2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
5 \section[OccName]{@OccName@}
9 -- The NameSpace type; abstact
10 NameSpace, tcName, clsName, tcClsName, dataName, varName, ipName,
11 tvName, nameSpaceString,
14 OccName, -- Abstract, instance of Outputable
17 mkOccFS, mkSysOcc, mkSysOccFS, mkCCallOcc, mkVarOcc, mkKindOccFS,
18 mkSuperDictSelOcc, mkDFunOcc, mkForeignExportOcc,
19 mkDictOcc, mkIPOcc, mkWorkerOcc, mkMethodOcc, mkDefaultMethodOcc,
20 mkDerivedTyConOcc, mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc,
23 isSysOcc, isTvOcc, isDataOcc, isDataSymOcc, isSymOcc, isIPOcc, isValOcc,
25 occNameFS, occNameString, occNameUserString, occNameSpace, occNameFlavour,
29 TidyOccEnv, emptyTidyOccEnv, tidyOccName, initTidyOccEnv,
32 EncodedString, EncodedFS, UserString, UserFS, encode, encodeFS, decode, pprEncodedFS,
34 -- The basic form of names
35 isLexCon, isLexVar, isLexId, isLexSym,
36 isLexConId, isLexConSym, isLexVarId, isLexVarSym,
37 isLowerISO, isUpperISO
41 #include "HsVersions.h"
43 import Char ( isDigit, isUpper, isLower, ISALPHANUM, ord, chr, digitToInt )
44 import Util ( thenCmp )
45 import FiniteMap ( FiniteMap, emptyFM, lookupFM, addToFM, elemFM )
50 We hold both module names and identifier names in a 'Z-encoded' form
51 that makes them acceptable both as a C identifier and as a Haskell
54 They can always be decoded again when printing error messages
55 or anything else for the user, but it does make sense for it
56 to be represented here in encoded form, so that when generating
57 code the encoding operation is not performed on each occurrence.
59 These type synonyms help documentation.
62 type UserFS = FAST_STRING -- As the user typed it
63 type EncodedFS = FAST_STRING -- Encoded form
65 type UserString = String -- As the user typed it
66 type EncodedString = String -- Encoded form
69 pprEncodedFS :: EncodedFS -> SDoc
71 = getPprStyle $ \ sty ->
73 -- ptext (decodeFS fs) would needlessly pack the string again
74 then text (decode (_UNPK_ fs))
78 %************************************************************************
80 \subsection{Name space}
82 %************************************************************************
85 data NameSpace = VarName -- Variables
86 | IPName -- Implicit Parameters
87 | DataName -- Data constructors
88 | TvName -- Type variables
89 | TcClsName -- Type constructors and classes; Haskell has them
90 -- in the same name space for now.
93 -- Though type constructors and classes are in the same name space now,
94 -- the NameSpace type is abstract, so we can easily separate them later
95 tcName = TcClsName -- Type constructors
96 clsName = TcClsName -- Classes
97 tcClsName = TcClsName -- Not sure which!
105 nameSpaceString :: NameSpace -> String
106 nameSpaceString DataName = "Data constructor"
107 nameSpaceString VarName = "Variable"
108 nameSpaceString IPName = "Implicit Param"
109 nameSpaceString TvName = "Type variable"
110 nameSpaceString TcClsName = "Type constructor or class"
114 %************************************************************************
116 \subsection[Name-pieces-datatypes]{The @OccName@ datatypes}
118 %************************************************************************
121 data OccName = OccName
128 instance Eq OccName where
129 (OccName sp1 s1) == (OccName sp2 s2) = s1 == s2 && sp1 == sp2
131 instance Ord OccName where
132 compare (OccName sp1 s1) (OccName sp2 s2) = (s1 `compare` s2) `thenCmp`
137 %************************************************************************
139 \subsection{Printing}
141 %************************************************************************
144 instance Outputable OccName where
147 pprOccName :: OccName -> SDoc
148 pprOccName (OccName sp occ) = pprEncodedFS occ
152 %************************************************************************
154 \subsection{Construction}
156 %************************************************************************
158 *Sys* things do no encoding; the caller should ensure that the thing is
162 mkSysOcc :: NameSpace -> EncodedString -> OccName
163 mkSysOcc occ_sp str = ASSERT2( alreadyEncoded str, text str )
164 OccName occ_sp (_PK_ str)
166 mkSysOccFS :: NameSpace -> EncodedFS -> OccName
167 mkSysOccFS occ_sp fs = ASSERT2( alreadyEncodedFS fs, ppr fs )
170 mkCCallOcc :: EncodedString -> OccName
171 -- This version of mkSysOcc doesn't check that the string is already encoded,
172 -- because it will be something like "{__ccall f dyn Int# -> Int#}"
173 -- This encodes a lot into something that then parses like an Id.
174 -- But then alreadyEncoded complains about the braces!
175 mkCCallOcc str = OccName varName (_PK_ str)
177 -- Kind constructors get a special function. Uniquely, they are not encoded,
178 -- so that they have names like '*'. This means that *even in interface files*
179 -- we'll get kinds like (* -> (* -> *)). We can't use mkSysOcc because it
180 -- has an ASSERT that doesn't hold.
181 mkKindOccFS :: NameSpace -> EncodedFS -> OccName
182 mkKindOccFS occ_sp fs = OccName occ_sp fs
185 *Source-code* things are encoded.
188 mkOccFS :: NameSpace -> UserFS -> OccName
189 mkOccFS occ_sp fs = mkSysOccFS occ_sp (encodeFS fs)
191 mkVarOcc :: UserFS -> OccName
192 mkVarOcc fs = mkSysOccFS varName (encodeFS fs)
197 %************************************************************************
199 \subsection{Predicates and taking them apart}
201 %************************************************************************
204 occNameFS :: OccName -> EncodedFS
205 occNameFS (OccName _ s) = s
207 occNameString :: OccName -> EncodedString
208 occNameString (OccName _ s) = _UNPK_ s
210 occNameUserString :: OccName -> UserString
211 occNameUserString occ = decode (occNameString occ)
213 occNameSpace :: OccName -> NameSpace
214 occNameSpace (OccName sp _) = sp
216 setOccNameSpace :: OccName -> NameSpace -> OccName
217 setOccNameSpace (OccName _ occ) sp = OccName sp occ
219 -- occNameFlavour is used only to generate good error messages
220 occNameFlavour :: OccName -> String
221 occNameFlavour (OccName sp _) = nameSpaceString sp
225 isTvOcc, isDataSymOcc, isSymOcc :: OccName -> Bool
227 isTvOcc (OccName TvName _) = True
228 isTvOcc other = False
230 isValOcc (OccName VarName _) = True
231 isValOcc (OccName DataName _) = True
232 isValOcc other = False
234 -- Data constructor operator (starts with ':', or '[]')
235 -- Pretty inefficient!
236 isDataSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
237 isDataSymOcc other = False
239 isDataOcc (OccName DataName _) = True
240 isDataOcc other = False
242 -- Any operator (data constructor or variable)
243 -- Pretty inefficient!
244 isSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
245 isSymOcc (OccName VarName s) = isLexSym (decodeFS s)
247 isIPOcc (OccName IPName _) = True
252 %************************************************************************
254 \subsection{Making system names}
256 %************************************************************************
258 Here's our convention for splitting up the interface file name space:
260 d... dictionary identifiers
261 (local variables, so no name-clash worries)
263 $f... dict-fun identifiers (from inst decls)
264 $dm... default methods
265 $p... superclass selectors
267 $T... compiler-generated tycons for dictionaries
268 $D... ...ditto data cons
269 $sf.. specialised version of f
271 in encoded form these appear as Zdfxxx etc
273 :... keywords (export:, letrec: etc.)
275 This knowledge is encoded in the following functions.
278 @mk_deriv@ generates an @OccName@ from the one-char prefix and a string.
279 NB: The string must already be encoded!
282 mk_deriv :: NameSpace
283 -> String -- Distinguishes one sort of derived name from another
284 -> EncodedString -- Must be already encoded!! We don't want to encode it a
285 -- second time because encoding isn't itempotent
288 mk_deriv occ_sp sys_prefix str = mkSysOcc occ_sp (encode sys_prefix ++ str)
292 mkDictOcc, mkIPOcc, mkWorkerOcc, mkDefaultMethodOcc,
293 mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc
294 :: OccName -> OccName
296 -- These derived variables have a prefix that no Haskell value could have
297 mkWorkerOcc = mk_simple_deriv varName "$w"
298 mkDefaultMethodOcc = mk_simple_deriv varName "$dm"
299 mkDerivedTyConOcc = mk_simple_deriv tcName ":" -- The : prefix makes sure it classifies
300 mkClassTyConOcc = mk_simple_deriv tcName ":T" -- as a tycon/datacon
301 mkClassDataConOcc = mk_simple_deriv dataName ":D" --
302 mkDictOcc = mk_simple_deriv varName "$d"
303 mkIPOcc = mk_simple_deriv varName "$i"
304 mkSpecOcc = mk_simple_deriv varName "$s"
305 mkForeignExportOcc = mk_simple_deriv varName "$f"
306 mkGenOcc1 = mk_simple_deriv varName "$gfrom" -- Generics
307 mkGenOcc2 = mk_simple_deriv varName "$gto" -- Generics
308 mk_simple_deriv sp px occ = mk_deriv sp px (occNameString occ)
311 isSysOcc :: OccName -> Bool -- True for all these '$' things
312 isSysOcc occ = case occNameUserString occ of
314 other -> False -- We don't care about the ':' ones
315 -- isSysOcc is only called for Ids anyway
319 mkSuperDictSelOcc :: Int -- Index of superclass, eg 3
320 -> OccName -- Class, eg "Ord"
321 -> OccName -- eg "p3Ord"
322 mkSuperDictSelOcc index cls_occ
323 = mk_deriv varName "$p" (show index ++ occNameString cls_occ)
328 mkDFunOcc :: EncodedString -- Typically the class and type glommed together e.g. "OrdMaybe"
329 -> OccName -- "$fOrdMaybe"
331 mkDFunOcc string = mk_deriv VarName "$f" string
334 We used to add a '$m' to indicate a method, but that gives rise to bad
335 error messages from the type checker when we print the function name or pattern
336 of an instance-decl binding. Why? Because the binding is zapped
337 to use the method name in place of the selector name.
338 (See TcClassDcl.tcMethodBind)
340 The way it is now, -ddump-xx output may look confusing, but
341 you can always say -dppr-debug to get the uniques.
343 However, we *do* have to zap the first character to be lower case,
344 because overloaded constructors (blarg) generate methods too.
345 And convert to VarName space
347 e.g. a call to constructor MkFoo where
348 data (Ord a) => Foo a = MkFoo a
350 If this is necessary, we do it by prefixing '$m'. These
351 guys never show up in error messages. What a hack.
354 mkMethodOcc :: OccName -> OccName
355 mkMethodOcc occ@(OccName VarName fs) = occ
356 mkMethodOcc occ = mk_simple_deriv varName "$m" occ
360 %************************************************************************
362 \subsection{Tidying them up}
364 %************************************************************************
366 Before we print chunks of code we like to rename it so that
367 we don't have to print lots of silly uniques in it. But we mustn't
368 accidentally introduce name clashes! So the idea is that we leave the
369 OccName alone unless it accidentally clashes with one that is already
370 in scope; if so, we tack on '1' at the end and try again, then '2', and
371 so on till we find a unique one.
373 There's a wrinkle for operators. Consider '>>='. We can't use '>>=1'
374 because that isn't a single lexeme. So we encode it to 'lle' and *then*
375 tack on the '1', if necessary.
378 type TidyOccEnv = FiniteMap FAST_STRING Int -- The in-scope OccNames
379 emptyTidyOccEnv = emptyFM
381 initTidyOccEnv :: [OccName] -> TidyOccEnv -- Initialise with names to avoid!
382 initTidyOccEnv = foldl (\env (OccName _ fs) -> addToFM env fs 1) emptyTidyOccEnv
384 tidyOccName :: TidyOccEnv -> OccName -> (TidyOccEnv, OccName)
386 tidyOccName in_scope occ@(OccName occ_sp fs)
387 | not (fs `elemFM` in_scope)
388 = (addToFM in_scope fs 1, occ) -- First occurrence
390 | otherwise -- Already occurs
391 = go in_scope (_UNPK_ fs)
394 go in_scope str = case lookupFM in_scope pk_str of
395 Just n -> go (addToFM in_scope pk_str (n+1)) (str ++ show n)
396 -- Need to go round again, just in case "t3" (say)
397 -- clashes with a "t3" that's already in scope
399 Nothing -> (addToFM in_scope pk_str 1, mkSysOccFS occ_sp pk_str)
406 %************************************************************************
408 \subsection{The 'Z' encoding}
410 %************************************************************************
412 This is the main name-encoding and decoding function. It encodes any
413 string into a string that is acceptable as a C name. This is the name
414 by which things are known right through the compiler.
416 The basic encoding scheme is this.
418 * Tuples (,,,) are coded as Z3T
420 * Alphabetic characters (upper and lower) and digits
421 all translate to themselves;
422 except 'Z', which translates to 'ZZ'
423 and 'z', which translates to 'zz'
424 We need both so that we can preserve the variable/tycon distinction
426 * Most other printable characters translate to 'zx' or 'Zx' for some
427 alphabetic character x
429 * The others translate as 'znnnU' where 'nnn' is the decimal number
433 --------------------------
448 -- alreadyEncoded is used in ASSERTs to check for encoded
449 -- strings. It isn't fail-safe, of course, because, say 'zh' might
450 -- be encoded or not.
451 alreadyEncoded :: String -> Bool
452 alreadyEncoded s = all ok s
455 -- This is a bit of a lie; if we really wanted spaces
456 -- in names we'd have to encode them. But we do put
457 -- spaces in ccall "occurrences", and we don't want to
459 ok ch = ISALPHANUM ch
461 alreadyEncodedFS :: FAST_STRING -> Bool
462 alreadyEncodedFS fs = alreadyEncoded (_UNPK_ fs)
464 encode :: UserString -> EncodedString
465 encode cs = case maybe_tuple cs of
466 Just n -> 'Z' : show n ++ "T" -- Tuples go to Z2T etc
470 go (c:cs) = encode_ch c ++ go cs
472 -- ToDo: Unboxed tuples too, perhaps?
473 maybe_tuple ('(' : cs) = check_tuple (0::Int) cs
474 maybe_tuple other = Nothing
476 check_tuple :: Int -> String -> Maybe Int
477 check_tuple n (',' : cs) = check_tuple (n+1) cs
478 check_tuple n ")" = Just n
479 check_tuple n other = Nothing
481 encodeFS :: UserFS -> EncodedFS
482 encodeFS fast_str | all unencodedChar str = fast_str
483 | otherwise = _PK_ (encode str)
485 str = _UNPK_ fast_str
487 unencodedChar :: Char -> Bool -- True for chars that don't need encoding
488 unencodedChar 'Z' = False
489 unencodedChar 'z' = False
490 unencodedChar c = ISALPHANUM c
492 encode_ch :: Char -> EncodedString
493 encode_ch c | unencodedChar c = [c] -- Common case first
496 encode_ch '(' = "ZL" -- Needed for things like (,), and (->)
497 encode_ch ')' = "ZR" -- For symmetry with (
517 encode_ch '\'' = "zq"
518 encode_ch '\\' = "zr"
523 encode_ch c = 'z' : shows (ord c) "U"
526 Decode is used for user printing.
529 decodeFS :: FAST_STRING -> FAST_STRING
530 decodeFS fs = _PK_ (decode (_UNPK_ fs))
532 decode :: EncodedString -> UserString
534 decode ('Z' : rest) = decode_escape rest
535 decode ('z' : rest) = decode_escape rest
536 decode (c : rest) = c : decode rest
538 decode_escape :: EncodedString -> UserString
540 decode_escape ('L' : rest) = '(' : decode rest
541 decode_escape ('R' : rest) = ')' : decode rest
542 decode_escape ('M' : rest) = '[' : decode rest
543 decode_escape ('N' : rest) = ']' : decode rest
544 decode_escape ('C' : rest) = ':' : decode rest
545 decode_escape ('Z' : rest) = 'Z' : decode rest
547 decode_escape ('z' : rest) = 'z' : decode rest
548 decode_escape ('a' : rest) = '&' : decode rest
549 decode_escape ('b' : rest) = '|' : decode rest
550 decode_escape ('c' : rest) = '^' : decode rest
551 decode_escape ('d' : rest) = '$' : decode rest
552 decode_escape ('e' : rest) = '=' : decode rest
553 decode_escape ('g' : rest) = '>' : decode rest
554 decode_escape ('h' : rest) = '#' : decode rest
555 decode_escape ('i' : rest) = '.' : decode rest
556 decode_escape ('l' : rest) = '<' : decode rest
557 decode_escape ('m' : rest) = '-' : decode rest
558 decode_escape ('n' : rest) = '!' : decode rest
559 decode_escape ('p' : rest) = '+' : decode rest
560 decode_escape ('q' : rest) = '\'' : decode rest
561 decode_escape ('r' : rest) = '\\' : decode rest
562 decode_escape ('s' : rest) = '/' : decode rest
563 decode_escape ('t' : rest) = '*' : decode rest
564 decode_escape ('u' : rest) = '_' : decode rest
565 decode_escape ('v' : rest) = '%' : decode rest
567 -- Tuples are coded as Z23T
568 -- Characters not having a specific code are coded as z224U
569 decode_escape (c : rest)
570 | isDigit c = go (digitToInt c) rest
572 go n (c : rest) | isDigit c = go (10*n + digitToInt c) rest
573 go n ('T' : rest) = '(' : replicate n ',' ++ ')' : decode rest
574 go n ('U' : rest) = chr n : decode rest
575 go n other = pprPanic "decode_escape" (ppr n <+> text (c:rest))
577 decode_escape (c : rest) = pprTrace "decode_escape" (char c) (decode rest)
581 %************************************************************************
583 n\subsection{Lexical categories}
585 %************************************************************************
587 These functions test strings to see if they fit the lexical categories
588 defined in the Haskell report.
591 isLexCon, isLexVar, isLexId, isLexSym :: FAST_STRING -> Bool
592 isLexConId, isLexConSym, isLexVarId, isLexVarSym :: FAST_STRING -> Bool
594 isLexCon cs = isLexConId cs || isLexConSym cs
595 isLexVar cs = isLexVarId cs || isLexVarSym cs
597 isLexId cs = isLexConId cs || isLexVarId cs
598 isLexSym cs = isLexConSym cs || isLexVarSym cs
602 isLexConId cs -- Prefix type or data constructors
603 | _NULL_ cs = False -- e.g. "Foo", "[]", "(,)"
604 | cs == SLIT("[]") = True
605 | otherwise = startsConId (_HEAD_ cs)
607 isLexVarId cs -- Ordinary prefix identifiers
608 | _NULL_ cs = False -- e.g. "x", "_x"
609 | otherwise = startsVarId (_HEAD_ cs)
611 isLexConSym cs -- Infix type or data constructors
612 | _NULL_ cs = False -- e.g. ":-:", ":", "->"
613 | cs == SLIT("->") = True
614 | otherwise = startsConSym (_HEAD_ cs)
616 isLexVarSym cs -- Infix identifiers
617 | _NULL_ cs = False -- e.g. "+"
618 | otherwise = startsVarSym (_HEAD_ cs)
621 startsVarSym, startsVarId, startsConSym, startsConId :: Char -> Bool
622 startsVarSym c = isSymbolASCII c || isSymbolISO c -- Infix Ids
623 startsConSym c = c == ':' -- Infix data constructors
624 startsVarId c = isLower c || isLowerISO c || c == '_' -- Ordinary Ids
625 startsConId c = isUpper c || isUpperISO c || c == '(' -- Ordinary type constructors and data constructors
628 isSymbolASCII c = c `elem` "!#$%&*+./<=>?@\\^|~-"
629 isSymbolISO c = ord c `elem` (0xd7 : 0xf7 : [0xa1 .. 0xbf])
630 isUpperISO (C# c#) = c# `geChar#` '\xc0'# && c# `leChar#` '\xde'# && c# `neChar#` '\xd7'#
631 --0xc0 <= oc && oc <= 0xde && oc /= 0xd7 where oc = ord c
632 isLowerISO (C# c#) = c# `geChar#` '\xdf'# && c# `leChar#` '\xff'# && c# `neChar#` '\xf7'#
633 --0xdf <= oc && oc <= 0xff && oc /= 0xf7 where oc = ord c