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,
11 tvName, nameSpaceString,
14 OccName, -- Abstract, instance of Outputable
17 mkOccFS, mkSysOcc, mkSysOccFS, mkFCallOcc, mkVarOcc, mkKindOccFS,
18 mkSuperDictSelOcc, mkDFunOcc, mkForeignExportOcc,
19 mkDictOcc, mkIPOcc, mkWorkerOcc, mkMethodOcc, mkDefaultMethodOcc,
20 mkDerivedTyConOcc, mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc,
23 isTvOcc, isDataOcc, isDataSymOcc, isSymOcc, 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 | DataName -- Data constructors
87 | TvName -- Type variables
88 | TcClsName -- Type constructors and classes; Haskell has them
89 -- in the same name space for now.
92 -- Though type constructors and classes are in the same name space now,
93 -- the NameSpace type is abstract, so we can easily separate them later
94 tcName = TcClsName -- Type constructors
95 clsName = TcClsName -- Classes
96 tcClsName = TcClsName -- Not sure which!
103 nameSpaceString :: NameSpace -> String
104 nameSpaceString DataName = "Data constructor"
105 nameSpaceString VarName = "Variable"
106 nameSpaceString TvName = "Type variable"
107 nameSpaceString TcClsName = "Type constructor or class"
111 %************************************************************************
113 \subsection[Name-pieces-datatypes]{The @OccName@ datatypes}
115 %************************************************************************
118 data OccName = OccName
125 instance Eq OccName where
126 (OccName sp1 s1) == (OccName sp2 s2) = s1 == s2 && sp1 == sp2
128 instance Ord OccName where
129 compare (OccName sp1 s1) (OccName sp2 s2) = (s1 `compare` s2) `thenCmp`
134 %************************************************************************
136 \subsection{Printing}
138 %************************************************************************
141 instance Outputable OccName where
144 pprOccName :: OccName -> SDoc
145 pprOccName (OccName sp occ) = pprEncodedFS occ
149 %************************************************************************
151 \subsection{Construction}
153 %************************************************************************
155 *Sys* things do no encoding; the caller should ensure that the thing is
159 mkSysOcc :: NameSpace -> EncodedString -> OccName
160 mkSysOcc occ_sp str = ASSERT2( alreadyEncoded str, text str )
161 OccName occ_sp (_PK_ str)
163 mkSysOccFS :: NameSpace -> EncodedFS -> OccName
164 mkSysOccFS occ_sp fs = ASSERT2( alreadyEncodedFS fs, ppr fs )
167 mkFCallOcc :: EncodedString -> OccName
168 -- This version of mkSysOcc doesn't check that the string is already encoded,
169 -- because it will be something like "{__ccall f dyn Int# -> Int#}"
170 -- This encodes a lot into something that then parses like an Id.
171 -- But then alreadyEncoded complains about the braces!
172 mkFCallOcc str = OccName varName (_PK_ str)
174 -- Kind constructors get a special function. Uniquely, they are not encoded,
175 -- so that they have names like '*'. This means that *even in interface files*
176 -- we'll get kinds like (* -> (* -> *)). We can't use mkSysOcc because it
177 -- has an ASSERT that doesn't hold.
178 mkKindOccFS :: NameSpace -> EncodedFS -> OccName
179 mkKindOccFS occ_sp fs = OccName occ_sp fs
182 *Source-code* things are encoded.
185 mkOccFS :: NameSpace -> UserFS -> OccName
186 mkOccFS occ_sp fs = mkSysOccFS occ_sp (encodeFS fs)
188 mkVarOcc :: UserFS -> OccName
189 mkVarOcc fs = mkSysOccFS varName (encodeFS fs)
194 %************************************************************************
196 \subsection{Predicates and taking them apart}
198 %************************************************************************
201 occNameFS :: OccName -> EncodedFS
202 occNameFS (OccName _ s) = s
204 occNameString :: OccName -> EncodedString
205 occNameString (OccName _ s) = _UNPK_ s
207 occNameUserString :: OccName -> UserString
208 occNameUserString occ = decode (occNameString occ)
210 occNameSpace :: OccName -> NameSpace
211 occNameSpace (OccName sp _) = sp
213 setOccNameSpace :: OccName -> NameSpace -> OccName
214 setOccNameSpace (OccName _ occ) sp = OccName sp occ
216 -- occNameFlavour is used only to generate good error messages
217 occNameFlavour :: OccName -> String
218 occNameFlavour (OccName sp _) = nameSpaceString sp
222 isTvOcc, isDataSymOcc, isSymOcc :: OccName -> Bool
224 isTvOcc (OccName TvName _) = True
225 isTvOcc other = False
227 isValOcc (OccName VarName _) = True
228 isValOcc (OccName DataName _) = True
229 isValOcc other = False
231 -- Data constructor operator (starts with ':', or '[]')
232 -- Pretty inefficient!
233 isDataSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
234 isDataSymOcc other = False
236 isDataOcc (OccName DataName _) = True
237 isDataOcc other = False
239 -- Any operator (data constructor or variable)
240 -- Pretty inefficient!
241 isSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
242 isSymOcc (OccName VarName s) = isLexSym (decodeFS s)
246 %************************************************************************
248 \subsection{Making system names}
250 %************************************************************************
252 Here's our convention for splitting up the interface file name space:
254 d... dictionary identifiers
255 (local variables, so no name-clash worries)
257 $f... dict-fun identifiers (from inst decls)
258 $dm... default methods
259 $p... superclass selectors
261 $T... compiler-generated tycons for dictionaries
262 $D... ...ditto data cons
263 $sf.. specialised version of f
265 in encoded form these appear as Zdfxxx etc
267 :... keywords (export:, letrec: etc.)
269 This knowledge is encoded in the following functions.
272 @mk_deriv@ generates an @OccName@ from the one-char prefix and a string.
273 NB: The string must already be encoded!
276 mk_deriv :: NameSpace
277 -> String -- Distinguishes one sort of derived name from another
278 -> EncodedString -- Must be already encoded!! We don't want to encode it a
279 -- second time because encoding isn't itempotent
282 mk_deriv occ_sp sys_prefix str = mkSysOcc occ_sp (encode sys_prefix ++ str)
286 mkDictOcc, mkIPOcc, mkWorkerOcc, mkDefaultMethodOcc,
287 mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc
288 :: OccName -> OccName
290 -- These derived variables have a prefix that no Haskell value could have
291 mkWorkerOcc = mk_simple_deriv varName "$w"
292 mkDefaultMethodOcc = mk_simple_deriv varName "$dm"
293 mkDerivedTyConOcc = mk_simple_deriv tcName ":" -- The : prefix makes sure it classifies
294 mkClassTyConOcc = mk_simple_deriv tcName ":T" -- as a tycon/datacon
295 mkClassDataConOcc = mk_simple_deriv dataName ":D" --
296 mkDictOcc = mk_simple_deriv varName "$d"
297 mkIPOcc = mk_simple_deriv varName "$i"
298 mkSpecOcc = mk_simple_deriv varName "$s"
299 mkForeignExportOcc = mk_simple_deriv varName "$f"
300 mkGenOcc1 = mk_simple_deriv varName "$gfrom" -- Generics
301 mkGenOcc2 = mk_simple_deriv varName "$gto" -- Generics
302 mk_simple_deriv sp px occ = mk_deriv sp px (occNameString occ)
306 mkSuperDictSelOcc :: Int -- Index of superclass, eg 3
307 -> OccName -- Class, eg "Ord"
308 -> OccName -- eg "p3Ord"
309 mkSuperDictSelOcc index cls_occ
310 = mk_deriv varName "$p" (show index ++ occNameString cls_occ)
315 mkDFunOcc :: EncodedString -- Typically the class and type glommed together e.g. "OrdMaybe"
316 -> OccName -- "$fOrdMaybe"
318 mkDFunOcc string = mk_deriv VarName "$f" string
321 We used to add a '$m' to indicate a method, but that gives rise to bad
322 error messages from the type checker when we print the function name or pattern
323 of an instance-decl binding. Why? Because the binding is zapped
324 to use the method name in place of the selector name.
325 (See TcClassDcl.tcMethodBind)
327 The way it is now, -ddump-xx output may look confusing, but
328 you can always say -dppr-debug to get the uniques.
330 However, we *do* have to zap the first character to be lower case,
331 because overloaded constructors (blarg) generate methods too.
332 And convert to VarName space
334 e.g. a call to constructor MkFoo where
335 data (Ord a) => Foo a = MkFoo a
337 If this is necessary, we do it by prefixing '$m'. These
338 guys never show up in error messages. What a hack.
341 mkMethodOcc :: OccName -> OccName
342 mkMethodOcc occ@(OccName VarName fs) = occ
343 mkMethodOcc occ = mk_simple_deriv varName "$m" occ
347 %************************************************************************
349 \subsection{Tidying them up}
351 %************************************************************************
353 Before we print chunks of code we like to rename it so that
354 we don't have to print lots of silly uniques in it. But we mustn't
355 accidentally introduce name clashes! So the idea is that we leave the
356 OccName alone unless it accidentally clashes with one that is already
357 in scope; if so, we tack on '1' at the end and try again, then '2', and
358 so on till we find a unique one.
360 There's a wrinkle for operators. Consider '>>='. We can't use '>>=1'
361 because that isn't a single lexeme. So we encode it to 'lle' and *then*
362 tack on the '1', if necessary.
365 type TidyOccEnv = FiniteMap FAST_STRING Int -- The in-scope OccNames
366 emptyTidyOccEnv = emptyFM
368 initTidyOccEnv :: [OccName] -> TidyOccEnv -- Initialise with names to avoid!
369 initTidyOccEnv = foldl (\env (OccName _ fs) -> addToFM env fs 1) emptyTidyOccEnv
371 tidyOccName :: TidyOccEnv -> OccName -> (TidyOccEnv, OccName)
373 tidyOccName in_scope occ@(OccName occ_sp fs)
374 | not (fs `elemFM` in_scope)
375 = (addToFM in_scope fs 1, occ) -- First occurrence
377 | otherwise -- Already occurs
378 = go in_scope (_UNPK_ fs)
381 go in_scope str = case lookupFM in_scope pk_str of
382 Just n -> go (addToFM in_scope pk_str (n+1)) (str ++ show n)
383 -- Need to go round again, just in case "t3" (say)
384 -- clashes with a "t3" that's already in scope
386 Nothing -> (addToFM in_scope pk_str 1, mkSysOccFS occ_sp pk_str)
393 %************************************************************************
395 \subsection{The 'Z' encoding}
397 %************************************************************************
399 This is the main name-encoding and decoding function. It encodes any
400 string into a string that is acceptable as a C name. This is the name
401 by which things are known right through the compiler.
403 The basic encoding scheme is this.
405 * Tuples (,,,) are coded as Z3T
407 * Alphabetic characters (upper and lower) and digits
408 all translate to themselves;
409 except 'Z', which translates to 'ZZ'
410 and 'z', which translates to 'zz'
411 We need both so that we can preserve the variable/tycon distinction
413 * Most other printable characters translate to 'zx' or 'Zx' for some
414 alphabetic character x
416 * The others translate as 'znnnU' where 'nnn' is the decimal number
420 --------------------------
432 (#,,,,#) Z4H unboxed 5-tuple
433 (NB: the number is one different to the number of
434 elements. No real reason except that () is a zero-tuple,
435 while (,) is a 2-tuple.)
438 -- alreadyEncoded is used in ASSERTs to check for encoded
439 -- strings. It isn't fail-safe, of course, because, say 'zh' might
440 -- be encoded or not.
441 alreadyEncoded :: String -> Bool
442 alreadyEncoded s = all ok s
445 -- This is a bit of a lie; if we really wanted spaces
446 -- in names we'd have to encode them. But we do put
447 -- spaces in ccall "occurrences", and we don't want to
449 ok ch = ISALPHANUM ch
451 alreadyEncodedFS :: FAST_STRING -> Bool
452 alreadyEncodedFS fs = alreadyEncoded (_UNPK_ fs)
454 encode :: UserString -> EncodedString
455 encode cs = case maybe_tuple cs of
456 Just n -> n -- Tuples go to Z2T etc
460 go (c:cs) = encode_ch c ++ go cs
462 maybe_tuple ('(' : '#' : cs) = case count_commas (0::Int) cs of
463 (n, '#' : ')' : cs) -> Just ('Z' : shows n "H")
465 maybe_tuple ('(' : cs) = case count_commas (0::Int) cs of
466 (n, ')' : cs) -> Just ('Z' : shows n "T")
468 maybe_tuple other = Nothing
470 count_commas :: Int -> String -> (Int, String)
471 count_commas n (',' : cs) = count_commas (n+1) cs
472 count_commas n cs = (n,cs)
474 encodeFS :: UserFS -> EncodedFS
475 encodeFS fast_str | all unencodedChar str = fast_str
476 | otherwise = _PK_ (encode str)
478 str = _UNPK_ fast_str
480 unencodedChar :: Char -> Bool -- True for chars that don't need encoding
481 unencodedChar 'Z' = False
482 unencodedChar 'z' = False
483 unencodedChar c = ISALPHANUM c
485 encode_ch :: Char -> EncodedString
486 encode_ch c | unencodedChar c = [c] -- Common case first
489 encode_ch '(' = "ZL" -- Needed for things like (,), and (->)
490 encode_ch ')' = "ZR" -- For symmetry with (
510 encode_ch '\'' = "zq"
511 encode_ch '\\' = "zr"
516 encode_ch c = 'z' : shows (ord c) "U"
519 Decode is used for user printing.
522 decodeFS :: FAST_STRING -> FAST_STRING
523 decodeFS fs = _PK_ (decode (_UNPK_ fs))
525 decode :: EncodedString -> UserString
527 decode ('Z' : rest) = decode_escape rest
528 decode ('z' : rest) = decode_escape rest
529 decode (c : rest) = c : decode rest
531 decode_escape :: EncodedString -> UserString
533 decode_escape ('L' : rest) = '(' : decode rest
534 decode_escape ('R' : rest) = ')' : decode rest
535 decode_escape ('M' : rest) = '[' : decode rest
536 decode_escape ('N' : rest) = ']' : decode rest
537 decode_escape ('C' : rest) = ':' : decode rest
538 decode_escape ('Z' : rest) = 'Z' : decode rest
540 decode_escape ('z' : rest) = 'z' : decode rest
541 decode_escape ('a' : rest) = '&' : decode rest
542 decode_escape ('b' : rest) = '|' : decode rest
543 decode_escape ('c' : rest) = '^' : decode rest
544 decode_escape ('d' : rest) = '$' : decode rest
545 decode_escape ('e' : rest) = '=' : decode rest
546 decode_escape ('g' : rest) = '>' : decode rest
547 decode_escape ('h' : rest) = '#' : decode rest
548 decode_escape ('i' : rest) = '.' : decode rest
549 decode_escape ('l' : rest) = '<' : decode rest
550 decode_escape ('m' : rest) = '-' : decode rest
551 decode_escape ('n' : rest) = '!' : decode rest
552 decode_escape ('p' : rest) = '+' : decode rest
553 decode_escape ('q' : rest) = '\'' : decode rest
554 decode_escape ('r' : rest) = '\\' : decode rest
555 decode_escape ('s' : rest) = '/' : decode rest
556 decode_escape ('t' : rest) = '*' : decode rest
557 decode_escape ('u' : rest) = '_' : decode rest
558 decode_escape ('v' : rest) = '%' : decode rest
560 -- Tuples are coded as Z23T
561 -- Characters not having a specific code are coded as z224U
562 decode_escape (c : rest)
563 | isDigit c = go (digitToInt c) rest
565 go n (c : rest) | isDigit c = go (10*n + digitToInt c) rest
566 go n ('T' : rest) = '(' : replicate n ',' ++ ')' : decode rest
567 go n ('H' : rest) = '(' : '#' : replicate n ',' ++ '#' : ')' : decode rest
568 go n ('U' : rest) = chr n : decode rest
569 go n other = pprPanic "decode_escape" (ppr n <+> text (c:rest))
571 decode_escape (c : rest) = pprTrace "decode_escape" (char c) (decode rest)
575 %************************************************************************
577 n\subsection{Lexical categories}
579 %************************************************************************
581 These functions test strings to see if they fit the lexical categories
582 defined in the Haskell report.
585 isLexCon, isLexVar, isLexId, isLexSym :: FAST_STRING -> Bool
586 isLexConId, isLexConSym, isLexVarId, isLexVarSym :: FAST_STRING -> Bool
588 isLexCon cs = isLexConId cs || isLexConSym cs
589 isLexVar cs = isLexVarId cs || isLexVarSym cs
591 isLexId cs = isLexConId cs || isLexVarId cs
592 isLexSym cs = isLexConSym cs || isLexVarSym cs
596 isLexConId cs -- Prefix type or data constructors
597 | _NULL_ cs = False -- e.g. "Foo", "[]", "(,)"
598 | cs == SLIT("[]") = True
599 | otherwise = startsConId (_HEAD_ cs)
601 isLexVarId cs -- Ordinary prefix identifiers
602 | _NULL_ cs = False -- e.g. "x", "_x"
603 | otherwise = startsVarId (_HEAD_ cs)
605 isLexConSym cs -- Infix type or data constructors
606 | _NULL_ cs = False -- e.g. ":-:", ":", "->"
607 | cs == SLIT("->") = True
608 | otherwise = startsConSym (_HEAD_ cs)
610 isLexVarSym cs -- Infix identifiers
611 | _NULL_ cs = False -- e.g. "+"
612 | otherwise = startsVarSym (_HEAD_ cs)
615 startsVarSym, startsVarId, startsConSym, startsConId :: Char -> Bool
616 startsVarSym c = isSymbolASCII c || isSymbolISO c -- Infix Ids
617 startsConSym c = c == ':' -- Infix data constructors
618 startsVarId c = isLower c || isLowerISO c || c == '_' -- Ordinary Ids
619 startsConId c = isUpper c || isUpperISO c || c == '(' -- Ordinary type constructors and data constructors
622 isSymbolASCII c = c `elem` "!#$%&*+./<=>?@\\^|~-"
623 isSymbolISO c = ord c `elem` (0xd7 : 0xf7 : [0xa1 .. 0xbf])
624 isUpperISO (C# c#) = c# `geChar#` '\xc0'# && c# `leChar#` '\xde'# && c# `neChar#` '\xd7'#
625 --0xc0 <= oc && oc <= 0xde && oc /= 0xd7 where oc = ord c
626 isLowerISO (C# c#) = c# `geChar#` '\xdf'# && c# `leChar#` '\xff'# && c# `neChar#` '\xf7'#
627 --0xdf <= oc && oc <= 0xff && oc /= 0xf7 where oc = ord c