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, tvName,
14 OccName, -- Abstract, instance of Outputable
17 mkSrcOccFS, mkSysOcc, mkSysOccFS, mkSrcVarOcc, mkKindOccFS,
18 mkSuperDictSelOcc, mkDFunOcc, mkForeignExportOcc,
19 mkDictOcc, mkWorkerOcc, mkMethodOcc, mkDefaultMethodOcc,
20 mkDerivedTyConOcc, mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc,
22 isTvOcc, isDataOcc, isDataSymOcc, isSymOcc,
24 occNameFS, occNameString, occNameUserString, occNameSpace, occNameFlavour,
28 TidyOccEnv, emptyTidyOccEnv, tidyOccName, initTidyOccEnv,
31 EncodedString, EncodedFS, UserString, UserFS, encode, encodeFS, decode, pprEncodedFS,
33 -- The basic form of names
34 isLexCon, isLexVar, isLexId, isLexSym,
35 isLexConId, isLexConSym, isLexVarId, isLexVarSym,
36 isLowerISO, isUpperISO
40 #include "HsVersions.h"
42 import Char ( isDigit, isAlpha, isUpper, isLower, ISALPHANUM, ord, chr, digitToInt, intToDigit )
43 import Util ( thenCmp )
44 import FiniteMap ( FiniteMap, emptyFM, lookupFM, addToFM, elemFM )
49 We hold both module names and identifier names in a 'Z-encoded' form
50 that makes them acceptable both as a C identifier and as a Haskell
53 They can always be decoded again when printing error messages
54 or anything else for the user, but it does make sense for it
55 to be represented here in encoded form, so that when generating
56 code the encoding operation is not performed on each occurrence.
58 These type synonyms help documentation.
61 type UserFS = FAST_STRING -- As the user typed it
62 type EncodedFS = FAST_STRING -- Encoded form
64 type UserString = String -- As the user typed it
65 type EncodedString = String -- Encoded form
68 pprEncodedFS :: EncodedFS -> SDoc
70 = getPprStyle $ \ sty ->
72 text (decode (_UNPK_ fs))
77 %************************************************************************
79 \subsection{Name space}
81 %************************************************************************
84 data NameSpace = VarName -- Variables
85 | DataName -- Data constructors
86 | TvName -- Type variables
87 | TcClsName -- Type constructors and classes; Haskell has them
88 -- in the same name space for now.
91 -- Though type constructors and classes are in the same name space now,
92 -- the NameSpace type is abstract, so we can easily separate them later
93 tcName = TcClsName -- Type constructors
94 clsName = TcClsName -- Classes
95 tcClsName = TcClsName -- Not sure which!
102 nameSpaceString :: NameSpace -> String
103 nameSpaceString DataName = "Data constructor"
104 nameSpaceString VarName = "Variable"
105 nameSpaceString TvName = "Type variable"
106 nameSpaceString TcClsName = "Type constructor or class"
110 %************************************************************************
112 \subsection[Name-pieces-datatypes]{The @OccName@ datatypes}
114 %************************************************************************
117 data OccName = OccName
124 instance Eq OccName where
125 (OccName sp1 s1) == (OccName sp2 s2) = s1 == s2 && sp1 == sp2
127 instance Ord OccName where
128 compare (OccName sp1 s1) (OccName sp2 s2) = (s1 `compare` s2) `thenCmp`
133 %************************************************************************
135 \subsection{Printing}
137 %************************************************************************
140 instance Outputable OccName where
143 pprOccName :: OccName -> SDoc
144 pprOccName (OccName sp occ) = pprEncodedFS occ
148 %************************************************************************
150 \subsection{Construction}
152 %************************************************************************
154 *Sys* things do no encoding; the caller should ensure that the thing is
158 mkSysOcc :: NameSpace -> EncodedString -> OccName
159 mkSysOcc occ_sp str = ASSERT( alreadyEncoded str )
160 OccName occ_sp (_PK_ str)
162 mkSysOccFS :: NameSpace -> EncodedFS -> OccName
163 mkSysOccFS occ_sp fs = ASSERT2( alreadyEncodedFS fs, ppr fs )
166 -- Kind constructors get a speical function. Uniquely, they are not encoded,
167 -- so that they have names like '*'. This means that *even in interface files*
168 -- we'll get kinds like (* -> (* -> *)). We can't use mkSysOcc because it
169 -- has an ASSERT that doesn't hold.
170 mkKindOccFS :: NameSpace -> EncodedFS -> OccName
171 mkKindOccFS occ_sp fs = OccName occ_sp fs
174 *Source-code* things are encoded.
177 mkSrcOccFS :: NameSpace -> UserFS -> OccName
178 mkSrcOccFS occ_sp fs = mkSysOccFS occ_sp (encodeFS fs)
180 mkSrcVarOcc :: UserFS -> OccName
181 mkSrcVarOcc fs = mkSysOccFS varName (encodeFS fs)
186 %************************************************************************
188 \subsection{Predicates and taking them apart}
190 %************************************************************************
193 occNameFS :: OccName -> EncodedFS
194 occNameFS (OccName _ s) = s
196 occNameString :: OccName -> EncodedString
197 occNameString (OccName _ s) = _UNPK_ s
199 occNameUserString :: OccName -> UserString
200 occNameUserString occ = decode (occNameString occ)
202 occNameSpace :: OccName -> NameSpace
203 occNameSpace (OccName sp _) = sp
205 setOccNameSpace :: OccName -> NameSpace -> OccName
206 setOccNameSpace (OccName _ occ) sp = OccName sp occ
208 -- occNameFlavour is used only to generate good error messages
209 occNameFlavour :: OccName -> String
210 occNameFlavour (OccName sp _) = nameSpaceString sp
214 isTvOcc, isDataSymOcc, isSymOcc :: OccName -> Bool
216 isTvOcc (OccName TvName _) = True
217 isTvOcc other = False
219 -- Data constructor operator (starts with ':', or '[]')
220 -- Pretty inefficient!
221 isDataSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
222 isDataSymOcc other = False
224 isDataOcc (OccName DataName _) = True
225 isDataOcc oter = False
227 -- Any operator (data constructor or variable)
228 -- Pretty inefficient!
229 isSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
230 isSymOcc (OccName VarName s) = isLexSym (decodeFS s)
234 %************************************************************************
236 \subsection{Making system names}
238 %************************************************************************
240 Here's our convention for splitting up the interface file name space:
242 d... dictionary identifiers
243 (local variables, so no name-clash worries)
245 $f... dict-fun identifiers (from inst decls)
246 $dm... default methods
247 $p... superclass selectors
249 $T... compiler-generated tycons for dictionaries
250 $D... ...ditto data cons
251 $sf.. specialised version of f
253 in encoded form these appear as Zdfxxx etc
255 :... keywords (export:, letrec: etc.)
257 This knowledge is encoded in the following functions.
260 @mk_deriv@ generates an @OccName@ from the one-char prefix and a string.
261 NB: The string must already be encoded!
264 mk_deriv :: NameSpace
265 -> String -- Distinguishes one sort of derived name from another
266 -> EncodedString -- Must be already encoded!! We don't want to encode it a
267 -- second time because encoding isn't itempotent
270 mk_deriv occ_sp sys_prefix str = mkSysOcc occ_sp (encode sys_prefix ++ str)
274 mkDictOcc, mkWorkerOcc, mkDefaultMethodOcc,
275 mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc
276 :: OccName -> OccName
278 -- These derived variables have a prefix that no Haskell value could have
279 mkWorkerOcc = mk_simple_deriv varName "$w"
280 mkDefaultMethodOcc = mk_simple_deriv varName "$dm"
281 mkDerivedTyConOcc = mk_simple_deriv tcName ":" -- The : prefix makes sure it classifies
282 mkClassTyConOcc = mk_simple_deriv tcName ":T" -- as a tycon/datacon
283 mkClassDataConOcc = mk_simple_deriv dataName ":D" --
284 mkDictOcc = mk_simple_deriv varName "$d"
285 mkSpecOcc = mk_simple_deriv varName "$s"
286 mkForeignExportOcc = mk_simple_deriv varName "$f"
288 mk_simple_deriv sp px occ = mk_deriv sp px (occNameString occ)
292 mkSuperDictSelOcc :: Int -- Index of superclass, eg 3
293 -> OccName -- Class, eg "Ord"
294 -> OccName -- eg "p3Ord"
295 mkSuperDictSelOcc index cls_occ
296 = mk_deriv varName "$p" (show index ++ occNameString cls_occ)
301 mkDFunOcc :: OccName -- class, eg "Ord"
302 -> OccName -- tycon (or something convenient from the instance type)
304 -> Int -- Unique to distinguish dfuns which share the previous two
306 -> OccName -- "dOrdMaybe3"
308 mkDFunOcc cls_occ tycon_occ index
309 = mk_deriv VarName "$f" (show_index ++ cls_str ++ tycon_str)
311 cls_str = occNameString cls_occ
312 tycon_str = occNameString tycon_occ
313 show_index | index == 0 = ""
314 | otherwise = show index
317 We used to add a '$m' to indicate a method, but that gives rise to bad
318 error messages from the type checker when we print the function name or pattern
319 of an instance-decl binding. Why? Because the binding is zapped
320 to use the method name in place of the selector name.
321 (See TcClassDcl.tcMethodBind)
323 The way it is now, -ddump-xx output may look confusing, but
324 you can always say -dppr-debug to get the uniques.
326 However, we *do* have to zap the first character to be lower case,
327 because overloaded constructors (blarg) generate methods too.
328 And convert to VarName space
330 e.g. a call to constructor MkFoo where
331 data (Ord a) => Foo a = MkFoo a
333 If this is necessary, we do it by prefixing '$m'. These
334 guys never show up in error messages. What a hack.
337 mkMethodOcc :: OccName -> OccName
338 mkMethodOcc occ@(OccName VarName fs) = occ
339 mkMethodOcc occ = mk_simple_deriv varName "$m" occ
343 %************************************************************************
345 \subsection{Tidying them up}
347 %************************************************************************
349 Before we print chunks of code we like to rename it so that
350 we don't have to print lots of silly uniques in it. But we mustn't
351 accidentally introduce name clashes! So the idea is that we leave the
352 OccName alone unless it accidentally clashes with one that is already
353 in scope; if so, we tack on '1' at the end and try again, then '2', and
354 so on till we find a unique one.
356 There's a wrinkle for operators. Consider '>>='. We can't use '>>=1'
357 because that isn't a single lexeme. So we encode it to 'lle' and *then*
358 tack on the '1', if necessary.
361 type TidyOccEnv = FiniteMap FAST_STRING Int -- The in-scope OccNames
362 emptyTidyOccEnv = emptyFM
364 initTidyOccEnv :: [OccName] -> TidyOccEnv -- Initialise with names to avoid!
365 initTidyOccEnv = foldl (\env (OccName _ fs) -> addToFM env fs 1) emptyTidyOccEnv
367 tidyOccName :: TidyOccEnv -> OccName -> (TidyOccEnv, OccName)
369 tidyOccName in_scope occ@(OccName occ_sp fs)
370 | not (fs `elemFM` in_scope)
371 = (addToFM in_scope fs 1, occ) -- First occurrence
373 | otherwise -- Already occurs
374 = go in_scope (_UNPK_ fs)
377 go in_scope str = case lookupFM in_scope pk_str of
378 Just n -> go (addToFM in_scope pk_str (n+1)) (str ++ show n)
379 -- Need to go round again, just in case "t3" (say)
380 -- clashes with a "t3" that's already in scope
382 Nothing -> (addToFM in_scope pk_str 1, mkSysOccFS occ_sp pk_str)
389 %************************************************************************
391 \subsection{The 'Z' encoding}
393 %************************************************************************
395 This is the main name-encoding and decoding function. It encodes any
396 string into a string that is acceptable as a C name. This is the name
397 by which things are known right through the compiler.
399 The basic encoding scheme is this.
401 * Tuples (,,,) are coded as Z3T
403 * Alphabetic characters (upper and lower), digits, and '_'
404 all translate to themselves;
405 except 'Z', which translates to 'ZZ'
406 and 'z', which translates to 'zz'
407 We need both so that we can preserve the variable/tycon distinction
409 * Most other printable characters translate to 'Zx' for some
410 alphabetic character x
412 * The others translate as 'Zxdd' where 'dd' is exactly two hexadecimal
413 digits for the ord of the character
416 --------------------------
431 -- alreadyEncoded is used in ASSERTs to check for encoded
432 -- strings. It isn't fail-safe, of course, because, say 'zh' might
433 -- be encoded or not.
434 alreadyEncoded :: String -> Bool
435 alreadyEncoded s = all ok s
438 ok ch = ISALPHANUM ch
440 alreadyEncodedFS :: FAST_STRING -> Bool
441 alreadyEncodedFS fs = alreadyEncoded (_UNPK_ fs)
443 encode :: UserString -> EncodedString
444 encode cs = case maybe_tuple cs of
445 Just n -> 'Z' : show n ++ "T" -- Tuples go to Z2T etc
449 go (c:cs) = encode_ch c ++ go cs
451 -- ToDo: Unboxed tuples too, perhaps?
452 maybe_tuple ('(' : cs) = check_tuple (0::Int) cs
453 maybe_tuple other = Nothing
455 check_tuple :: Int -> String -> Maybe Int
456 check_tuple n (',' : cs) = check_tuple (n+1) cs
457 check_tuple n ")" = Just n
458 check_tuple n other = Nothing
460 encodeFS :: UserFS -> EncodedFS
461 encodeFS fast_str | all unencodedChar str = fast_str
462 | otherwise = _PK_ (encode str)
464 str = _UNPK_ fast_str
466 unencodedChar :: Char -> Bool -- True for chars that don't need encoding
467 unencodedChar '_' = True
468 unencodedChar 'Z' = False
469 unencodedChar 'z' = False
470 unencodedChar c = ISALPHANUM c
472 encode_ch :: Char -> EncodedString
473 encode_ch c | unencodedChar c = [c] -- Common case first
476 encode_ch '(' = "ZL" -- Needed for things like (,), and (->)
477 encode_ch ')' = "ZR" -- For symmetry with (
496 encode_ch '\'' = "zq"
497 encode_ch '\\' = "zr"
502 encode_ch c = ['z', 'x', intToDigit hi, intToDigit lo]
504 (hi,lo) = ord c `quotRem` 16
507 Decode is used for user printing.
510 decodeFS :: FAST_STRING -> FAST_STRING
511 decodeFS fs = _PK_ (decode (_UNPK_ fs))
513 decode :: EncodedString -> UserString
515 decode ('Z' : rest) = decode_escape rest
516 decode ('z' : rest) = decode_escape rest
517 decode (c : rest) = c : decode rest
519 decode_escape :: EncodedString -> UserString
521 decode_escape ('Z' : rest) = 'Z' : decode rest
522 decode_escape ('C' : rest) = ':' : decode rest
523 decode_escape ('L' : rest) = '(' : decode rest
524 decode_escape ('R' : rest) = ')' : decode rest
525 decode_escape ('M' : rest) = '[' : decode rest
526 decode_escape ('N' : rest) = ']' : decode rest
528 decode_escape ('z' : rest) = 'z' : decode rest
529 decode_escape ('a' : rest) = '&' : decode rest
530 decode_escape ('b' : rest) = '|' : decode rest
531 decode_escape ('d' : rest) = '$' : decode rest
532 decode_escape ('e' : rest) = '=' : decode rest
533 decode_escape ('g' : rest) = '>' : decode rest
534 decode_escape ('h' : rest) = '#' : decode rest
535 decode_escape ('i' : rest) = '.' : decode rest
536 decode_escape ('l' : rest) = '<' : decode rest
537 decode_escape ('m' : rest) = '-' : decode rest
538 decode_escape ('n' : rest) = '!' : decode rest
539 decode_escape ('p' : rest) = '+' : decode rest
540 decode_escape ('q' : rest) = '\'' : decode rest
541 decode_escape ('r' : rest) = '\\' : decode rest
542 decode_escape ('s' : rest) = '/' : decode rest
543 decode_escape ('t' : rest) = '*' : decode rest
544 decode_escape ('u' : rest) = '^' : decode rest
545 decode_escape ('v' : rest) = '%' : decode rest
546 decode_escape ('x' : d1 : d2 : rest) = chr (digitToInt d1 * 16 + digitToInt d2) : decode rest
548 -- Tuples are coded as Z23T
549 decode_escape (c : rest)
550 | isDigit c = go (digitToInt c) rest
552 go n (c : rest) | isDigit c = go (10*n + digitToInt c) rest
553 go n ('T' : rest) = '(' : replicate n ',' ++ ')' : decode rest
554 go n other = pprPanic "decode_escape" (ppr n <+> text (c:rest))
556 decode_escape (c : rest) = pprTrace "decode_escape" (char c) (decode rest)
560 %************************************************************************
562 n\subsection{Lexical categories}
564 %************************************************************************
566 These functions test strings to see if they fit the lexical categories
567 defined in the Haskell report.
570 isLexCon, isLexVar, isLexId, isLexSym :: FAST_STRING -> Bool
571 isLexConId, isLexConSym, isLexVarId, isLexVarSym :: FAST_STRING -> Bool
573 isLexCon cs = isLexConId cs || isLexConSym cs
574 isLexVar cs = isLexVarId cs || isLexVarSym cs
576 isLexId cs = isLexConId cs || isLexVarId cs
577 isLexSym cs = isLexConSym cs || isLexVarSym cs
581 isLexConId cs -- Prefix type or data constructors
582 | _NULL_ cs = False -- e.g. "Foo", "[]", "(,)"
583 | cs == SLIT("[]") = True
584 | c == '(' = True -- (), (,), (,,), ...
585 | otherwise = isUpper c || isUpperISO c
589 isLexVarId cs -- Ordinary prefix identifiers
590 | _NULL_ cs = False -- e.g. "x", "_x"
591 | otherwise = isLower c || isLowerISO c || c == '_'
595 isLexConSym cs -- Infix type or data constructors
596 | _NULL_ cs = False -- e.g. ":-:", ":", "->"
597 | otherwise = c == ':'
602 isLexVarSym cs -- Infix identifiers
603 | _NULL_ cs = False -- e.g. "+"
604 | otherwise = isSymbolASCII c
610 isSymbolASCII c = c `elem` "!#$%&*+./<=>?@\\^|~-"
611 isSymbolISO c = ord c `elem` (0xd7 : 0xf7 : [0xa1 .. 0xbf])
612 isUpperISO (C# c#) = c# `geChar#` '\xc0'# && c# `leChar#` '\xde'# && c# `neChar#` '\xd7'#
613 --0xc0 <= oc && oc <= 0xde && oc /= 0xd7 where oc = ord c
614 isLowerISO (C# c#) = c# `geChar#` '\xdf'# && c# `leChar#` '\xff'# && c# `neChar#` '\xf7'#
615 --0xdf <= oc && oc <= 0xff && oc /= 0xf7 where oc = ord c