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, uvName, nameSpaceString,
14 OccName, -- Abstract, instance of Outputable
17 mkSrcOccFS, mkSysOcc, mkSysOccFS, mkCCallOcc, mkSrcVarOcc, mkKindOccFS,
18 mkSuperDictSelOcc, mkDFunOcc, mkForeignExportOcc,
19 mkDictOcc, mkIPOcc, mkWorkerOcc, mkMethodOcc, mkDefaultMethodOcc,
20 mkDerivedTyConOcc, mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc,
22 isSysOcc, isTvOcc, isUvOcc, isDataOcc, isDataSymOcc, isSymOcc, isIPOcc, isValOcc,
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 | IPName -- Implicit Parameters
86 | DataName -- Data constructors
87 | TvName -- Type variables
88 | UvName -- Usage 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!
106 nameSpaceString :: NameSpace -> String
107 nameSpaceString DataName = "Data constructor"
108 nameSpaceString VarName = "Variable"
109 nameSpaceString IPName = "Implicit Param"
110 nameSpaceString TvName = "Type variable"
111 nameSpaceString UvName = "Usage variable"
112 nameSpaceString TcClsName = "Type constructor or class"
116 %************************************************************************
118 \subsection[Name-pieces-datatypes]{The @OccName@ datatypes}
120 %************************************************************************
123 data OccName = OccName
130 instance Eq OccName where
131 (OccName sp1 s1) == (OccName sp2 s2) = s1 == s2 && sp1 == sp2
133 instance Ord OccName where
134 compare (OccName sp1 s1) (OccName sp2 s2) = (s1 `compare` s2) `thenCmp`
139 %************************************************************************
141 \subsection{Printing}
143 %************************************************************************
146 instance Outputable OccName where
149 pprOccName :: OccName -> SDoc
150 pprOccName (OccName sp occ) = pprEncodedFS occ
154 %************************************************************************
156 \subsection{Construction}
158 %************************************************************************
160 *Sys* things do no encoding; the caller should ensure that the thing is
164 mkSysOcc :: NameSpace -> EncodedString -> OccName
165 mkSysOcc occ_sp str = ASSERT2( alreadyEncoded str, text str )
166 OccName occ_sp (_PK_ str)
168 mkSysOccFS :: NameSpace -> EncodedFS -> OccName
169 mkSysOccFS occ_sp fs = ASSERT2( alreadyEncodedFS fs, ppr fs )
172 mkCCallOcc :: EncodedString -> OccName
173 -- This version of mkSysOcc doesn't check that the string is already encoded,
174 -- because it will be something like "{__ccall f dyn Int# -> Int#}"
175 -- This encodes a lot into something that then parses like an Id.
176 -- But then alreadyEncoded complains about the braces!
177 mkCCallOcc str = OccName varName (_PK_ str)
179 -- Kind constructors get a speical function. Uniquely, they are not encoded,
180 -- so that they have names like '*'. This means that *even in interface files*
181 -- we'll get kinds like (* -> (* -> *)). We can't use mkSysOcc because it
182 -- has an ASSERT that doesn't hold.
183 mkKindOccFS :: NameSpace -> EncodedFS -> OccName
184 mkKindOccFS occ_sp fs = OccName occ_sp fs
187 *Source-code* things are encoded.
190 mkSrcOccFS :: NameSpace -> UserFS -> OccName
191 mkSrcOccFS occ_sp fs = mkSysOccFS occ_sp (encodeFS fs)
193 mkSrcVarOcc :: UserFS -> OccName
194 mkSrcVarOcc fs = mkSysOccFS varName (encodeFS fs)
199 %************************************************************************
201 \subsection{Predicates and taking them apart}
203 %************************************************************************
206 occNameFS :: OccName -> EncodedFS
207 occNameFS (OccName _ s) = s
209 occNameString :: OccName -> EncodedString
210 occNameString (OccName _ s) = _UNPK_ s
212 occNameUserString :: OccName -> UserString
213 occNameUserString occ = decode (occNameString occ)
215 occNameSpace :: OccName -> NameSpace
216 occNameSpace (OccName sp _) = sp
218 setOccNameSpace :: OccName -> NameSpace -> OccName
219 setOccNameSpace (OccName _ occ) sp = OccName sp occ
221 -- occNameFlavour is used only to generate good error messages
222 occNameFlavour :: OccName -> String
223 occNameFlavour (OccName sp _) = nameSpaceString sp
227 isTvOcc, isDataSymOcc, isSymOcc, isUvOcc :: OccName -> Bool
229 isTvOcc (OccName TvName _) = True
230 isTvOcc other = False
232 isUvOcc (OccName UvName _) = True
233 isUvOcc other = False
235 isValOcc (OccName VarName _) = True
236 isValOcc (OccName DataName _) = True
237 isValOcc other = False
239 -- Data constructor operator (starts with ':', or '[]')
240 -- Pretty inefficient!
241 isDataSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
242 isDataSymOcc other = False
244 isDataOcc (OccName DataName _) = True
245 isDataOcc other = False
247 -- Any operator (data constructor or variable)
248 -- Pretty inefficient!
249 isSymOcc (OccName DataName s) = isLexConSym (decodeFS s)
250 isSymOcc (OccName VarName s) = isLexSym (decodeFS s)
252 isIPOcc (OccName IPName _) = True
257 %************************************************************************
259 \subsection{Making system names}
261 %************************************************************************
263 Here's our convention for splitting up the interface file name space:
265 d... dictionary identifiers
266 (local variables, so no name-clash worries)
268 $f... dict-fun identifiers (from inst decls)
269 $dm... default methods
270 $p... superclass selectors
272 $T... compiler-generated tycons for dictionaries
273 $D... ...ditto data cons
274 $sf.. specialised version of f
276 in encoded form these appear as Zdfxxx etc
278 :... keywords (export:, letrec: etc.)
280 This knowledge is encoded in the following functions.
283 @mk_deriv@ generates an @OccName@ from the one-char prefix and a string.
284 NB: The string must already be encoded!
287 mk_deriv :: NameSpace
288 -> String -- Distinguishes one sort of derived name from another
289 -> EncodedString -- Must be already encoded!! We don't want to encode it a
290 -- second time because encoding isn't itempotent
293 mk_deriv occ_sp sys_prefix str = mkSysOcc occ_sp (encode sys_prefix ++ str)
297 mkDictOcc, mkIPOcc, mkWorkerOcc, mkDefaultMethodOcc,
298 mkClassTyConOcc, mkClassDataConOcc, mkSpecOcc
299 :: OccName -> OccName
301 -- These derived variables have a prefix that no Haskell value could have
302 mkWorkerOcc = mk_simple_deriv varName "$w"
303 mkDefaultMethodOcc = mk_simple_deriv varName "$dm"
304 mkDerivedTyConOcc = mk_simple_deriv tcName ":" -- The : prefix makes sure it classifies
305 mkClassTyConOcc = mk_simple_deriv tcName ":T" -- as a tycon/datacon
306 mkClassDataConOcc = mk_simple_deriv dataName ":D" --
307 mkDictOcc = mk_simple_deriv varName "$d"
308 mkIPOcc = mk_simple_deriv varName "$i"
309 mkSpecOcc = mk_simple_deriv varName "$s"
310 mkForeignExportOcc = mk_simple_deriv varName "$f"
312 mk_simple_deriv sp px occ = mk_deriv sp px (occNameString occ)
315 isSysOcc :: OccName -> Bool -- True for all these '$' things
316 isSysOcc occ = case occNameUserString occ of
318 other -> False -- We don't care about the ':' ones
319 -- isSysOcc is only called for Ids anyway
323 mkSuperDictSelOcc :: Int -- Index of superclass, eg 3
324 -> OccName -- Class, eg "Ord"
325 -> OccName -- eg "p3Ord"
326 mkSuperDictSelOcc index cls_occ
327 = mk_deriv varName "$p" (show index ++ occNameString cls_occ)
332 mkDFunOcc :: EncodedString -- Typically the class and type glommed together e.g. "OrdMaybe"
333 -> Int -- Unique to distinguish dfuns which share the previous two
335 -- The requirement is that the (string,index) pair be unique in this module
337 -> OccName -- "$fOrdMaybe3"
339 mkDFunOcc string index
340 = mk_deriv VarName "$f" (show_index ++ string)
342 show_index | index == 0 = ""
343 | otherwise = show index
346 We used to add a '$m' to indicate a method, but that gives rise to bad
347 error messages from the type checker when we print the function name or pattern
348 of an instance-decl binding. Why? Because the binding is zapped
349 to use the method name in place of the selector name.
350 (See TcClassDcl.tcMethodBind)
352 The way it is now, -ddump-xx output may look confusing, but
353 you can always say -dppr-debug to get the uniques.
355 However, we *do* have to zap the first character to be lower case,
356 because overloaded constructors (blarg) generate methods too.
357 And convert to VarName space
359 e.g. a call to constructor MkFoo where
360 data (Ord a) => Foo a = MkFoo a
362 If this is necessary, we do it by prefixing '$m'. These
363 guys never show up in error messages. What a hack.
366 mkMethodOcc :: OccName -> OccName
367 mkMethodOcc occ@(OccName VarName fs) = occ
368 mkMethodOcc occ = mk_simple_deriv varName "$m" occ
372 %************************************************************************
374 \subsection{Tidying them up}
376 %************************************************************************
378 Before we print chunks of code we like to rename it so that
379 we don't have to print lots of silly uniques in it. But we mustn't
380 accidentally introduce name clashes! So the idea is that we leave the
381 OccName alone unless it accidentally clashes with one that is already
382 in scope; if so, we tack on '1' at the end and try again, then '2', and
383 so on till we find a unique one.
385 There's a wrinkle for operators. Consider '>>='. We can't use '>>=1'
386 because that isn't a single lexeme. So we encode it to 'lle' and *then*
387 tack on the '1', if necessary.
390 type TidyOccEnv = FiniteMap FAST_STRING Int -- The in-scope OccNames
391 emptyTidyOccEnv = emptyFM
393 initTidyOccEnv :: [OccName] -> TidyOccEnv -- Initialise with names to avoid!
394 initTidyOccEnv = foldl (\env (OccName _ fs) -> addToFM env fs 1) emptyTidyOccEnv
396 tidyOccName :: TidyOccEnv -> OccName -> (TidyOccEnv, OccName)
398 tidyOccName in_scope occ@(OccName occ_sp fs)
399 | not (fs `elemFM` in_scope)
400 = (addToFM in_scope fs 1, occ) -- First occurrence
402 | otherwise -- Already occurs
403 = go in_scope (_UNPK_ fs)
406 go in_scope str = case lookupFM in_scope pk_str of
407 Just n -> go (addToFM in_scope pk_str (n+1)) (str ++ show n)
408 -- Need to go round again, just in case "t3" (say)
409 -- clashes with a "t3" that's already in scope
411 Nothing -> (addToFM in_scope pk_str 1, mkSysOccFS occ_sp pk_str)
418 %************************************************************************
420 \subsection{The 'Z' encoding}
422 %************************************************************************
424 This is the main name-encoding and decoding function. It encodes any
425 string into a string that is acceptable as a C name. This is the name
426 by which things are known right through the compiler.
428 The basic encoding scheme is this.
430 * Tuples (,,,) are coded as Z3T
432 * Alphabetic characters (upper and lower) and digits
433 all translate to themselves;
434 except 'Z', which translates to 'ZZ'
435 and 'z', which translates to 'zz'
436 We need both so that we can preserve the variable/tycon distinction
438 * Most other printable characters translate to 'zx' or 'Zx' for some
439 alphabetic character x
441 * The others translate as 'zxdd' where 'dd' is exactly two hexadecimal
442 digits for the ord of the character
445 --------------------------
460 -- alreadyEncoded is used in ASSERTs to check for encoded
461 -- strings. It isn't fail-safe, of course, because, say 'zh' might
462 -- be encoded or not.
463 alreadyEncoded :: String -> Bool
464 alreadyEncoded s = all ok s
466 ok ' ' = True -- This is a bit of a lie; if we really wanted spaces
467 -- in names we'd have to encode them. But we do put
468 -- spaces in ccall "occurrences", and we don't want to
470 ok ch = ISALPHANUM ch
472 alreadyEncodedFS :: FAST_STRING -> Bool
473 alreadyEncodedFS fs = alreadyEncoded (_UNPK_ fs)
475 encode :: UserString -> EncodedString
476 encode cs = case maybe_tuple cs of
477 Just n -> 'Z' : show n ++ "T" -- Tuples go to Z2T etc
481 go (c:cs) = encode_ch c ++ go cs
483 -- ToDo: Unboxed tuples too, perhaps?
484 maybe_tuple ('(' : cs) = check_tuple (0::Int) cs
485 maybe_tuple other = Nothing
487 check_tuple :: Int -> String -> Maybe Int
488 check_tuple n (',' : cs) = check_tuple (n+1) cs
489 check_tuple n ")" = Just n
490 check_tuple n other = Nothing
492 encodeFS :: UserFS -> EncodedFS
493 encodeFS fast_str | all unencodedChar str = fast_str
494 | otherwise = _PK_ (encode str)
496 str = _UNPK_ fast_str
498 unencodedChar :: Char -> Bool -- True for chars that don't need encoding
499 unencodedChar 'Z' = False
500 unencodedChar 'z' = False
501 unencodedChar c = ISALPHANUM c
503 encode_ch :: Char -> EncodedString
504 encode_ch c | unencodedChar c = [c] -- Common case first
507 encode_ch '(' = "ZL" -- Needed for things like (,), and (->)
508 encode_ch ')' = "ZR" -- For symmetry with (
528 encode_ch '\'' = "zq"
529 encode_ch '\\' = "zr"
534 encode_ch c = ['z', 'x', intToDigit hi, intToDigit lo]
536 (hi,lo) = ord c `quotRem` 16
539 Decode is used for user printing.
542 decodeFS :: FAST_STRING -> FAST_STRING
543 decodeFS fs = _PK_ (decode (_UNPK_ fs))
545 decode :: EncodedString -> UserString
547 decode ('Z' : rest) = decode_escape rest
548 decode ('z' : rest) = decode_escape rest
549 decode (c : rest) = c : decode rest
551 decode_escape :: EncodedString -> UserString
553 decode_escape ('L' : rest) = '(' : decode rest
554 decode_escape ('R' : rest) = ')' : decode rest
555 decode_escape ('M' : rest) = '[' : decode rest
556 decode_escape ('N' : rest) = ']' : decode rest
557 decode_escape ('C' : rest) = ':' : decode rest
558 decode_escape ('Z' : rest) = 'Z' : decode rest
560 decode_escape ('z' : rest) = 'z' : decode rest
561 decode_escape ('a' : rest) = '&' : decode rest
562 decode_escape ('b' : rest) = '|' : decode rest
563 decode_escape ('c' : rest) = '^' : decode rest
564 decode_escape ('d' : rest) = '$' : decode rest
565 decode_escape ('e' : rest) = '=' : decode rest
566 decode_escape ('g' : rest) = '>' : decode rest
567 decode_escape ('h' : rest) = '#' : decode rest
568 decode_escape ('i' : rest) = '.' : decode rest
569 decode_escape ('l' : rest) = '<' : decode rest
570 decode_escape ('m' : rest) = '-' : decode rest
571 decode_escape ('n' : rest) = '!' : decode rest
572 decode_escape ('p' : rest) = '+' : decode rest
573 decode_escape ('q' : rest) = '\'' : decode rest
574 decode_escape ('r' : rest) = '\\' : decode rest
575 decode_escape ('s' : rest) = '/' : decode rest
576 decode_escape ('t' : rest) = '*' : decode rest
577 decode_escape ('u' : rest) = '_' : decode rest
578 decode_escape ('v' : rest) = '%' : decode rest
579 decode_escape ('x' : d1 : d2 : rest) = chr (digitToInt d1 * 16 + digitToInt d2) : decode rest
581 -- Tuples are coded as Z23T
582 decode_escape (c : rest)
583 | isDigit c = go (digitToInt c) rest
585 go n (c : rest) | isDigit c = go (10*n + digitToInt c) rest
586 go n ('T' : rest) = '(' : replicate n ',' ++ ')' : decode rest
587 go n other = pprPanic "decode_escape" (ppr n <+> text (c:rest))
589 decode_escape (c : rest) = pprTrace "decode_escape" (char c) (decode rest)
593 %************************************************************************
595 n\subsection{Lexical categories}
597 %************************************************************************
599 These functions test strings to see if they fit the lexical categories
600 defined in the Haskell report.
603 isLexCon, isLexVar, isLexId, isLexSym :: FAST_STRING -> Bool
604 isLexConId, isLexConSym, isLexVarId, isLexVarSym :: FAST_STRING -> Bool
606 isLexCon cs = isLexConId cs || isLexConSym cs
607 isLexVar cs = isLexVarId cs || isLexVarSym cs
609 isLexId cs = isLexConId cs || isLexVarId cs
610 isLexSym cs = isLexConSym cs || isLexVarSym cs
614 isLexConId cs -- Prefix type or data constructors
615 | _NULL_ cs = False -- e.g. "Foo", "[]", "(,)"
616 | cs == SLIT("[]") = True
617 | c == '(' = True -- (), (,), (,,), ...
618 | otherwise = isUpper c || isUpperISO c
622 isLexVarId cs -- Ordinary prefix identifiers
623 | _NULL_ cs = False -- e.g. "x", "_x"
624 | otherwise = isLower c || isLowerISO c || c == '_'
628 isLexConSym cs -- Infix type or data constructors
629 | _NULL_ cs = False -- e.g. ":-:", ":", "->"
630 | otherwise = c == ':'
635 isLexVarSym cs -- Infix identifiers
636 | _NULL_ cs = False -- e.g. "+"
637 | otherwise = isSymbolASCII c
643 isSymbolASCII c = c `elem` "!#$%&*+./<=>?@\\^|~-"
644 isSymbolISO c = ord c `elem` (0xd7 : 0xf7 : [0xa1 .. 0xbf])
645 isUpperISO (C# c#) = c# `geChar#` '\xc0'# && c# `leChar#` '\xde'# && c# `neChar#` '\xd7'#
646 --0xc0 <= oc && oc <= 0xde && oc /= 0xd7 where oc = ord c
647 isLowerISO (C# c#) = c# `geChar#` '\xdf'# && c# `leChar#` '\xff'# && c# `neChar#` '\xf7'#
648 --0xdf <= oc && oc <= 0xff && oc /= 0xf7 where oc = ord c