible stuff for dealing with M.(,,,)
1102 lex_tuple cont mod buf back_off =
1106 case currentChar# buf of
1107 ','# -> go (n+1) (stepOn buf)
1108 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Boxed n))) (stepOn buf)
1111 lex_ubx_tuple cont mod buf back_off =
1115 case currentChar# buf of
1116 ','# -> go (n+1) (stepOn buf)
1117 '#'# -> case lookAhead# buf 1# of
1118 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Unboxed n)))
1124 -----------------------------------------------------------------------------
1125 doDiscard rips along really fast, looking for a '##-}',
1126 indicating the end of the pragma we're skipping
1129 doDiscard inStr buf =
1130 case currentChar# buf of
1131 '#'# | inStr ==# 0# ->
1132 case lookAhead# buf 1# of { '#'# ->
1133 case lookAhead# buf 2# of { '-'# ->
1134 case lookAhead# buf 3# of { '}'# ->
1135 (lexemeToBuffer buf, stepOverLexeme (setCurrentPos# buf 4#));
1136 _ -> doDiscard inStr (incLexeme buf) };
1137 _ -> doDiscard inStr (incLexeme buf) };
1138 _ -> doDiscard inStr (incLexeme buf) }
1142 odd_slashes buf flg i# =
1143 case lookAhead# buf i# of
1144 '\\'# -> odd_slashes buf (not flg) (i# -# 1#)
1147 not_inStr = if inStr ==# 0# then 1# else 0#
1149 case lookAhead# buf (negateInt# 1#) of --backwards, actually
1150 '\\'# -> -- escaping something..
1151 if odd_slashes buf True (negateInt# 2#)
1152 then -- odd number of slashes, " is escaped.
1153 doDiscard inStr (incLexeme buf)
1154 else -- even number of slashes, \ is escaped.
1155 doDiscard not_inStr (incLexeme buf)
1156 _ -> doDiscard not_inStr (incLexeme buf)
1158 '\''# | inStr ==# 0# ->
1159 case lookAhead# buf 1# of { '"'# ->
1160 case lookAhead# buf 2# of { '\''# ->
1161 doDiscard inStr (setCurrentPos# buf 3#);
1162 _ -> doDiscard inStr (incLexeme buf) };
1163 _ -> doDiscard inStr (incLexeme buf) }
1165 _ -> doDiscard inStr (incLexeme buf)
1169 -----------------------------------------------------------------------------
1180 data PState = PState {
1182 glasgow_exts :: Int#,
1185 context :: [LayoutContext]
1188 type P a = StringBuffer -- Input string
1193 returnP a buf s = POk s a
1195 thenP :: P a -> (a -> P b) -> P b
1196 m `thenP` k = \ buf s ->
1198 POk s1 a -> k a buf s1
1199 PFailed err -> PFailed err
1201 thenP_ :: P a -> P b -> P b
1202 m `thenP_` k = m `thenP` \_ -> k
1204 mapP :: (a -> P b) -> [a] -> P [b]
1205 mapP f [] = returnP []
1208 mapP f as `thenP` \bs ->
1211 failP :: String -> P a
1212 failP msg buf s = PFailed (text msg)
1214 failMsgP :: Message -> P a
1215 failMsgP msg buf s = PFailed msg
1217 lexError :: String -> P a
1218 lexError str buf s@PState{ loc = loc }
1219 = failMsgP (hcat [ppr loc, text ": ", text str]) buf s
1221 getSrcLocP :: P SrcLoc
1222 getSrcLocP buf s@(PState{ loc = loc }) = POk s loc
1224 -- use a temporary SrcLoc for the duration of the argument
1225 setSrcLocP :: SrcLoc -> P a -> P a
1226 setSrcLocP new_loc p buf s =
1227 case p buf s{ loc=new_loc } of
1229 PFailed e -> PFailed e
1231 getSrcFile :: P FAST_STRING
1232 getSrcFile buf s@(PState{ loc = loc }) = POk s (srcLocFile loc)
1234 pushContext :: LayoutContext -> P ()
1235 pushContext ctxt buf s@(PState{ context = ctx }) = POk s{context = ctxt:ctx} ()
1239 This special case in layoutOn is to handle layout contexts with are
1240 indented the same or less than the current context. This is illegal
1241 according to the Haskell spec, so we have to arrange to close the
1242 current context. eg.
1247 after the first 'where', the sequence of events is:
1249 - layout system inserts a ';' (column 0)
1250 - parser begins a new context at column 0
1251 - parser shifts ';' (legal empty declaration)
1252 - parser sees 'class': parse error (we're still in the inner context)
1254 trouble is, by the time we know we need a new context, the lexer has
1255 already generated the ';'. Hacky solution is as follows: since we
1256 know the column of the next token (it's the column number of the new
1257 context), we set the ACTUAL column number of the new context to this
1258 numer plus one. Hence the next time the lexer is called, a '}' will
1259 be generated to close the new context straight away. Furthermore, we
1260 have to set the atbol flag so that the ';' that the parser shifted as
1261 part of the new context is re-generated.
1263 when the new context is *less* indented than the current one:
1265 f = f where g = g where
1268 - current context: column 12.
1269 - on seeing 'h' (column 0), the layout system inserts '}'
1270 - parser starts a new context, column 0
1271 - parser sees '}', uses it to close new context
1272 - we still need to insert another '}' followed by a ';',
1273 hence the atbol trick.
1275 There's also a special hack in here to deal with
1282 i.e. the inner context is at the same indentation level as the outer
1283 context. This is strictly illegal according to Haskell 98, but
1284 there's a lot of existing code using this style and it doesn't make
1285 any sense to disallow it, since empty 'do' lists don't make sense.
1288 layoutOn :: Bool -> P ()
1289 layoutOn strict buf s@(PState{ bol = bol, context = ctx }) =
1290 let offset = lexemeIndex buf -# bol in
1293 | if strict then prev_off >=# offset else prev_off ># offset ->
1294 --trace ("layout on, column: " ++ show (I# offset)) $
1295 POk s{ context = Layout (offset +# 1#) : ctx, atbol = 1# } ()
1297 --trace ("layout on, column: " ++ show (I# offset)) $
1298 POk s{ context = Layout offset : ctx } ()
1301 layoutOff buf s@(PState{ context = ctx }) =
1302 POk s{ context = NoLayout:ctx } ()
1305 popContext = \ buf s@(PState{ context = ctx, loc = loc }) ->
1307 (_:tl) -> POk s{ context = tl } ()
1308 [] -> PFailed (srcParseErr buf loc)
1311 Note that if the name of the file we're processing ends
1312 with `hi-boot', we accept it on faith as having the right
1313 version. This is done so that .hi-boot files that comes
1314 with hsc don't have to be updated before every release,
1315 *and* it allows us to share .hi-boot files with versions
1316 of hsc that don't have .hi version checking (e.g., ghc-2.10's)
1318 If the version number is 0, the checking is also turned off.
1319 (needed to deal with GHC.hi only!)
1321 Once we can assume we're compiling with a version of ghc that
1322 supports interface file checking, we can drop the special
1325 checkVersion :: Maybe Integer -> P ()
1326 checkVersion mb@(Just v) buf s@(PState{loc = loc})
1327 | (v==0) || (v == fromInt opt_HiVersion) || opt_NoHiCheck = POk s ()
1328 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1329 checkVersion mb@Nothing buf s@(PState{loc = loc})
1330 | "hi-boot" `isSuffixOf` (_UNPK_ (srcLocFile loc)) = POk s ()
1331 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1333 -----------------------------------------------------------------
1335 ifaceParseErr :: StringBuffer -> SrcLoc -> Message
1337 = hsep [ppr l, ptext SLIT("Interface file parse error; on input `"),
1338 text (lexemeToString s), char '\'']
1340 ifaceVersionErr hi_vers l toks
1341 = hsep [ppr l, ptext SLIT("Interface file version error;"),
1342 ptext SLIT("Expected"), int opt_HiVersion,
1343 ptext SLIT("found "), pp_version]
1347 Nothing -> ptext SLIT("pre ghc-3.02 version")
1348 Just v -> ptext SLIT("version") <+> integer v
1350 -----------------------------------------------------------------------------
1352 srcParseErr :: StringBuffer -> SrcLoc -> Message
1356 then ptext SLIT(": parse error (possibly incorrect indentation)")
1357 else hcat [ptext SLIT(": parse error on input "),
1358 char '`', text token, char '\'']
1361 token = lexemeToString s