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 PrelNames ( mkTupNameStr )
40 import CmdLineOpts ( opt_HiVersion, opt_NoHiCheck )
41 import ForeignCall ( Safety(..) )
42 import NewDemand ( StrictSig(..), Demand(..), Demands(..),
43 DmdResult(..), mkTopDmdType, evalDmd, lazyDmd )
44 import UniqFM ( listToUFM, lookupUFM )
45 import BasicTypes ( Boxity(..) )
46 import SrcLoc ( SrcLoc, incSrcLine, srcLocFile, srcLocLine,
47 replaceSrcLine, mkSrcLoc )
49 import ErrUtils ( Message )
56 import Char ( chr, ord )
57 import PrelRead ( readRational__ ) -- Glasgow non-std
60 %************************************************************************
62 \subsection{Data types}
64 %************************************************************************
66 The token data type, fairly un-interesting except from one
67 constructor, @ITidinfo@, which is used to lazily lex id info (arity,
68 strictness, unfolding etc).
70 The Idea/Observation here is that the renamer needs to scan through
71 all of an interface file before it can continue. But only a fraction
72 of the information contained in the file turns out to be useful, so
73 delaying as much as possible of the scanning and parsing of an
74 interface file Makes Sense (Heap profiles of the compiler
75 show a reduction in heap usage by at least a factor of two,
78 Hence, the interface file lexer spots when value declarations are
79 being scanned and return the @ITidinfo@ and @ITtype@ constructors
80 for the type and any other id info for that binding (unfolding, strictness
81 etc). These constructors are applied to the result of lexing these sub-chunks.
83 The lexing of the type and id info is all done lazily, of course, so
84 the scanning (and subsequent parsing) will be done *only* on the ids the
85 renamer finds out that it is interested in. The rest will just be junked.
86 Laziness, you know it makes sense :-)
90 = ITas -- Haskell keywords
114 | ITscc -- ToDo: remove (we use {-# SCC "..." #-} now)
116 | ITforall -- GHC extension keywords
127 | ITinterface -- interface keywords
135 | ITccall (Bool,Bool,Safety) -- (is_dyn, is_casm, may_gc)
162 | ITspecialise_prag -- Pragmas
172 | ITdotdot -- reserved symbols
186 | ITbiglam -- GHC-extension symbols
188 | ITocurly -- special symbols
190 | ITocurlybar -- {|, for type applications
191 | ITccurlybar -- |}, for type applications
204 | ITvarid FAST_STRING -- identifiers
205 | ITconid FAST_STRING
206 | ITvarsym FAST_STRING
207 | ITconsym FAST_STRING
208 | ITqvarid (FAST_STRING,FAST_STRING)
209 | ITqconid (FAST_STRING,FAST_STRING)
210 | ITqvarsym (FAST_STRING,FAST_STRING)
211 | ITqconsym (FAST_STRING,FAST_STRING)
213 | ITdupipvarid FAST_STRING -- GHC extension: implicit param: ?x
214 | ITsplitipvarid 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),
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 let buf1 | lookAhead# buf 1# `eqChar#` 'l'# &&
459 lookAhead# buf 2# `eqChar#` 'i'# &&
460 lookAhead# buf 3# `eqChar#` 'n'# &&
461 lookAhead# buf 4# `eqChar#` 'e'# = stepOnBy# buf 5#
462 | otherwise = stepOn buf
464 case expandWhile# is_space buf1 of { buf2 ->
465 if is_digit (currentChar# buf2)
466 then line_prag next_line buf2 s'
470 next_line buf = lexer cont (stepOnUntilChar# buf '\n'#)
472 -- tabs have been expanded beforehand
473 c | is_space c -> tab y bol atbol (stepOn buf)
474 | otherwise -> is_a_token
476 where s' = s{loc = replaceSrcLine loc y,
480 is_a_token | atbol /=# 0# = lexBOL cont buf s'
481 | otherwise = lexToken cont glaexts buf s'
483 -- {-# LINE .. #-} pragmas. yeuch.
484 line_prag cont buf s@PState{loc=loc} =
485 case expandWhile# is_space buf of { buf1 ->
486 case scanNumLit 0 (stepOverLexeme buf1) of { (line,buf2) ->
487 -- subtract one: the line number refers to the *following* line.
488 let real_line = line - 1 in
489 case fromInteger real_line of { i@(I# l) ->
490 -- ToDo, if no filename then we skip the newline.... d'oh
491 case expandWhile# is_space buf2 of { buf3 ->
492 case currentChar# buf3 of
494 case untilEndOfString# (stepOn (stepOverLexeme buf3)) of { buf4 ->
496 file = lexemeToFastString buf4
497 new_buf = stepOn (stepOverLexeme buf4)
499 if nullFastString file
500 then cont new_buf s{loc = replaceSrcLine loc l}
501 else cont new_buf s{loc = mkSrcLoc file i}
503 _other -> cont (stepOverLexeme buf3) s{loc = replaceSrcLine loc l}
506 skipNestedComment :: P a -> P a
507 skipNestedComment cont buf state = skipNestedComment' (loc state) cont buf state
509 skipNestedComment' :: SrcLoc -> P a -> P a
510 skipNestedComment' orig_loc cont buf = loop buf
513 case currentChar# buf of
514 '-'# | lookAhead# buf 1# `eqChar#` '}'# -> cont (stepOnBy# buf 2#)
516 '{'# | lookAhead# buf 1# `eqChar#` '-'# ->
518 (skipNestedComment' orig_loc cont)
521 '\n'# -> \ s@PState{loc=loc} ->
522 let buf' = stepOn buf in
523 loop buf' s{loc = incSrcLine loc,
524 bol = currentIndex# buf',
527 -- pass the original SrcLoc to lexError so that the error is
528 -- reported at the line it was originally on, not the line at
529 -- the end of the file.
530 '\NUL'# | bufferExhausted (stepOn buf) ->
531 \s -> lexError "unterminated `{-'" buf s{loc=orig_loc} -- -}
533 _ -> loop (stepOn buf)
535 -- When we are lexing the first token of a line, check whether we need to
536 -- insert virtual semicolons or close braces due to layout.
538 lexBOL :: (Token -> P a) -> P a
539 lexBOL cont buf s@(PState{
541 glasgow_exts = glaexts,
546 if need_close_curly then
547 --trace ("col = " ++ show (I# col) ++ ", layout: inserting '}'") $
548 cont ITvccurly buf s{atbol = 1#, context = tail ctx}
549 else if need_semi_colon then
550 --trace ("col = " ++ show (I# col) ++ ", layout: inserting ';'") $
551 cont ITsemi buf s{atbol = 0#}
553 lexToken cont glaexts buf s{atbol = 0#}
555 col = currentIndex# buf -# bol
568 Layout n -> col ==# n
571 lexToken :: (Token -> P a) -> Int# -> P a
572 lexToken cont glaexts buf =
573 -- trace "lexToken" $
574 case currentChar# buf of
576 -- special symbols ----------------------------------------------------
577 '('# | flag glaexts && lookAhead# buf 1# `eqChar#` '#'#
578 -> cont IToubxparen (setCurrentPos# buf 2#)
580 -> cont IToparen (incLexeme buf)
582 ')'# -> cont ITcparen (incLexeme buf)
583 '['# -> cont ITobrack (incLexeme buf)
584 ']'# -> cont ITcbrack (incLexeme buf)
585 ','# -> cont ITcomma (incLexeme buf)
586 ';'# -> cont ITsemi (incLexeme buf)
587 '}'# -> \ s@PState{context = ctx} ->
589 (_:ctx') -> cont ITccurly (incLexeme buf) s{context=ctx'}
590 _ -> lexError "too many '}'s" buf s
591 '|'# -> case lookAhead# buf 1# of
592 '}'# | flag glaexts -> cont ITccurlybar
593 (setCurrentPos# buf 2#)
594 _ -> lex_sym cont (incLexeme buf)
597 '#'# -> case lookAhead# buf 1# of
598 ')'# | flag glaexts -> cont ITcubxparen (setCurrentPos# buf 2#)
599 '-'# -> case lookAhead# buf 2# of
600 '}'# -> cont ITclose_prag (setCurrentPos# buf 3#)
601 _ -> lex_sym cont (incLexeme buf)
602 _ -> lex_sym cont (incLexeme buf)
604 '`'# | flag glaexts && lookAhead# buf 1# `eqChar#` '`'#
605 -> lex_cstring cont (setCurrentPos# buf 2#)
607 -> cont ITbackquote (incLexeme buf)
609 '{'# -> -- look for "{-##" special iface pragma
610 case lookAhead# buf 1# of
612 -> cont ITocurlybar (setCurrentPos# buf 2#)
613 '-'# -> case lookAhead# buf 2# of
614 '#'# -> case lookAhead# buf 3# of
618 (\ cont lexeme buf' -> cont (ITpragma lexeme) buf')
620 (stepOnBy# (stepOverLexeme buf) 4#)
621 _ -> lex_prag cont (setCurrentPos# buf 3#)
622 _ -> cont ITocurly (incLexeme buf)
623 _ -> (layoutOff `thenP_` cont ITocurly) (incLexeme buf)
625 -- strings/characters -------------------------------------------------
626 '\"'#{-"-} -> lex_string cont glaexts [] (incLexeme buf)
627 '\''# -> lex_char (char_end cont) glaexts (incLexeme buf)
629 -- strictness and cpr pragmas and __scc treated specially.
630 '_'# | flag glaexts ->
631 case lookAhead# buf 1# of
632 '_'# -> case lookAhead# buf 2# of
634 lex_demand cont (stepOnUntil (not . isSpace)
635 (stepOnBy# buf 3#)) -- past __S
637 cont ITcprinfo (stepOnBy# buf 3#) -- past __M
640 case prefixMatch (stepOnBy# buf 3#) "cc" of
641 Just buf' -> lex_scc cont (stepOverLexeme buf')
642 Nothing -> lex_id cont glaexts buf
643 _ -> lex_id cont glaexts buf
644 _ -> lex_id cont glaexts buf
646 -- Hexadecimal and octal constants
647 '0'# | (ch `eqChar#` 'x'# || ch `eqChar#` 'X'#) && is_hexdigit ch2
648 -> readNum (after_lexnum cont glaexts) buf' is_hexdigit 16 hex
649 | (ch `eqChar#` 'o'# || ch `eqChar#` 'O'#) && is_octdigit ch2
650 -> readNum (after_lexnum cont glaexts) buf' is_octdigit 8 oct_or_dec
651 where ch = lookAhead# buf 1#
652 ch2 = lookAhead# buf 2#
653 buf' = setCurrentPos# buf 2#
656 if bufferExhausted (stepOn buf) then
659 trace "lexIface: misplaced NUL?" $
660 cont (ITunknown "\NUL") (stepOn buf)
662 '?'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
663 lex_ip ITdupipvarid cont (incLexeme buf)
664 '%'# | flag glaexts && is_lower (lookAhead# buf 1#) ->
665 lex_ip ITsplitipvarid cont (incLexeme buf)
666 c | is_digit c -> lex_num cont glaexts 0 buf
667 | is_symbol c -> lex_sym cont buf
668 | is_upper c -> lex_con cont glaexts buf
669 | is_ident c -> lex_id cont glaexts buf
670 | otherwise -> lexError "illegal character" buf
672 -- Int# is unlifted, and therefore faster than Bool for flags.
678 -------------------------------------------------------------------------------
682 = case expandWhile# is_space buf of { buf1 ->
683 case expandWhile# is_ident (stepOverLexeme buf1) of { buf2 ->
684 let lexeme = mkFastString (map toUpper (lexemeToString buf2)) in
685 case lookupUFM pragmaKeywordsFM lexeme of
686 Just kw -> cont kw (mergeLexemes buf buf2)
687 Nothing -> panic "lex_prag"
690 -------------------------------------------------------------------------------
693 lex_string cont glaexts s buf
694 = case currentChar# buf of
696 let buf' = incLexeme buf
697 s' = mkFastStringNarrow (map chr (reverse s))
698 in case currentChar# buf' of
699 '#'# | flag glaexts -> if all (<= 0xFF) s
700 then cont (ITprimstring s') (incLexeme buf')
701 else lexError "primitive string literal must contain only characters <= \'\\xFF\'" buf'
702 _ -> cont (ITstring s') buf'
704 -- ignore \& in a string, deal with string gaps
705 '\\'# | next_ch `eqChar#` '&'#
706 -> lex_string cont glaexts s buf'
708 -> lex_stringgap cont glaexts s (incLexeme buf)
710 where next_ch = lookAhead# buf 1#
711 buf' = setCurrentPos# buf 2#
713 _ -> lex_char (lex_next_string cont s) glaexts buf
715 lex_stringgap cont glaexts s buf
716 = let buf' = incLexeme buf in
717 case currentChar# buf of
718 '\n'# -> \st@PState{loc = loc} -> lex_stringgap cont glaexts s buf'
719 st{loc = incSrcLine loc}
720 '\\'# -> lex_string cont glaexts s buf'
721 c | is_space c -> lex_stringgap cont glaexts s buf'
722 other -> charError buf'
724 lex_next_string cont s glaexts c buf = lex_string cont glaexts (c:s) buf
726 lex_char :: (Int# -> Int -> P a) -> Int# -> P a
727 lex_char cont glaexts buf
728 = case currentChar# buf of
729 '\\'# -> lex_escape (cont glaexts) (incLexeme buf)
730 c | is_any c -> cont glaexts (I# (ord# c)) (incLexeme buf)
731 other -> charError buf
733 char_end cont glaexts c buf
734 = case currentChar# buf of
735 '\''# -> let buf' = incLexeme buf in
736 case currentChar# buf' of
738 -> cont (ITprimchar c) (incLexeme buf')
739 _ -> cont (ITchar c) buf'
743 = let buf' = incLexeme buf in
744 case currentChar# buf of
745 'a'# -> cont (ord '\a') buf'
746 'b'# -> cont (ord '\b') buf'
747 'f'# -> cont (ord '\f') buf'
748 'n'# -> cont (ord '\n') buf'
749 'r'# -> cont (ord '\r') buf'
750 't'# -> cont (ord '\t') buf'
751 'v'# -> cont (ord '\v') buf'
752 '\\'# -> cont (ord '\\') buf'
753 '"'# -> cont (ord '\"') buf'
754 '\''# -> cont (ord '\'') buf'
755 '^'# -> let c = currentChar# buf' in
756 if c `geChar#` '@'# && c `leChar#` '_'#
757 then cont (I# (ord# c -# ord# '@'#)) (incLexeme buf')
760 'x'# -> readNum (after_charnum cont) buf' is_hexdigit 16 hex
761 'o'# -> readNum (after_charnum cont) buf' is_octdigit 8 oct_or_dec
763 -> readNum (after_charnum cont) buf is_digit 10 oct_or_dec
765 _ -> case [ (c,buf2) | (p,c) <- silly_escape_chars,
766 Just buf2 <- [prefixMatch buf p] ] of
767 (c,buf2):_ -> cont (ord c) buf2
770 after_charnum cont i buf
771 = if i >= 0 && i <= 0x10FFFF
772 then cont (fromInteger i) buf
775 readNum cont buf is_digit base conv = read buf 0
777 = case currentChar# buf of { c ->
779 then read (incLexeme buf) (i*base + (toInteger (I# (conv c))))
785 || (c `geChar#` 'a'# && c `leChar#` 'f'#)
786 || (c `geChar#` 'A'# && c `leChar#` 'F'#)
788 hex c | is_digit c = ord# c -# ord# '0'#
789 | otherwise = ord# (to_lower c) -# ord# 'a'# +# 10#
790 oct_or_dec c = ord# c -# ord# '0'#
792 is_octdigit c = c `geChar#` '0'# && c `leChar#` '7'#
795 | c `geChar#` 'A'# && c `leChar#` 'Z'#
796 = chr# (ord# c -# (ord# 'A'# -# ord# 'a'#))
799 charError buf = lexError "error in character literal" buf
801 silly_escape_chars = [
838 -------------------------------------------------------------------------------
840 lex_demand cont buf =
841 case read_em [] buf of { (ls,buf') ->
842 case currentChar# buf' of
843 'b'# -> cont (ITstrict (StrictSig (mkTopDmdType ls BotRes))) (incLexeme buf')
844 'm'# -> cont (ITstrict (StrictSig (mkTopDmdType ls RetCPR))) (incLexeme buf')
845 _ -> cont (ITstrict (StrictSig (mkTopDmdType ls TopRes))) buf'
849 case currentChar# buf of
850 'T'# -> read_em (Top : acc) (stepOn buf)
851 'L'# -> read_em (lazyDmd : acc) (stepOn buf)
852 'A'# -> read_em (Abs : acc) (stepOn buf)
853 'V'# -> read_em (evalDmd : acc) (stepOn buf) -- Temporary, until
854 -- we've recompiled prelude etc
855 'C'# -> do_unary Call acc (stepOnBy# buf 2#) -- Skip 'C('
857 'U'# -> do_seq1 Eval acc (stepOnBy# buf 1#)
858 'D'# -> do_seq1 Defer acc (stepOnBy# buf 1#)
859 'S'# -> do_seq1 (Box . Eval) acc (stepOnBy# buf 1#)
861 _ -> (reverse acc, buf)
864 = case currentChar# buf of
865 '('# -> do_seq2 fn acc (stepOnBy# buf 1#)
866 _ -> read_em (fn (Poly Abs) : acc) buf
869 = case read_em [] buf of { (dmds, buf) ->
870 case currentChar# buf of
871 ')'# -> read_em (fn (Prod dmds) : acc)
873 '*'# -> ASSERT( length dmds == 1 )
874 read_em (fn (Poly (head dmds)) : acc)
875 (stepOnBy# buf 2#) -- Skip '*)'
879 = case read_em [] buf of
880 ([dmd], rest) -> read_em (fn dmd : acc) (stepOn rest) -- Skip ')'
884 case currentChar# buf of
885 'C'# -> cont ITsccAllCafs (incLexeme buf)
886 other -> cont ITscc buf
888 -----------------------------------------------------------------------------
891 lex_num :: (Token -> P a) -> Int# -> Integer -> P a
892 lex_num cont glaexts acc buf =
893 case scanNumLit acc buf of
895 case currentChar# buf' of
896 '.'# | is_digit (lookAhead# buf' 1#) ->
897 -- this case is not optimised at all, as the
898 -- presence of floating point numbers in interface
899 -- files is not that common. (ToDo)
900 case expandWhile# is_digit (incLexeme buf') of
901 buf2 -> -- points to first non digit char
903 let l = case currentChar# buf2 of
909 = let buf3 = incLexeme buf2 in
910 case currentChar# buf3 of
911 '-'# -> expandWhile# is_digit (incLexeme buf3)
912 '+'# -> expandWhile# is_digit (incLexeme buf3)
913 x | is_digit x -> expandWhile# is_digit buf3
916 v = readRational__ (lexemeToString l)
918 in case currentChar# l of -- glasgow exts only
919 '#'# | flag glaexts -> let l' = incLexeme l in
920 case currentChar# l' of
921 '#'# -> cont (ITprimdouble v) (incLexeme l')
922 _ -> cont (ITprimfloat v) l'
923 _ -> cont (ITrational v) l
925 _ -> after_lexnum cont glaexts acc' buf'
927 after_lexnum cont glaexts i buf
928 = case currentChar# buf of
929 '#'# | flag glaexts -> cont (ITprimint i) (incLexeme buf)
930 _ -> cont (ITinteger i) buf
932 -----------------------------------------------------------------------------
933 -- C "literal literal"s (i.e. things like ``NULL'', ``stdout'' etc.)
935 -- we lexemeToFastString on the bit between the ``''s, but include the
936 -- quotes in the full lexeme.
938 lex_cstring cont buf =
939 case expandUntilMatch (stepOverLexeme buf) "\'\'" of
940 Just buf' -> cont (ITlitlit (lexemeToFastString
941 (setCurrentPos# buf' (negateInt# 2#))))
942 (mergeLexemes buf buf')
943 Nothing -> lexError "unterminated ``" buf
945 -----------------------------------------------------------------------------
946 -- identifiers, symbols etc.
948 lex_ip ip_constr cont buf =
949 case expandWhile# is_ident buf of
950 buf' -> cont (ip_constr (tailFS lexeme)) buf'
951 where lexeme = lexemeToFastString buf'
953 lex_id cont glaexts buf =
954 let buf1 = expandWhile# is_ident buf in
957 case (if flag glaexts
958 then expandWhile# (eqChar# '#'#) buf1 -- slurp trailing hashes
959 else buf1) of { buf' ->
961 let lexeme = lexemeToFastString buf' in
963 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
964 Just kwd_token -> --trace ("hkeywd: "++_UNPK_(lexeme)) $
968 let var_token = cont (ITvarid lexeme) buf' in
970 if not (flag glaexts)
974 case lookupUFM ghcExtensionKeywordsFM lexeme of {
975 Just kwd_token -> cont kwd_token buf';
982 case expandWhile# is_symbol buf of
983 buf' -> case lookupUFM haskellKeySymsFM lexeme of {
984 Just kwd_token -> --trace ("keysym: "++unpackFS lexeme) $
985 cont kwd_token buf' ;
986 Nothing -> --trace ("sym: "++unpackFS lexeme) $
987 cont (mk_var_token lexeme) buf'
989 where lexeme = lexemeToFastString buf'
992 -- lex_con recursively collects components of a qualified identifer.
993 -- The argument buf is the StringBuffer representing the lexeme
994 -- identified so far, where the next character is upper-case.
996 lex_con cont glaexts buf =
997 -- trace ("con: "{-++unpackFS lexeme-}) $
998 let empty_buf = stepOverLexeme buf in
999 case expandWhile# is_ident empty_buf of { buf1 ->
1000 case slurp_trailing_hashes buf1 glaexts of { con_buf ->
1002 let all_buf = mergeLexemes buf con_buf
1004 con_lexeme = lexemeToFastString con_buf
1005 mod_lexeme = lexemeToFastString (decLexeme buf)
1006 all_lexeme = lexemeToFastString all_buf
1009 | emptyLexeme buf = cont (ITconid con_lexeme) all_buf
1010 | otherwise = cont (ITqconid (mod_lexeme,con_lexeme)) all_buf
1013 case currentChar# all_buf of
1014 '.'# -> maybe_qualified cont glaexts all_lexeme
1015 (incLexeme all_buf) just_a_conid
1020 maybe_qualified cont glaexts mod buf just_a_conid =
1021 -- trace ("qid: "{-++unpackFS lexeme-}) $
1022 case currentChar# buf of
1023 '['# -> -- Special case for []
1024 case lookAhead# buf 1# of
1025 ']'# -> cont (ITqconid (mod,SLIT("[]"))) (setCurrentPos# buf 2#)
1028 '('# -> -- Special case for (,,,)
1029 -- This *is* necessary to deal with e.g. "instance C PrelBase.(,,)"
1030 case lookAhead# buf 1# of
1031 '#'# | flag glaexts -> case lookAhead# buf 2# of
1032 ','# -> lex_ubx_tuple cont mod (setCurrentPos# buf 3#)
1035 ')'# -> cont (ITqconid (mod,SLIT("()"))) (setCurrentPos# buf 2#)
1036 ','# -> lex_tuple cont mod (setCurrentPos# buf 2#) just_a_conid
1039 '-'# -> case lookAhead# buf 1# of
1040 '>'# -> cont (ITqconid (mod,SLIT("(->)"))) (setCurrentPos# buf 2#)
1041 _ -> lex_id3 cont glaexts mod buf just_a_conid
1043 _ -> lex_id3 cont glaexts mod buf just_a_conid
1046 lex_id3 cont glaexts mod buf just_a_conid
1047 | is_upper (currentChar# buf) =
1048 lex_con cont glaexts buf
1050 | is_symbol (currentChar# buf) =
1052 start_new_lexeme = stepOverLexeme buf
1054 -- trace ("lex_id31 "{-++unpackFS lexeme-}) $
1055 case expandWhile# is_symbol start_new_lexeme of { buf' ->
1057 lexeme = lexemeToFastString buf'
1058 -- real lexeme is M.<sym>
1059 new_buf = mergeLexemes buf buf'
1061 cont (mk_qvar_token mod lexeme) new_buf
1062 -- wrong, but arguably morally right: M... is now a qvarsym
1067 start_new_lexeme = stepOverLexeme buf
1069 -- trace ("lex_id32 "{-++unpackFS lexeme-}) $
1070 case expandWhile# is_ident start_new_lexeme of { buf1 ->
1075 case slurp_trailing_hashes buf1 glaexts of { buf' ->
1078 lexeme = lexemeToFastString buf'
1079 new_buf = mergeLexemes buf buf'
1080 is_a_qvarid = cont (mk_qvar_token mod lexeme) new_buf
1082 case _scc_ "Lex.haskellKeyword" lookupUFM haskellKeywordsFM lexeme of {
1083 Nothing -> is_a_qvarid ;
1085 Just kwd_token | isSpecial kwd_token -- special ids (as, qualified, hiding) shouldn't be
1086 -> is_a_qvarid -- recognised as keywords here.
1088 -> just_a_conid -- avoid M.where etc.
1091 slurp_trailing_hashes buf glaexts
1092 | flag glaexts = expandWhile# (`eqChar#` '#'#) buf
1097 | is_upper f = ITconid pk_str
1098 | is_ident f = ITvarid pk_str
1099 | f `eqChar#` ':'# = ITconsym pk_str
1100 | otherwise = ITvarsym pk_str
1102 (C# f) = _HEAD_ pk_str
1103 -- tl = _TAIL_ pk_str
1105 mk_qvar_token m token =
1106 -- trace ("mk_qvar ") $
1107 case mk_var_token token of
1108 ITconid n -> ITqconid (m,n)
1109 ITvarid n -> ITqvarid (m,n)
1110 ITconsym n -> ITqconsym (m,n)
1111 ITvarsym n -> ITqvarsym (m,n)
1112 _ -> ITunknown (show token)
1115 ----------------------------------------------------------------------------
1116 Horrible stuff for dealing with M.(,,,)
1119 lex_tuple cont mod buf back_off =
1123 case currentChar# buf of
1124 ','# -> go (n+1) (stepOn buf)
1125 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Boxed n))) (stepOn buf)
1128 lex_ubx_tuple cont mod buf back_off =
1132 case currentChar# buf of
1133 ','# -> go (n+1) (stepOn buf)
1134 '#'# -> case lookAhead# buf 1# of
1135 ')'# -> cont (ITqconid (mod, snd (mkTupNameStr Unboxed n)))
1141 -----------------------------------------------------------------------------
1142 'lexPragma' rips along really fast, looking for a '##-}',
1143 indicating the end of the pragma we're skipping
1146 lexPragma cont contf inStr buf =
1147 case currentChar# buf of
1148 '#'# | inStr ==# 0# ->
1149 case lookAhead# buf 1# of { '#'# ->
1150 case lookAhead# buf 2# of { '-'# ->
1151 case lookAhead# buf 3# of { '}'# ->
1152 contf cont (lexemeToBuffer buf)
1153 (stepOverLexeme (setCurrentPos# buf 4#));
1154 _ -> lexPragma cont contf inStr (incLexeme buf) };
1155 _ -> lexPragma cont contf inStr (incLexeme buf) };
1156 _ -> lexPragma cont contf inStr (incLexeme buf) }
1160 odd_slashes buf flg i# =
1161 case lookAhead# buf i# of
1162 '\\'# -> odd_slashes buf (not flg) (i# -# 1#)
1165 not_inStr = if inStr ==# 0# then 1# else 0#
1167 case lookAhead# buf (negateInt# 1#) of --backwards, actually
1168 '\\'# -> -- escaping something..
1169 if odd_slashes buf True (negateInt# 2#)
1170 then -- odd number of slashes, " is escaped.
1171 lexPragma cont contf inStr (incLexeme buf)
1172 else -- even number of slashes, \ is escaped.
1173 lexPragma cont contf not_inStr (incLexeme buf)
1174 _ -> lexPragma cont contf not_inStr (incLexeme buf)
1176 '\''# | inStr ==# 0# ->
1177 case lookAhead# buf 1# of { '"'# ->
1178 case lookAhead# buf 2# of { '\''# ->
1179 lexPragma cont contf inStr (setCurrentPos# buf 3#);
1180 _ -> lexPragma cont contf inStr (incLexeme buf) };
1181 _ -> lexPragma cont contf inStr (incLexeme buf) }
1183 -- a sign that the input is ill-formed, since pragmas are
1184 -- assumed to always be properly closed (in .hi files).
1185 '\NUL'# -> trace "lexPragma: unexpected end-of-file" $
1186 cont (ITunknown "\NUL") buf
1188 _ -> lexPragma cont contf inStr (incLexeme buf)
1192 -----------------------------------------------------------------------------
1203 data PState = PState {
1205 glasgow_exts :: Int#,
1208 context :: [LayoutContext]
1211 type P a = StringBuffer -- Input string
1216 returnP a buf s = POk s a
1218 thenP :: P a -> (a -> P b) -> P b
1219 m `thenP` k = \ buf s ->
1221 POk s1 a -> k a buf s1
1222 PFailed err -> PFailed err
1224 thenP_ :: P a -> P b -> P b
1225 m `thenP_` k = m `thenP` \_ -> k
1227 mapP :: (a -> P b) -> [a] -> P [b]
1228 mapP f [] = returnP []
1231 mapP f as `thenP` \bs ->
1234 failP :: String -> P a
1235 failP msg buf s = PFailed (text msg)
1237 failMsgP :: Message -> P a
1238 failMsgP msg buf s = PFailed msg
1240 lexError :: String -> P a
1241 lexError str buf s@PState{ loc = loc }
1242 = failMsgP (hcat [ppr loc, text ": ", text str]) buf s
1244 getSrcLocP :: P SrcLoc
1245 getSrcLocP buf s@(PState{ loc = loc }) = POk s loc
1247 -- use a temporary SrcLoc for the duration of the argument
1248 setSrcLocP :: SrcLoc -> P a -> P a
1249 setSrcLocP new_loc p buf s =
1250 case p buf s{ loc=new_loc } of
1252 PFailed e -> PFailed e
1254 getSrcFile :: P FAST_STRING
1255 getSrcFile buf s@(PState{ loc = loc }) = POk s (srcLocFile loc)
1257 pushContext :: LayoutContext -> P ()
1258 pushContext ctxt buf s@(PState{ context = ctx }) = POk s{context = ctxt:ctx} ()
1262 This special case in layoutOn is to handle layout contexts with are
1263 indented the same or less than the current context. This is illegal
1264 according to the Haskell spec, so we have to arrange to close the
1265 current context. eg.
1270 after the first 'where', the sequence of events is:
1272 - layout system inserts a ';' (column 0)
1273 - parser begins a new context at column 0
1274 - parser shifts ';' (legal empty declaration)
1275 - parser sees 'class': parse error (we're still in the inner context)
1277 trouble is, by the time we know we need a new context, the lexer has
1278 already generated the ';'. Hacky solution is as follows: since we
1279 know the column of the next token (it's the column number of the new
1280 context), we set the ACTUAL column number of the new context to this
1281 numer plus one. Hence the next time the lexer is called, a '}' will
1282 be generated to close the new context straight away. Furthermore, we
1283 have to set the atbol flag so that the ';' that the parser shifted as
1284 part of the new context is re-generated.
1286 when the new context is *less* indented than the current one:
1288 f = f where g = g where
1291 - current context: column 12.
1292 - on seeing 'h' (column 0), the layout system inserts '}'
1293 - parser starts a new context, column 0
1294 - parser sees '}', uses it to close new context
1295 - we still need to insert another '}' followed by a ';',
1296 hence the atbol trick.
1298 There's also a special hack in here to deal with
1305 i.e. the inner context is at the same indentation level as the outer
1306 context. This is strictly illegal according to Haskell 98, but
1307 there's a lot of existing code using this style and it doesn't make
1308 any sense to disallow it, since empty 'do' lists don't make sense.
1311 layoutOn :: Bool -> P ()
1312 layoutOn strict buf s@(PState{ bol = bol, context = ctx }) =
1313 let offset = lexemeIndex buf -# bol in
1316 | if strict then prev_off >=# offset else prev_off ># offset ->
1317 --trace ("layout on, column: " ++ show (I# offset)) $
1318 POk s{ context = Layout (offset +# 1#) : ctx, atbol = 1# } ()
1320 --trace ("layout on, column: " ++ show (I# offset)) $
1321 POk s{ context = Layout offset : ctx } ()
1324 layoutOff buf s@(PState{ context = ctx }) =
1325 POk s{ context = NoLayout:ctx } ()
1328 popContext = \ buf s@(PState{ context = ctx, loc = loc }) ->
1330 (_:tl) -> POk s{ context = tl } ()
1331 [] -> PFailed (srcParseErr buf loc)
1334 Note that if the name of the file we're processing ends
1335 with `hi-boot', we accept it on faith as having the right
1336 version. This is done so that .hi-boot files that comes
1337 with hsc don't have to be updated before every release,
1338 *and* it allows us to share .hi-boot files with versions
1339 of hsc that don't have .hi version checking (e.g., ghc-2.10's)
1341 If the version number is 0, the checking is also turned off.
1342 (needed to deal with GHC.hi only!)
1344 Once we can assume we're compiling with a version of ghc that
1345 supports interface file checking, we can drop the special
1348 checkVersion :: Maybe Integer -> P ()
1349 checkVersion mb@(Just v) buf s@(PState{loc = loc})
1350 | (v==0) || (v == fromInt opt_HiVersion) || opt_NoHiCheck = POk s ()
1351 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1352 checkVersion mb@Nothing buf s@(PState{loc = loc})
1353 | "hi-boot" `isSuffixOf` (_UNPK_ (srcLocFile loc)) = POk s ()
1354 | otherwise = PFailed (ifaceVersionErr mb loc ([]::[Token]){-Todo-})
1356 -----------------------------------------------------------------
1358 ifaceParseErr :: StringBuffer -> SrcLoc -> Message
1360 = hsep [ppr l, ptext SLIT("Interface file parse error; on input `"),
1361 text (lexemeToString s), char '\'']
1363 ifaceVersionErr hi_vers l toks
1364 = hsep [ppr l, ptext SLIT("Interface file version error;"),
1365 ptext SLIT("Expected"), int opt_HiVersion,
1366 ptext SLIT("found "), pp_version]
1370 Nothing -> ptext SLIT("pre ghc-3.02 version")
1371 Just v -> ptext SLIT("version") <+> integer v
1373 -----------------------------------------------------------------------------
1375 srcParseErr :: StringBuffer -> SrcLoc -> Message
1379 then ptext SLIT(": parse error (possibly incorrect indentation)")
1380 else hcat [ptext SLIT(": parse error on input "),
1381 char '`', text token, char '\'']
1384 token = lexemeToString s