2 % (c) The University of Glasgow 2006
3 % (c) The AQUA Project, Glasgow University, 1998
6 This module contains definitions for the IdInfo for things that
7 have a standard form, namely:
11 - method and superclass selectors
12 - primitive operations
16 mkDictFunId, mkDefaultMethodId,
20 mkPrimOpId, mkFCallId, mkTickBoxOpId, mkBreakPointOpId,
22 mkReboxingAlt, wrapNewTypeBody, unwrapNewTypeBody,
23 wrapFamInstBody, unwrapFamInstScrut,
24 mkUnpackCase, mkProductBox,
26 -- And some particular Ids; see below for why they are wired in
27 wiredInIds, ghcPrimIds,
28 unsafeCoerceName, unsafeCoerceId, realWorldPrimId,
29 voidArgId, nullAddrId, seqId, lazyId, lazyIdKey,
31 -- Re-export error Ids
35 #include "HsVersions.h"
44 import CoreUtils ( exprType, mkCoerce )
55 import Var ( Var, TyVar, mkCoVar, mkExportedLocalVar )
61 import BasicTypes hiding ( SuccessFlag(..) )
69 %************************************************************************
71 \subsection{Wired in Ids}
73 %************************************************************************
77 There are several reasons why an Id might appear in the wiredInIds:
79 (1) The ghcPrimIds are wired in because they can't be defined in
80 Haskell at all, although the can be defined in Core. They have
81 compulsory unfoldings, so they are always inlined and they have
82 no definition site. Their home module is GHC.Prim, so they
83 also have a description in primops.txt.pp, where they are called
86 (2) The 'error' function, eRROR_ID, is wired in because we don't yet have
87 a way to express in an interface file that the result type variable
88 is 'open'; that is can be unified with an unboxed type
90 [The interface file format now carry such information, but there's
91 no way yet of expressing at the definition site for these
92 error-reporting functions that they have an 'open'
93 result type. -- sof 1/99]
95 (3) Other error functions (rUNTIME_ERROR_ID) are wired in (a) because
96 the desugarer generates code that mentiones them directly, and
97 (b) for the same reason as eRROR_ID
99 (4) lazyId is wired in because the wired-in version overrides the
100 strictness of the version defined in GHC.Base
102 In cases (2-4), the function has a definition in a library module, and
103 can be called; but the wired-in version means that the details are
104 never read from that module's interface file; instead, the full definition
112 eRROR_ID, -- This one isn't used anywhere else in the compiler
113 -- But we still need it in wiredInIds so that when GHC
114 -- compiles a program that mentions 'error' we don't
115 -- import its type from the interface file; we just get
116 -- the Id defined here. Which has an 'open-tyvar' type.
119 iRREFUT_PAT_ERROR_ID,
120 nON_EXHAUSTIVE_GUARDS_ERROR_ID,
121 nO_METHOD_BINDING_ERROR_ID,
129 -- These Ids are exported from GHC.Prim
132 = [ -- These can't be defined in Haskell, but they have
133 -- perfectly reasonable unfoldings in Core
141 %************************************************************************
143 \subsection{Data constructors}
145 %************************************************************************
147 The wrapper for a constructor is an ordinary top-level binding that evaluates
148 any strict args, unboxes any args that are going to be flattened, and calls
151 We're going to build a constructor that looks like:
153 data (Data a, C b) => T a b = T1 !a !Int b
156 \d1::Data a, d2::C b ->
157 \p q r -> case p of { p ->
159 Con T1 [a,b] [p,q,r]}}
163 * d2 is thrown away --- a context in a data decl is used to make sure
164 one *could* construct dictionaries at the site the constructor
165 is used, but the dictionary isn't actually used.
167 * We have to check that we can construct Data dictionaries for
168 the types a and Int. Once we've done that we can throw d1 away too.
170 * We use (case p of q -> ...) to evaluate p, rather than "seq" because
171 all that matters is that the arguments are evaluated. "seq" is
172 very careful to preserve evaluation order, which we don't need
175 You might think that we could simply give constructors some strictness
176 info, like PrimOps, and let CoreToStg do the let-to-case transformation.
177 But we don't do that because in the case of primops and functions strictness
178 is a *property* not a *requirement*. In the case of constructors we need to
179 do something active to evaluate the argument.
181 Making an explicit case expression allows the simplifier to eliminate
182 it in the (common) case where the constructor arg is already evaluated.
184 Note [Wrappers for data instance tycons]
185 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
186 In the case of data instances, the wrapper also applies the coercion turning
187 the representation type into the family instance type to cast the result of
188 the wrapper. For example, consider the declarations
190 data family Map k :: * -> *
191 data instance Map (a, b) v = MapPair (Map a (Pair b v))
193 The tycon to which the datacon MapPair belongs gets a unique internal
194 name of the form :R123Map, and we call it the representation tycon.
195 In contrast, Map is the family tycon (accessible via
196 tyConFamInst_maybe). A coercion allows you to move between
197 representation and family type. It is accessible from :R123Map via
198 tyConFamilyCoercion_maybe and has kind
200 Co123Map a b v :: {Map (a, b) v ~ :R123Map a b v}
202 The wrapper and worker of MapPair get the types
205 $WMapPair :: forall a b v. Map a (Map a b v) -> Map (a, b) v
206 $WMapPair a b v = MapPair a b v `cast` sym (Co123Map a b v)
209 MapPair :: forall a b v. Map a (Map a b v) -> :R123Map a b v
211 This coercion is conditionally applied by wrapFamInstBody.
213 It's a bit more complicated if the data instance is a GADT as well!
215 data instance T [a] where
216 T1 :: forall b. b -> T [Maybe b]
218 Hence we translate to
221 $WT1 :: forall b. b -> T [Maybe b]
222 $WT1 b v = T1 (Maybe b) b (Maybe b) v
223 `cast` sym (Co7T (Maybe b))
226 T1 :: forall c b. (c ~ Maybe b) => b -> :R7T c
228 -- Coercion from family type to representation type
229 Co7T a :: T [a] ~ :R7T a
232 mkDataConIds :: Name -> Name -> DataCon -> DataConIds
233 mkDataConIds wrap_name wkr_name data_con
234 | isNewTyCon tycon -- Newtype, only has a worker
235 = DCIds Nothing nt_work_id
237 | any isBanged all_strict_marks -- Algebraic, needs wrapper
238 || not (null eq_spec) -- NB: LoadIface.ifaceDeclSubBndrs
239 || isFamInstTyCon tycon -- depends on this test
240 = DCIds (Just alg_wrap_id) wrk_id
242 | otherwise -- Algebraic, no wrapper
243 = DCIds Nothing wrk_id
245 (univ_tvs, ex_tvs, eq_spec,
246 eq_theta, dict_theta, orig_arg_tys, res_ty) = dataConFullSig data_con
247 tycon = dataConTyCon data_con -- The representation TyCon (not family)
249 ----------- Worker (algebraic data types only) --------------
250 -- The *worker* for the data constructor is the function that
251 -- takes the representation arguments and builds the constructor.
252 wrk_id = mkGlobalId (DataConWorkId data_con) wkr_name
253 (dataConRepType data_con) wkr_info
255 wkr_arity = dataConRepArity data_con
256 wkr_info = noCafIdInfo
257 `setArityInfo` wkr_arity
258 `setStrictnessInfo` Just wkr_sig
259 `setUnfoldingInfo` evaldUnfolding -- Record that it's evaluated,
262 wkr_sig = mkStrictSig (mkTopDmdType (replicate wkr_arity topDmd) cpr_info)
263 -- Note [Data-con worker strictness]
264 -- Notice that we do *not* say the worker is strict
265 -- even if the data constructor is declared strict
266 -- e.g. data T = MkT !(Int,Int)
267 -- Why? Because the *wrapper* is strict (and its unfolding has case
268 -- expresssions that do the evals) but the *worker* itself is not.
269 -- If we pretend it is strict then when we see
270 -- case x of y -> $wMkT y
271 -- the simplifier thinks that y is "sure to be evaluated" (because
272 -- $wMkT is strict) and drops the case. No, $wMkT is not strict.
274 -- When the simplifer sees a pattern
275 -- case e of MkT x -> ...
276 -- it uses the dataConRepStrictness of MkT to mark x as evaluated;
277 -- but that's fine... dataConRepStrictness comes from the data con
278 -- not from the worker Id.
280 cpr_info | isProductTyCon tycon &&
283 wkr_arity <= mAX_CPR_SIZE = retCPR
285 -- RetCPR is only true for products that are real data types;
286 -- that is, not unboxed tuples or [non-recursive] newtypes
288 ----------- Workers for newtypes --------------
289 nt_work_id = mkGlobalId (DataConWrapId data_con) wkr_name wrap_ty nt_work_info
290 nt_work_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
291 `setArityInfo` 1 -- Arity 1
292 `setUnfoldingInfo` newtype_unf
293 id_arg1 = mkTemplateLocal 1 (head orig_arg_tys)
294 newtype_unf = ASSERT2( isVanillaDataCon data_con &&
295 isSingleton orig_arg_tys, ppr data_con )
296 -- Note [Newtype datacons]
297 mkCompulsoryUnfolding $
298 mkLams wrap_tvs $ Lam id_arg1 $
299 wrapNewTypeBody tycon res_ty_args (Var id_arg1)
302 ----------- Wrapper --------------
303 -- We used to include the stupid theta in the wrapper's args
304 -- but now we don't. Instead the type checker just injects these
305 -- extra constraints where necessary.
306 wrap_tvs = (univ_tvs `minusList` map fst eq_spec) ++ ex_tvs
307 res_ty_args = substTyVars (mkTopTvSubst eq_spec) univ_tvs
308 eq_tys = mkPredTys eq_theta
309 dict_tys = mkPredTys dict_theta
310 wrap_ty = mkForAllTys wrap_tvs $ mkFunTys eq_tys $ mkFunTys dict_tys $
311 mkFunTys orig_arg_tys $ res_ty
312 -- NB: watch out here if you allow user-written equality
313 -- constraints in data constructor signatures
315 ----------- Wrappers for algebraic data types --------------
316 alg_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty alg_wrap_info
317 alg_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
318 `setArityInfo` wrap_arity
319 -- It's important to specify the arity, so that partial
320 -- applications are treated as values
321 `setUnfoldingInfo` wrap_unf
322 `setStrictnessInfo` Just wrap_sig
324 all_strict_marks = dataConExStricts data_con ++ dataConStrictMarks data_con
325 wrap_sig = mkStrictSig (mkTopDmdType arg_dmds cpr_info)
326 arg_dmds = map mk_dmd all_strict_marks
327 mk_dmd str | isBanged str = evalDmd
328 | otherwise = lazyDmd
329 -- The Cpr info can be important inside INLINE rhss, where the
330 -- wrapper constructor isn't inlined.
331 -- And the argument strictness can be important too; we
332 -- may not inline a contructor when it is partially applied.
334 -- data W = C !Int !Int !Int
335 -- ...(let w = C x in ...(w p q)...)...
336 -- we want to see that w is strict in its two arguments
338 wrap_unf = mkInlineRule wrap_rhs (Just (length dict_args + length id_args))
339 wrap_rhs = mkLams wrap_tvs $
341 mkLams dict_args $ mkLams id_args $
342 foldr mk_case con_app
343 (zip (dict_args ++ id_args) all_strict_marks)
346 con_app _ rep_ids = wrapFamInstBody tycon res_ty_args $
347 Var wrk_id `mkTyApps` res_ty_args
349 -- Equality evidence:
350 `mkTyApps` map snd eq_spec
352 `mkVarApps` reverse rep_ids
354 (dict_args,i2) = mkLocals 1 dict_tys
355 (id_args,i3) = mkLocals i2 orig_arg_tys
357 (eq_args,_) = mkCoVarLocals i3 eq_tys
359 mkCoVarLocals i [] = ([],i)
360 mkCoVarLocals i (x:xs) = let (ys,j) = mkCoVarLocals (i+1) xs
361 y = mkCoVar (mkSysTvName (mkBuiltinUnique i)
366 :: (Id, HsBang) -- Arg, strictness
367 -> (Int -> [Id] -> CoreExpr) -- Body
368 -> Int -- Next rep arg id
369 -> [Id] -- Rep args so far, reversed
371 mk_case (arg,strict) body i rep_args
373 HsNoBang -> body i (arg:rep_args)
374 HsUnpack -> unboxProduct i (Var arg) (idType arg) the_body
376 the_body i con_args = body i (reverse con_args ++ rep_args)
377 _other -- HsUnpackFailed and HsStrict
378 | isUnLiftedType (idType arg) -> body i (arg:rep_args)
379 | otherwise -> Case (Var arg) arg res_ty
380 [(DEFAULT,[], body i (arg:rep_args))]
382 mAX_CPR_SIZE :: Arity
384 -- We do not treat very big tuples as CPR-ish:
385 -- a) for a start we get into trouble because there aren't
386 -- "enough" unboxed tuple types (a tiresome restriction,
388 -- b) more importantly, big unboxed tuples get returned mainly
389 -- on the stack, and are often then allocated in the heap
390 -- by the caller. So doing CPR for them may in fact make
393 mkLocals :: Int -> [Type] -> ([Id], Int)
394 mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n)
399 Note [Newtype datacons]
400 ~~~~~~~~~~~~~~~~~~~~~~~
401 The "data constructor" for a newtype should always be vanilla. At one
402 point this wasn't true, because the newtype arising from
405 newtype T:D a = D:D (C a)
406 so the data constructor for T:C had a single argument, namely the
407 predicate (C a). But now we treat that as an ordinary argument, not
408 part of the theta-type, so all is well.
411 %************************************************************************
413 \subsection{Dictionary selectors}
415 %************************************************************************
417 Selecting a field for a dictionary. If there is just one field, then
418 there's nothing to do.
420 Dictionary selectors may get nested forall-types. Thus:
423 op :: forall b. Ord b => a -> b -> b
425 Then the top-level type for op is
427 op :: forall a. Foo a =>
431 This is unlike ordinary record selectors, which have all the for-alls
432 at the outside. When dealing with classes it's very convenient to
433 recover the original type signature from the class op selector.
436 mkDictSelId :: Bool -- True <=> don't include the unfolding
437 -- Little point on imports without -O, because the
438 -- dictionary itself won't be visible
439 -> Name -- Name of one of the *value* selectors
440 -- (dictionary superclass or method)
442 mkDictSelId no_unf name clas
443 = mkGlobalId (ClassOpId clas) name sel_ty info
445 sel_ty = mkForAllTys tyvars (mkFunTy (idType dict_id) (idType the_arg_id))
446 -- We can't just say (exprType rhs), because that would give a type
448 -- for a single-op class (after all, the selector is the identity)
449 -- But it's type must expose the representation of the dictionary
450 -- to get (say) C a -> (a -> a)
452 base_info = noCafIdInfo
454 `setStrictnessInfo` Just strict_sig
455 `setUnfoldingInfo` (if no_unf then noUnfolding
456 else mkImplicitUnfolding rhs)
457 -- In module where class op is defined, we must add
458 -- the unfolding, even though it'll never be inlined
459 -- becuase we use that to generate a top-level binding
462 info = base_info `setSpecInfo` mkSpecInfo [rule]
463 `setInlinePragInfo` neverInlinePragma
464 -- Add a magic BuiltinRule, and never inline it
465 -- so that the rule is always available to fire.
466 -- See Note [ClassOp/DFun selection] in TcInstDcls
468 n_ty_args = length tyvars
470 -- This is the built-in rule that goes
471 -- op (dfT d1 d2) ---> opT d1 d2
472 rule = BuiltinRule { ru_name = fsLit "Class op " `appendFS`
473 occNameFS (getOccName name)
475 , ru_nargs = n_ty_args + 1
476 , ru_try = dictSelRule val_index n_ty_args n_eq_args }
478 -- The strictness signature is of the form U(AAAVAAAA) -> T
479 -- where the V depends on which item we are selecting
480 -- It's worth giving one, so that absence info etc is generated
481 -- even if the selector isn't inlined
482 strict_sig = mkStrictSig (mkTopDmdType [arg_dmd] TopRes)
483 arg_dmd | new_tycon = evalDmd
484 | otherwise = Eval (Prod [ if the_arg_id == id then evalDmd else Abs
487 tycon = classTyCon clas
488 new_tycon = isNewTyCon tycon
489 [data_con] = tyConDataCons tycon
490 tyvars = dataConUnivTyVars data_con
491 arg_tys = dataConRepArgTys data_con -- Includes the dictionary superclasses
492 eq_theta = dataConEqTheta data_con
493 n_eq_args = length eq_theta
495 -- 'index' is a 0-index into the *value* arguments of the dictionary
496 val_index = assoc "MkId.mkDictSelId" sel_index_prs name
497 sel_index_prs = map idName (classAllSelIds clas) `zip` [0..]
499 the_arg_id = arg_ids !! val_index
500 pred = mkClassPred clas (mkTyVarTys tyvars)
501 dict_id = mkTemplateLocal 1 $ mkPredTy pred
502 arg_ids = mkTemplateLocalsNum 2 arg_tys
503 eq_ids = map mkWildEvBinder eq_theta
505 rhs = mkLams tyvars (Lam dict_id rhs_body)
506 rhs_body | new_tycon = unwrapNewTypeBody tycon (map mkTyVarTy tyvars) (Var dict_id)
507 | otherwise = Case (Var dict_id) dict_id (idType the_arg_id)
508 [(DataAlt data_con, eq_ids ++ arg_ids, Var the_arg_id)]
510 dictSelRule :: Int -> Arity -> Arity
511 -> IdUnfoldingFun -> [CoreExpr] -> Maybe CoreExpr
513 -- sel_i t1..tk (df s1..sn d1..dm) = op_i_helper s1..sn d1..dm
514 -- sel_i t1..tk (D t1..tk op1 ... opm) = opi
516 -- NB: the data constructor has the same number of type and
517 -- coercion args as the selector
519 -- This only works for *value* superclasses
520 -- There are no selector functions for equality superclasses
521 dictSelRule val_index n_ty_args n_eq_args id_unf args
522 | (dict_arg : _) <- drop n_ty_args args
523 , Just (_, _, con_args) <- exprIsConApp_maybe id_unf dict_arg
524 , let val_args = drop n_eq_args con_args
525 = Just (val_args !! val_index)
531 %************************************************************************
535 %************************************************************************
538 -- unbox a product type...
539 -- we will recurse into newtypes, casting along the way, and unbox at the
540 -- first product data constructor we find. e.g.
542 -- data PairInt = PairInt Int Int
543 -- newtype S = MkS PairInt
546 -- If we have e = MkT (MkS (PairInt 0 1)) and some body expecting a list of
547 -- ids, we get (modulo int passing)
549 -- case (e `cast` CoT) `cast` CoS of
550 -- PairInt a b -> body [a,b]
552 -- The Ints passed around are just for creating fresh locals
553 unboxProduct :: Int -> CoreExpr -> Type -> (Int -> [Id] -> CoreExpr) -> CoreExpr
554 unboxProduct i arg arg_ty body
557 result = mkUnpackCase the_id arg con_args boxing_con rhs
558 (_tycon, _tycon_args, boxing_con, tys) = deepSplitProductType "unboxProduct" arg_ty
559 ([the_id], i') = mkLocals i [arg_ty]
560 (con_args, i'') = mkLocals i' tys
561 rhs = body i'' con_args
563 mkUnpackCase :: Id -> CoreExpr -> [Id] -> DataCon -> CoreExpr -> CoreExpr
564 -- (mkUnpackCase x e args Con body)
566 -- case (e `cast` ...) of bndr { Con args -> body }
568 -- the type of the bndr passed in is irrelevent
569 mkUnpackCase bndr arg unpk_args boxing_con body
570 = Case cast_arg (setIdType bndr bndr_ty) (exprType body) [(DataAlt boxing_con, unpk_args, body)]
572 (cast_arg, bndr_ty) = go (idType bndr) arg
574 | (tycon, tycon_args, _, _) <- splitProductType "mkUnpackCase" ty
575 , isNewTyCon tycon && not (isRecursiveTyCon tycon)
576 = go (newTyConInstRhs tycon tycon_args)
577 (unwrapNewTypeBody tycon tycon_args arg)
578 | otherwise = (arg, ty)
581 reboxProduct :: [Unique] -- uniques to create new local binders
582 -> Type -- type of product to box
583 -> ([Unique], -- remaining uniques
584 CoreExpr, -- boxed product
585 [Id]) -- Ids being boxed into product
588 (_tycon, _tycon_args, _pack_con, con_arg_tys) = deepSplitProductType "reboxProduct" ty
590 us' = dropList con_arg_tys us
592 arg_ids = zipWith (mkSysLocal (fsLit "rb")) us con_arg_tys
594 bind_rhs = mkProductBox arg_ids ty
597 (us', bind_rhs, arg_ids)
599 mkProductBox :: [Id] -> Type -> CoreExpr
600 mkProductBox arg_ids ty
603 (tycon, tycon_args, pack_con, _con_arg_tys) = splitProductType "mkProductBox" ty
606 | isNewTyCon tycon && not (isRecursiveTyCon tycon)
607 = wrap (mkProductBox arg_ids (newTyConInstRhs tycon tycon_args))
608 | otherwise = mkConApp pack_con (map Type tycon_args ++ map Var arg_ids)
610 wrap expr = wrapNewTypeBody tycon tycon_args expr
613 -- (mkReboxingAlt us con xs rhs) basically constructs the case
614 -- alternative (con, xs, rhs)
615 -- but it does the reboxing necessary to construct the *source*
616 -- arguments, xs, from the representation arguments ys.
618 -- data T = MkT !(Int,Int) Bool
620 -- mkReboxingAlt MkT [x,b] r
621 -- = (DataAlt MkT, [y::Int,z::Int,b], let x = (y,z) in r)
623 -- mkDataAlt should really be in DataCon, but it can't because
624 -- it manipulates CoreSyn.
627 :: [Unique] -- Uniques for the new Ids
629 -> [Var] -- Source-level args, including existential dicts
633 mkReboxingAlt us con args rhs
634 | not (any isMarkedUnboxed stricts)
635 = (DataAlt con, args, rhs)
639 (binds, args') = go args stricts us
641 (DataAlt con, args', mkLets binds rhs)
644 stricts = dataConExStricts con ++ dataConStrictMarks con
646 go [] _stricts _us = ([], [])
648 -- Type variable case
649 go (arg:args) stricts us
651 = let (binds, args') = go args stricts us
652 in (binds, arg:args')
654 -- Term variable case
655 go (arg:args) (str:stricts) us
656 | isMarkedUnboxed str
658 let (binds, unpacked_args') = go args stricts us'
659 (us', bind_rhs, unpacked_args) = reboxProduct us (idType arg)
661 (NonRec arg bind_rhs : binds, unpacked_args ++ unpacked_args')
663 = let (binds, args') = go args stricts us
664 in (binds, arg:args')
665 go (_ : _) [] _ = panic "mkReboxingAlt"
669 %************************************************************************
671 Wrapping and unwrapping newtypes and type families
673 %************************************************************************
676 wrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
677 -- The wrapper for the data constructor for a newtype looks like this:
678 -- newtype T a = MkT (a,Int)
679 -- MkT :: forall a. (a,Int) -> T a
680 -- MkT = /\a. \(x:(a,Int)). x `cast` sym (CoT a)
681 -- where CoT is the coercion TyCon assoicated with the newtype
683 -- The call (wrapNewTypeBody T [a] e) returns the
684 -- body of the wrapper, namely
685 -- e `cast` (CoT [a])
687 -- If a coercion constructor is provided in the newtype, then we use
688 -- it, otherwise the wrap/unwrap are both no-ops
690 -- If the we are dealing with a newtype *instance*, we have a second coercion
691 -- identifying the family instance with the constructor of the newtype
692 -- instance. This coercion is applied in any case (ie, composed with the
693 -- coercion constructor of the newtype or applied by itself).
695 wrapNewTypeBody tycon args result_expr
696 = wrapFamInstBody tycon args inner
699 | Just co_con <- newTyConCo_maybe tycon
700 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) result_expr
704 -- When unwrapping, we do *not* apply any family coercion, because this will
705 -- be done via a CoPat by the type checker. We have to do it this way as
706 -- computing the right type arguments for the coercion requires more than just
707 -- a spliting operation (cf, TcPat.tcConPat).
709 unwrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
710 unwrapNewTypeBody tycon args result_expr
711 | Just co_con <- newTyConCo_maybe tycon
712 = mkCoerce (mkTyConApp co_con args) result_expr
716 -- If the type constructor is a representation type of a data instance, wrap
717 -- the expression into a cast adjusting the expression type, which is an
718 -- instance of the representation type, to the corresponding instance of the
719 -- family instance type.
720 -- See Note [Wrappers for data instance tycons]
721 wrapFamInstBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
722 wrapFamInstBody tycon args body
723 | Just co_con <- tyConFamilyCoercion_maybe tycon
724 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) body
728 unwrapFamInstScrut :: TyCon -> [Type] -> CoreExpr -> CoreExpr
729 unwrapFamInstScrut tycon args scrut
730 | Just co_con <- tyConFamilyCoercion_maybe tycon
731 = mkCoerce (mkTyConApp co_con args) scrut
737 %************************************************************************
739 \subsection{Primitive operations}
741 %************************************************************************
744 mkPrimOpId :: PrimOp -> Id
748 (tyvars,arg_tys,res_ty, arity, strict_sig) = primOpSig prim_op
749 ty = mkForAllTys tyvars (mkFunTys arg_tys res_ty)
750 name = mkWiredInName gHC_PRIM (primOpOcc prim_op)
751 (mkPrimOpIdUnique (primOpTag prim_op))
753 id = mkGlobalId (PrimOpId prim_op) name ty info
756 `setSpecInfo` mkSpecInfo (primOpRules prim_op name)
758 `setStrictnessInfo` Just strict_sig
760 -- For each ccall we manufacture a separate CCallOpId, giving it
761 -- a fresh unique, a type that is correct for this particular ccall,
762 -- and a CCall structure that gives the correct details about calling
765 -- The *name* of this Id is a local name whose OccName gives the full
766 -- details of the ccall, type and all. This means that the interface
767 -- file reader can reconstruct a suitable Id
769 mkFCallId :: Unique -> ForeignCall -> Type -> Id
770 mkFCallId uniq fcall ty
771 = ASSERT( isEmptyVarSet (tyVarsOfType ty) )
772 -- A CCallOpId should have no free type variables;
773 -- when doing substitutions won't substitute over it
774 mkGlobalId (FCallId fcall) name ty info
776 occ_str = showSDoc (braces (ppr fcall <+> ppr ty))
777 -- The "occurrence name" of a ccall is the full info about the
778 -- ccall; it is encoded, but may have embedded spaces etc!
780 name = mkFCallName uniq occ_str
784 `setStrictnessInfo` Just strict_sig
786 (_, tau) = tcSplitForAllTys ty
787 (arg_tys, _) = tcSplitFunTys tau
788 arity = length arg_tys
789 strict_sig = mkStrictSig (mkTopDmdType (replicate arity evalDmd) TopRes)
791 -- Tick boxes and breakpoints are both represented as TickBoxOpIds,
792 -- except for the type:
794 -- a plain HPC tick box has type (State# RealWorld)
795 -- a breakpoint Id has type forall a.a
797 -- The breakpoint Id will be applied to a list of arbitrary free variables,
798 -- which is why it needs a polymorphic type.
800 mkTickBoxOpId :: Unique -> Module -> TickBoxId -> Id
801 mkTickBoxOpId uniq mod ix = mkTickBox' uniq mod ix realWorldStatePrimTy
803 mkBreakPointOpId :: Unique -> Module -> TickBoxId -> Id
804 mkBreakPointOpId uniq mod ix = mkTickBox' uniq mod ix ty
805 where ty = mkSigmaTy [openAlphaTyVar] [] openAlphaTy
807 mkTickBox' :: Unique -> Module -> TickBoxId -> Type -> Id
808 mkTickBox' uniq mod ix ty = mkGlobalId (TickBoxOpId tickbox) name ty info
810 tickbox = TickBox mod ix
811 occ_str = showSDoc (braces (ppr tickbox))
812 name = mkTickBoxOpName uniq occ_str
817 %************************************************************************
819 \subsection{DictFuns and default methods}
821 %************************************************************************
823 Important notes about dict funs and default methods
824 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
825 Dict funs and default methods are *not* ImplicitIds. Their definition
826 involves user-written code, so we can't figure out their strictness etc
827 based on fixed info, as we can for constructors and record selectors (say).
829 We build them as LocalIds, but with External Names. This ensures that
830 they are taken to account by free-variable finding and dependency
831 analysis (e.g. CoreFVs.exprFreeVars).
833 Why shouldn't they be bound as GlobalIds? Because, in particular, if
834 they are globals, the specialiser floats dict uses above their defns,
835 which prevents good simplifications happening. Also the strictness
836 analyser treats a occurrence of a GlobalId as imported and assumes it
837 contains strictness in its IdInfo, which isn't true if the thing is
838 bound in the same module as the occurrence.
840 It's OK for dfuns to be LocalIds, because we form the instance-env to
841 pass on to the next module (md_insts) in CoreTidy, afer tidying
842 and globalising the top-level Ids.
844 BUT make sure they are *exported* LocalIds (mkExportedLocalId) so
845 that they aren't discarded by the occurrence analyser.
848 mkDefaultMethodId :: Id -- Selector Id
849 -> Name -- Default method name
850 -> Id -- Default method Id
851 mkDefaultMethodId sel_id dm_name = mkExportedLocalId dm_name (idType sel_id)
853 mkDictFunId :: Name -- Name to use for the dict fun;
860 mkDictFunId dfun_name inst_tyvars dfun_theta clas inst_tys
861 = mkExportedLocalVar (DFunId is_nt) dfun_name dfun_ty vanillaIdInfo
863 is_nt = isNewTyCon (classTyCon clas)
864 dfun_ty = mkSigmaTy inst_tyvars dfun_theta (mkDictTy clas inst_tys)
868 %************************************************************************
870 \subsection{Un-definable}
872 %************************************************************************
874 These Ids can't be defined in Haskell. They could be defined in
875 unfoldings in the wired-in GHC.Prim interface file, but we'd have to
876 ensure that they were definitely, definitely inlined, because there is
877 no curried identifier for them. That's what mkCompulsoryUnfolding
878 does. If we had a way to get a compulsory unfolding from an interface
879 file, we could do that, but we don't right now.
881 unsafeCoerce# isn't so much a PrimOp as a phantom identifier, that
882 just gets expanded into a type coercion wherever it occurs. Hence we
883 add it as a built-in Id with an unfolding here.
885 The type variables we use here are "open" type variables: this means
886 they can unify with both unlifted and lifted types. Hence we provide
887 another gun with which to shoot yourself in the foot.
890 lazyIdName, unsafeCoerceName, nullAddrName, seqName, realWorldName :: Name
891 unsafeCoerceName = mkWiredInIdName gHC_PRIM (fsLit "unsafeCoerce#") unsafeCoerceIdKey unsafeCoerceId
892 nullAddrName = mkWiredInIdName gHC_PRIM (fsLit "nullAddr#") nullAddrIdKey nullAddrId
893 seqName = mkWiredInIdName gHC_PRIM (fsLit "seq") seqIdKey seqId
894 realWorldName = mkWiredInIdName gHC_PRIM (fsLit "realWorld#") realWorldPrimIdKey realWorldPrimId
895 lazyIdName = mkWiredInIdName gHC_BASE (fsLit "lazy") lazyIdKey lazyId
899 ------------------------------------------------
900 -- unsafeCoerce# :: forall a b. a -> b
903 = pcMiscPrelId unsafeCoerceName ty info
905 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
908 ty = mkForAllTys [argAlphaTyVar,openBetaTyVar]
909 (mkFunTy argAlphaTy openBetaTy)
910 [x] = mkTemplateLocals [argAlphaTy]
911 rhs = mkLams [argAlphaTyVar,openBetaTyVar,x] $
912 Cast (Var x) (mkUnsafeCoercion argAlphaTy openBetaTy)
914 ------------------------------------------------
916 -- nullAddr# :: Addr#
917 -- The reason is is here is because we don't provide
918 -- a way to write this literal in Haskell.
919 nullAddrId = pcMiscPrelId nullAddrName addrPrimTy info
921 info = noCafIdInfo `setUnfoldingInfo`
922 mkCompulsoryUnfolding (Lit nullAddrLit)
924 ------------------------------------------------
925 seqId :: Id -- See Note [seqId magic]
926 seqId = pcMiscPrelId seqName ty info
928 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
929 `setSpecInfo` mkSpecInfo [seq_cast_rule]
932 ty = mkForAllTys [alphaTyVar,argBetaTyVar]
933 (mkFunTy alphaTy (mkFunTy argBetaTy argBetaTy))
934 [x,y] = mkTemplateLocals [alphaTy, argBetaTy]
935 rhs = mkLams [alphaTyVar,argBetaTyVar,x,y] (Case (Var x) x argBetaTy [(DEFAULT, [], Var y)])
937 -- See Note [Built-in RULES for seq]
938 seq_cast_rule = BuiltinRule { ru_name = fsLit "seq of cast"
941 , ru_try = match_seq_of_cast
944 match_seq_of_cast :: IdUnfoldingFun -> [CoreExpr] -> Maybe CoreExpr
945 -- See Note [Built-in RULES for seq]
946 match_seq_of_cast _ [Type _, Type res_ty, Cast scrut co, expr]
947 = Just (Var seqId `mkApps` [Type (fst (coercionKind co)), Type res_ty,
949 match_seq_of_cast _ _ = Nothing
951 ------------------------------------------------
952 lazyId :: Id -- See Note [lazyId magic]
953 lazyId = pcMiscPrelId lazyIdName ty info
956 ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy)
961 'GHC.Prim.seq' is special in several ways.
963 a) Its second arg can have an unboxed type
966 b) Its fixity is set in LoadIface.ghcPrimIface
968 c) It has quite a bit of desugaring magic.
969 See DsUtils.lhs Note [Desugaring seq (1)] and (2) and (3)
971 d) There is some special rule handing: Note [User-defined RULES for seq]
973 e) See Note [Typing rule for seq] in TcExpr.
975 Note [User-defined RULES for seq]
976 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
977 Roman found situations where he had
979 where he knew that f (which was strict in n) would terminate if n did.
980 Notice that the result of (f n) is discarded. So it makes sense to
984 Rather than attempt some general analysis to support this, I've added
985 enough support that you can do this using a rewrite rule:
987 RULE "f/seq" forall n. seq (f n) e = seq n e
989 You write that rule. When GHC sees a case expression that discards
990 its result, it mentally transforms it to a call to 'seq' and looks for
991 a RULE. (This is done in Simplify.rebuildCase.) As usual, the
992 correctness of the rule is up to you.
994 To make this work, we need to be careful that the magical desugaring
995 done in Note [seqId magic] item (c) is *not* done on the LHS of a rule.
996 Or rather, we arrange to un-do it, in DsBinds.decomposeRuleLhs.
998 Note [Built-in RULES for seq]
999 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1000 We also have the following built-in rule for seq
1002 seq (x `cast` co) y = seq x y
1004 This eliminates unnecessary casts and also allows other seq rules to
1005 match more often. Notably,
1007 seq (f x `cast` co) y --> seq (f x) y
1009 and now a user-defined rule for seq (see Note [User-defined RULES for seq])
1015 lazy :: forall a?. a? -> a? (i.e. works for unboxed types too)
1017 Used to lazify pseq: pseq a b = a `seq` lazy b
1019 Also, no strictness: by being a built-in Id, all the info about lazyId comes from here,
1020 not from GHC.Base.hi. This is important, because the strictness
1021 analyser will spot it as strict!
1023 Also no unfolding in lazyId: it gets "inlined" by a HACK in CorePrep.
1024 It's very important to do this inlining *after* unfoldings are exposed
1025 in the interface file. Otherwise, the unfolding for (say) pseq in the
1026 interface file will not mention 'lazy', so if we inline 'pseq' we'll totally
1027 miss the very thing that 'lazy' was there for in the first place.
1028 See Trac #3259 for a real world example.
1030 lazyId is defined in GHC.Base, so we don't *have* to inline it. If it
1031 appears un-applied, we'll end up just calling it.
1033 -------------------------------------------------------------
1034 @realWorld#@ used to be a magic literal, \tr{void#}. If things get
1035 nasty as-is, change it back to a literal (@Literal@).
1037 voidArgId is a Local Id used simply as an argument in functions
1038 where we just want an arg to avoid having a thunk of unlifted type.
1040 x = \ void :: State# RealWorld -> (# p, q #)
1042 This comes up in strictness analysis
1045 realWorldPrimId :: Id
1046 realWorldPrimId -- :: State# RealWorld
1047 = pcMiscPrelId realWorldName realWorldStatePrimTy
1048 (noCafIdInfo `setUnfoldingInfo` evaldUnfolding)
1049 -- The evaldUnfolding makes it look that realWorld# is evaluated
1050 -- which in turn makes Simplify.interestingArg return True,
1051 -- which in turn makes INLINE things applied to realWorld# likely
1055 voidArgId -- :: State# RealWorld
1056 = mkSysLocal (fsLit "void") voidArgIdKey realWorldStatePrimTy
1061 pcMiscPrelId :: Name -> Type -> IdInfo -> Id
1062 pcMiscPrelId name ty info
1063 = mkVanillaGlobalWithInfo name ty info
1064 -- We lie and say the thing is imported; otherwise, we get into
1065 -- a mess with dependency analysis; e.g., core2stg may heave in
1066 -- random calls to GHCbase.unpackPS__. If GHCbase is the module
1067 -- being compiled, then it's just a matter of luck if the definition
1068 -- will be in "the right place" to be in scope.