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, mkDictFunTy, mkDefaultMethodId, mkDictSelId,
19 mkPrimOpId, mkFCallId, mkTickBoxOpId, mkBreakPointOpId,
21 mkReboxingAlt, wrapNewTypeBody, unwrapNewTypeBody,
22 wrapFamInstBody, unwrapFamInstScrut,
23 mkUnpackCase, mkProductBox,
25 -- And some particular Ids; see below for why they are wired in
26 wiredInIds, ghcPrimIds,
27 unsafeCoerceName, unsafeCoerceId, realWorldPrimId,
28 voidArgId, nullAddrId, seqId, lazyId, lazyIdKey,
31 -- Re-export error Ids
35 #include "HsVersions.h"
39 import TysWiredIn ( unitTy )
45 import CoreUtils ( exprType, mkCoerce )
56 import Var ( mkExportedLocalVar )
62 import BasicTypes hiding ( SuccessFlag(..) )
71 %************************************************************************
73 \subsection{Wired in Ids}
75 %************************************************************************
79 There are several reasons why an Id might appear in the wiredInIds:
81 (1) The ghcPrimIds are wired in because they can't be defined in
82 Haskell at all, although the can be defined in Core. They have
83 compulsory unfoldings, so they are always inlined and they have
84 no definition site. Their home module is GHC.Prim, so they
85 also have a description in primops.txt.pp, where they are called
88 (2) The 'error' function, eRROR_ID, is wired in because we don't yet have
89 a way to express in an interface file that the result type variable
90 is 'open'; that is can be unified with an unboxed type
92 [The interface file format now carry such information, but there's
93 no way yet of expressing at the definition site for these
94 error-reporting functions that they have an 'open'
95 result type. -- sof 1/99]
97 (3) Other error functions (rUNTIME_ERROR_ID) are wired in (a) because
98 the desugarer generates code that mentiones them directly, and
99 (b) for the same reason as eRROR_ID
101 (4) lazyId is wired in because the wired-in version overrides the
102 strictness of the version defined in GHC.Base
104 In cases (2-4), the function has a definition in a library module, and
105 can be called; but the wired-in version means that the details are
106 never read from that module's interface file; instead, the full definition
113 ++ errorIds -- Defined in MkCore
116 -- These Ids are exported from GHC.Prim
119 = [ -- These can't be defined in Haskell, but they have
120 -- perfectly reasonable unfoldings in Core
128 %************************************************************************
130 \subsection{Data constructors}
132 %************************************************************************
134 The wrapper for a constructor is an ordinary top-level binding that evaluates
135 any strict args, unboxes any args that are going to be flattened, and calls
138 We're going to build a constructor that looks like:
140 data (Data a, C b) => T a b = T1 !a !Int b
143 \d1::Data a, d2::C b ->
144 \p q r -> case p of { p ->
146 Con T1 [a,b] [p,q,r]}}
150 * d2 is thrown away --- a context in a data decl is used to make sure
151 one *could* construct dictionaries at the site the constructor
152 is used, but the dictionary isn't actually used.
154 * We have to check that we can construct Data dictionaries for
155 the types a and Int. Once we've done that we can throw d1 away too.
157 * We use (case p of q -> ...) to evaluate p, rather than "seq" because
158 all that matters is that the arguments are evaluated. "seq" is
159 very careful to preserve evaluation order, which we don't need
162 You might think that we could simply give constructors some strictness
163 info, like PrimOps, and let CoreToStg do the let-to-case transformation.
164 But we don't do that because in the case of primops and functions strictness
165 is a *property* not a *requirement*. In the case of constructors we need to
166 do something active to evaluate the argument.
168 Making an explicit case expression allows the simplifier to eliminate
169 it in the (common) case where the constructor arg is already evaluated.
171 Note [Wrappers for data instance tycons]
172 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
173 In the case of data instances, the wrapper also applies the coercion turning
174 the representation type into the family instance type to cast the result of
175 the wrapper. For example, consider the declarations
177 data family Map k :: * -> *
178 data instance Map (a, b) v = MapPair (Map a (Pair b v))
180 The tycon to which the datacon MapPair belongs gets a unique internal
181 name of the form :R123Map, and we call it the representation tycon.
182 In contrast, Map is the family tycon (accessible via
183 tyConFamInst_maybe). A coercion allows you to move between
184 representation and family type. It is accessible from :R123Map via
185 tyConFamilyCoercion_maybe and has kind
187 Co123Map a b v :: {Map (a, b) v ~ :R123Map a b v}
189 The wrapper and worker of MapPair get the types
192 $WMapPair :: forall a b v. Map a (Map a b v) -> Map (a, b) v
193 $WMapPair a b v = MapPair a b v `cast` sym (Co123Map a b v)
196 MapPair :: forall a b v. Map a (Map a b v) -> :R123Map a b v
198 This coercion is conditionally applied by wrapFamInstBody.
200 It's a bit more complicated if the data instance is a GADT as well!
202 data instance T [a] where
203 T1 :: forall b. b -> T [Maybe b]
205 Hence we translate to
208 $WT1 :: forall b. b -> T [Maybe b]
209 $WT1 b v = T1 (Maybe b) b (Maybe b) v
210 `cast` sym (Co7T (Maybe b))
213 T1 :: forall c b. (c ~ Maybe b) => b -> :R7T c
215 -- Coercion from family type to representation type
216 Co7T a :: T [a] ~ :R7T a
219 mkDataConIds :: Name -> Name -> DataCon -> DataConIds
220 mkDataConIds wrap_name wkr_name data_con
221 | isNewTyCon tycon -- Newtype, only has a worker
222 = DCIds Nothing nt_work_id
224 | any isBanged all_strict_marks -- Algebraic, needs wrapper
225 || not (null eq_spec) -- NB: LoadIface.ifaceDeclSubBndrs
226 || isFamInstTyCon tycon -- depends on this test
227 = DCIds (Just alg_wrap_id) wrk_id
229 | otherwise -- Algebraic, no wrapper
230 = DCIds Nothing wrk_id
232 (univ_tvs, ex_tvs, eq_spec,
233 theta, orig_arg_tys, res_ty) = dataConFullSig data_con
234 tycon = dataConTyCon data_con -- The representation TyCon (not family)
236 ----------- Worker (algebraic data types only) --------------
237 -- The *worker* for the data constructor is the function that
238 -- takes the representation arguments and builds the constructor.
239 wrk_id = mkGlobalId (DataConWorkId data_con) wkr_name
240 (dataConRepType data_con) wkr_info
242 wkr_arity = dataConRepArity data_con
243 wkr_info = noCafIdInfo
244 `setArityInfo` wkr_arity
245 `setStrictnessInfo` Just wkr_sig
246 `setUnfoldingInfo` evaldUnfolding -- Record that it's evaluated,
249 wkr_sig = mkStrictSig (mkTopDmdType (replicate wkr_arity topDmd) cpr_info)
250 -- Note [Data-con worker strictness]
251 -- Notice that we do *not* say the worker is strict
252 -- even if the data constructor is declared strict
253 -- e.g. data T = MkT !(Int,Int)
254 -- Why? Because the *wrapper* is strict (and its unfolding has case
255 -- expresssions that do the evals) but the *worker* itself is not.
256 -- If we pretend it is strict then when we see
257 -- case x of y -> $wMkT y
258 -- the simplifier thinks that y is "sure to be evaluated" (because
259 -- $wMkT is strict) and drops the case. No, $wMkT is not strict.
261 -- When the simplifer sees a pattern
262 -- case e of MkT x -> ...
263 -- it uses the dataConRepStrictness of MkT to mark x as evaluated;
264 -- but that's fine... dataConRepStrictness comes from the data con
265 -- not from the worker Id.
267 cpr_info | isProductTyCon tycon &&
270 wkr_arity <= mAX_CPR_SIZE = retCPR
272 -- RetCPR is only true for products that are real data types;
273 -- that is, not unboxed tuples or [non-recursive] newtypes
275 ----------- Workers for newtypes --------------
276 nt_work_id = mkGlobalId (DataConWrapId data_con) wkr_name wrap_ty nt_work_info
277 nt_work_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
278 `setArityInfo` 1 -- Arity 1
279 `setInlinePragInfo` alwaysInlinePragma
280 `setUnfoldingInfo` newtype_unf
281 id_arg1 = mkTemplateLocal 1 (head orig_arg_tys)
282 newtype_unf = ASSERT2( isVanillaDataCon data_con &&
283 isSingleton orig_arg_tys, ppr data_con )
284 -- Note [Newtype datacons]
285 mkCompulsoryUnfolding $
286 mkLams wrap_tvs $ Lam id_arg1 $
287 wrapNewTypeBody tycon res_ty_args (Var id_arg1)
290 ----------- Wrapper --------------
291 -- We used to include the stupid theta in the wrapper's args
292 -- but now we don't. Instead the type checker just injects these
293 -- extra constraints where necessary.
294 wrap_tvs = (univ_tvs `minusList` map fst eq_spec) ++ ex_tvs
295 res_ty_args = substTyVars (mkTopTvSubst eq_spec) univ_tvs
296 ev_tys = mkPredTys theta
297 wrap_ty = mkForAllTys wrap_tvs $
299 mkFunTys orig_arg_tys $ res_ty
301 ----------- Wrappers for algebraic data types --------------
302 alg_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty alg_wrap_info
303 alg_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
304 `setArityInfo` wrap_arity
305 -- It's important to specify the arity, so that partial
306 -- applications are treated as values
307 `setInlinePragInfo` alwaysInlinePragma
308 `setUnfoldingInfo` wrap_unf
309 `setStrictnessInfo` Just wrap_sig
311 all_strict_marks = dataConExStricts data_con ++ dataConStrictMarks data_con
312 wrap_sig = mkStrictSig (mkTopDmdType arg_dmds cpr_info)
313 arg_dmds = map mk_dmd all_strict_marks
314 mk_dmd str | isBanged str = evalDmd
315 | otherwise = lazyDmd
316 -- The Cpr info can be important inside INLINE rhss, where the
317 -- wrapper constructor isn't inlined.
318 -- And the argument strictness can be important too; we
319 -- may not inline a contructor when it is partially applied.
321 -- data W = C !Int !Int !Int
322 -- ...(let w = C x in ...(w p q)...)...
323 -- we want to see that w is strict in its two arguments
325 wrap_unf = mkInlineUnfolding (Just (length ev_args + length id_args)) wrap_rhs
326 wrap_rhs = mkLams wrap_tvs $
329 foldr mk_case con_app
330 (zip (ev_args ++ id_args) all_strict_marks)
333 con_app _ rep_ids = wrapFamInstBody tycon res_ty_args $
334 Var wrk_id `mkTyApps` res_ty_args
336 `mkCoApps` map (mkReflCo . snd) eq_spec
337 `mkVarApps` reverse rep_ids
339 (ev_args,i2) = mkLocals 1 ev_tys
340 (id_args,i3) = mkLocals i2 orig_arg_tys
344 :: (Id, HsBang) -- Arg, strictness
345 -> (Int -> [Id] -> CoreExpr) -- Body
346 -> Int -- Next rep arg id
347 -> [Id] -- Rep args so far, reversed
349 mk_case (arg,strict) body i rep_args
351 HsNoBang -> body i (arg:rep_args)
352 HsUnpack -> unboxProduct i (Var arg) (idType arg) the_body
354 the_body i con_args = body i (reverse con_args ++ rep_args)
355 _other -- HsUnpackFailed and HsStrict
356 | isUnLiftedType (idType arg) -> body i (arg:rep_args)
357 | otherwise -> Case (Var arg) arg res_ty
358 [(DEFAULT,[], body i (arg:rep_args))]
360 mAX_CPR_SIZE :: Arity
362 -- We do not treat very big tuples as CPR-ish:
363 -- a) for a start we get into trouble because there aren't
364 -- "enough" unboxed tuple types (a tiresome restriction,
366 -- b) more importantly, big unboxed tuples get returned mainly
367 -- on the stack, and are often then allocated in the heap
368 -- by the caller. So doing CPR for them may in fact make
371 mkLocals :: Int -> [Type] -> ([Id], Int)
372 mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n)
377 Note [Newtype datacons]
378 ~~~~~~~~~~~~~~~~~~~~~~~
379 The "data constructor" for a newtype should always be vanilla. At one
380 point this wasn't true, because the newtype arising from
383 newtype T:D a = D:D (C a)
384 so the data constructor for T:C had a single argument, namely the
385 predicate (C a). But now we treat that as an ordinary argument, not
386 part of the theta-type, so all is well.
389 %************************************************************************
391 \subsection{Dictionary selectors}
393 %************************************************************************
395 Selecting a field for a dictionary. If there is just one field, then
396 there's nothing to do.
398 Dictionary selectors may get nested forall-types. Thus:
401 op :: forall b. Ord b => a -> b -> b
403 Then the top-level type for op is
405 op :: forall a. Foo a =>
409 This is unlike ordinary record selectors, which have all the for-alls
410 at the outside. When dealing with classes it's very convenient to
411 recover the original type signature from the class op selector.
414 mkDictSelId :: Bool -- True <=> don't include the unfolding
415 -- Little point on imports without -O, because the
416 -- dictionary itself won't be visible
417 -> Name -- Name of one of the *value* selectors
418 -- (dictionary superclass or method)
420 mkDictSelId no_unf name clas
421 = mkGlobalId (ClassOpId clas) name sel_ty info
423 sel_ty = mkForAllTys tyvars (mkFunTy (idType dict_id) (idType the_arg_id))
424 -- We can't just say (exprType rhs), because that would give a type
426 -- for a single-op class (after all, the selector is the identity)
427 -- But it's type must expose the representation of the dictionary
428 -- to get (say) C a -> (a -> a)
430 base_info = noCafIdInfo
432 `setStrictnessInfo` Just strict_sig
433 `setUnfoldingInfo` (if no_unf then noUnfolding
434 else mkImplicitUnfolding rhs)
435 -- In module where class op is defined, we must add
436 -- the unfolding, even though it'll never be inlined
437 -- becuase we use that to generate a top-level binding
440 info | new_tycon = base_info `setInlinePragInfo` alwaysInlinePragma
441 -- See Note [Single-method classes] for why alwaysInlinePragma
442 | otherwise = base_info `setSpecInfo` mkSpecInfo [rule]
443 `setInlinePragInfo` neverInlinePragma
444 -- Add a magic BuiltinRule, and never inline it
445 -- so that the rule is always available to fire.
446 -- See Note [ClassOp/DFun selection] in TcInstDcls
448 n_ty_args = length tyvars
450 -- This is the built-in rule that goes
451 -- op (dfT d1 d2) ---> opT d1 d2
452 rule = BuiltinRule { ru_name = fsLit "Class op " `appendFS`
453 occNameFS (getOccName name)
455 , ru_nargs = n_ty_args + 1
456 , ru_try = dictSelRule val_index n_ty_args }
458 -- The strictness signature is of the form U(AAAVAAAA) -> T
459 -- where the V depends on which item we are selecting
460 -- It's worth giving one, so that absence info etc is generated
461 -- even if the selector isn't inlined
462 strict_sig = mkStrictSig (mkTopDmdType [arg_dmd] TopRes)
463 arg_dmd | new_tycon = evalDmd
464 | otherwise = Eval (Prod [ if the_arg_id == id then evalDmd else Abs
467 tycon = classTyCon clas
468 new_tycon = isNewTyCon tycon
469 [data_con] = tyConDataCons tycon
470 tyvars = dataConUnivTyVars data_con
471 arg_tys = dataConRepArgTys data_con -- Includes the dictionary superclasses
473 -- 'index' is a 0-index into the *value* arguments of the dictionary
474 val_index = assoc "MkId.mkDictSelId" sel_index_prs name
475 sel_index_prs = map idName (classAllSelIds clas) `zip` [0..]
477 the_arg_id = arg_ids !! val_index
478 pred = mkClassPred clas (mkTyVarTys tyvars)
479 dict_id = mkTemplateLocal 1 $ mkPredTy pred
480 arg_ids = mkTemplateLocalsNum 2 arg_tys
482 rhs = mkLams tyvars (Lam dict_id rhs_body)
483 rhs_body | new_tycon = unwrapNewTypeBody tycon (map mkTyVarTy tyvars) (Var dict_id)
484 | otherwise = Case (Var dict_id) dict_id (idType the_arg_id)
485 [(DataAlt data_con, arg_ids, Var the_arg_id)]
487 dictSelRule :: Int -> Arity
488 -> IdUnfoldingFun -> [CoreExpr] -> Maybe CoreExpr
489 -- Tries to persuade the argument to look like a constructor
490 -- application, using exprIsConApp_maybe, and then selects
492 -- sel_i t1..tk (D t1..tk op1 ... opm) = opi
494 dictSelRule val_index n_ty_args id_unf args
495 | (dict_arg : _) <- drop n_ty_args args
496 , Just (_, _, con_args) <- exprIsConApp_maybe id_unf dict_arg
497 = Just (con_args !! val_index)
503 %************************************************************************
507 %************************************************************************
510 -- unbox a product type...
511 -- we will recurse into newtypes, casting along the way, and unbox at the
512 -- first product data constructor we find. e.g.
514 -- data PairInt = PairInt Int Int
515 -- newtype S = MkS PairInt
518 -- If we have e = MkT (MkS (PairInt 0 1)) and some body expecting a list of
519 -- ids, we get (modulo int passing)
521 -- case (e `cast` CoT) `cast` CoS of
522 -- PairInt a b -> body [a,b]
524 -- The Ints passed around are just for creating fresh locals
525 unboxProduct :: Int -> CoreExpr -> Type -> (Int -> [Id] -> CoreExpr) -> CoreExpr
526 unboxProduct i arg arg_ty body
529 result = mkUnpackCase the_id arg con_args boxing_con rhs
530 (_tycon, _tycon_args, boxing_con, tys) = deepSplitProductType "unboxProduct" arg_ty
531 ([the_id], i') = mkLocals i [arg_ty]
532 (con_args, i'') = mkLocals i' tys
533 rhs = body i'' con_args
535 mkUnpackCase :: Id -> CoreExpr -> [Id] -> DataCon -> CoreExpr -> CoreExpr
536 -- (mkUnpackCase x e args Con body)
538 -- case (e `cast` ...) of bndr { Con args -> body }
540 -- the type of the bndr passed in is irrelevent
541 mkUnpackCase bndr arg unpk_args boxing_con body
542 = Case cast_arg (setIdType bndr bndr_ty) (exprType body) [(DataAlt boxing_con, unpk_args, body)]
544 (cast_arg, bndr_ty) = go (idType bndr) arg
546 | (tycon, tycon_args, _, _) <- splitProductType "mkUnpackCase" ty
547 , isNewTyCon tycon && not (isRecursiveTyCon tycon)
548 = go (newTyConInstRhs tycon tycon_args)
549 (unwrapNewTypeBody tycon tycon_args arg)
550 | otherwise = (arg, ty)
553 reboxProduct :: [Unique] -- uniques to create new local binders
554 -> Type -- type of product to box
555 -> ([Unique], -- remaining uniques
556 CoreExpr, -- boxed product
557 [Id]) -- Ids being boxed into product
560 (_tycon, _tycon_args, _pack_con, con_arg_tys) = deepSplitProductType "reboxProduct" ty
562 us' = dropList con_arg_tys us
564 arg_ids = zipWith (mkSysLocal (fsLit "rb")) us con_arg_tys
566 bind_rhs = mkProductBox arg_ids ty
569 (us', bind_rhs, arg_ids)
571 mkProductBox :: [Id] -> Type -> CoreExpr
572 mkProductBox arg_ids ty
575 (tycon, tycon_args, pack_con, _con_arg_tys) = splitProductType "mkProductBox" ty
578 | isNewTyCon tycon && not (isRecursiveTyCon tycon)
579 = wrap (mkProductBox arg_ids (newTyConInstRhs tycon tycon_args))
580 | otherwise = mkConApp pack_con (map Type tycon_args ++ map Var arg_ids)
582 wrap expr = wrapNewTypeBody tycon tycon_args expr
585 -- (mkReboxingAlt us con xs rhs) basically constructs the case
586 -- alternative (con, xs, rhs)
587 -- but it does the reboxing necessary to construct the *source*
588 -- arguments, xs, from the representation arguments ys.
590 -- data T = MkT !(Int,Int) Bool
592 -- mkReboxingAlt MkT [x,b] r
593 -- = (DataAlt MkT, [y::Int,z::Int,b], let x = (y,z) in r)
595 -- mkDataAlt should really be in DataCon, but it can't because
596 -- it manipulates CoreSyn.
599 :: [Unique] -- Uniques for the new Ids
601 -> [Var] -- Source-level args, including existential dicts
605 mkReboxingAlt us con args rhs
606 | not (any isMarkedUnboxed stricts)
607 = (DataAlt con, args, rhs)
611 (binds, args') = go args stricts us
613 (DataAlt con, args', mkLets binds rhs)
616 stricts = dataConExStricts con ++ dataConStrictMarks con
618 go [] _stricts _us = ([], [])
620 -- Type variable case
621 go (arg:args) stricts us
623 = let (binds, args') = go args stricts us
624 in (binds, arg:args')
626 -- Term variable case
627 go (arg:args) (str:stricts) us
628 | isMarkedUnboxed str
630 let (binds, unpacked_args') = go args stricts us'
631 (us', bind_rhs, unpacked_args) = reboxProduct us (idType arg)
633 (NonRec arg bind_rhs : binds, unpacked_args ++ unpacked_args')
635 = let (binds, args') = go args stricts us
636 in (binds, arg:args')
637 go (_ : _) [] _ = panic "mkReboxingAlt"
641 %************************************************************************
643 Wrapping and unwrapping newtypes and type families
645 %************************************************************************
648 wrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
649 -- The wrapper for the data constructor for a newtype looks like this:
650 -- newtype T a = MkT (a,Int)
651 -- MkT :: forall a. (a,Int) -> T a
652 -- MkT = /\a. \(x:(a,Int)). x `cast` sym (CoT a)
653 -- where CoT is the coercion TyCon assoicated with the newtype
655 -- The call (wrapNewTypeBody T [a] e) returns the
656 -- body of the wrapper, namely
657 -- e `cast` (CoT [a])
659 -- If a coercion constructor is provided in the newtype, then we use
660 -- it, otherwise the wrap/unwrap are both no-ops
662 -- If the we are dealing with a newtype *instance*, we have a second coercion
663 -- identifying the family instance with the constructor of the newtype
664 -- instance. This coercion is applied in any case (ie, composed with the
665 -- coercion constructor of the newtype or applied by itself).
667 wrapNewTypeBody tycon args result_expr
668 = ASSERT( isNewTyCon tycon )
669 wrapFamInstBody tycon args $
670 mkCoerce (mkSymCo co) result_expr
672 co = mkAxInstCo (newTyConCo tycon) args
674 -- When unwrapping, we do *not* apply any family coercion, because this will
675 -- be done via a CoPat by the type checker. We have to do it this way as
676 -- computing the right type arguments for the coercion requires more than just
677 -- a spliting operation (cf, TcPat.tcConPat).
679 unwrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
680 unwrapNewTypeBody tycon args result_expr
681 = ASSERT( isNewTyCon tycon )
682 mkCoerce (mkAxInstCo (newTyConCo tycon) args) result_expr
684 -- If the type constructor is a representation type of a data instance, wrap
685 -- the expression into a cast adjusting the expression type, which is an
686 -- instance of the representation type, to the corresponding instance of the
687 -- family instance type.
688 -- See Note [Wrappers for data instance tycons]
689 wrapFamInstBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
690 wrapFamInstBody tycon args body
691 | Just co_con <- tyConFamilyCoercion_maybe tycon
692 = mkCoerce (mkSymCo (mkAxInstCo co_con args)) body
696 unwrapFamInstScrut :: TyCon -> [Type] -> CoreExpr -> CoreExpr
697 unwrapFamInstScrut tycon args scrut
698 | Just co_con <- tyConFamilyCoercion_maybe tycon
699 = mkCoerce (mkAxInstCo co_con args) scrut
705 %************************************************************************
707 \subsection{Primitive operations}
709 %************************************************************************
712 mkPrimOpId :: PrimOp -> Id
716 (tyvars,arg_tys,res_ty, arity, strict_sig) = primOpSig prim_op
717 ty = mkForAllTys tyvars (mkFunTys arg_tys res_ty)
718 name = mkWiredInName gHC_PRIM (primOpOcc prim_op)
719 (mkPrimOpIdUnique (primOpTag prim_op))
721 id = mkGlobalId (PrimOpId prim_op) name ty info
724 `setSpecInfo` mkSpecInfo (primOpRules prim_op name)
726 `setStrictnessInfo` Just strict_sig
728 -- For each ccall we manufacture a separate CCallOpId, giving it
729 -- a fresh unique, a type that is correct for this particular ccall,
730 -- and a CCall structure that gives the correct details about calling
733 -- The *name* of this Id is a local name whose OccName gives the full
734 -- details of the ccall, type and all. This means that the interface
735 -- file reader can reconstruct a suitable Id
737 mkFCallId :: Unique -> ForeignCall -> Type -> Id
738 mkFCallId uniq fcall ty
739 = ASSERT( isEmptyVarSet (tyVarsOfType ty) )
740 -- A CCallOpId should have no free type variables;
741 -- when doing substitutions won't substitute over it
742 mkGlobalId (FCallId fcall) name ty info
744 occ_str = showSDoc (braces (ppr fcall <+> ppr ty))
745 -- The "occurrence name" of a ccall is the full info about the
746 -- ccall; it is encoded, but may have embedded spaces etc!
748 name = mkFCallName uniq occ_str
752 `setStrictnessInfo` Just strict_sig
754 (_, tau) = tcSplitForAllTys ty
755 (arg_tys, _) = tcSplitFunTys tau
756 arity = length arg_tys
757 strict_sig = mkStrictSig (mkTopDmdType (replicate arity evalDmd) TopRes)
759 -- Tick boxes and breakpoints are both represented as TickBoxOpIds,
760 -- except for the type:
762 -- a plain HPC tick box has type (State# RealWorld)
763 -- a breakpoint Id has type forall a.a
765 -- The breakpoint Id will be applied to a list of arbitrary free variables,
766 -- which is why it needs a polymorphic type.
768 mkTickBoxOpId :: Unique -> Module -> TickBoxId -> Id
769 mkTickBoxOpId uniq mod ix = mkTickBox' uniq mod ix realWorldStatePrimTy
771 mkBreakPointOpId :: Unique -> Module -> TickBoxId -> Id
772 mkBreakPointOpId uniq mod ix = mkTickBox' uniq mod ix ty
773 where ty = mkSigmaTy [openAlphaTyVar] [] openAlphaTy
775 mkTickBox' :: Unique -> Module -> TickBoxId -> Type -> Id
776 mkTickBox' uniq mod ix ty = mkGlobalId (TickBoxOpId tickbox) name ty info
778 tickbox = TickBox mod ix
779 occ_str = showSDoc (braces (ppr tickbox))
780 name = mkTickBoxOpName uniq occ_str
785 %************************************************************************
787 \subsection{DictFuns and default methods}
789 %************************************************************************
791 Important notes about dict funs and default methods
792 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
793 Dict funs and default methods are *not* ImplicitIds. Their definition
794 involves user-written code, so we can't figure out their strictness etc
795 based on fixed info, as we can for constructors and record selectors (say).
797 We build them as LocalIds, but with External Names. This ensures that
798 they are taken to account by free-variable finding and dependency
799 analysis (e.g. CoreFVs.exprFreeVars).
801 Why shouldn't they be bound as GlobalIds? Because, in particular, if
802 they are globals, the specialiser floats dict uses above their defns,
803 which prevents good simplifications happening. Also the strictness
804 analyser treats a occurrence of a GlobalId as imported and assumes it
805 contains strictness in its IdInfo, which isn't true if the thing is
806 bound in the same module as the occurrence.
808 It's OK for dfuns to be LocalIds, because we form the instance-env to
809 pass on to the next module (md_insts) in CoreTidy, afer tidying
810 and globalising the top-level Ids.
812 BUT make sure they are *exported* LocalIds (mkExportedLocalId) so
813 that they aren't discarded by the occurrence analyser.
816 mkDefaultMethodId :: Id -- Selector Id
817 -> Name -- Default method name
818 -> Id -- Default method Id
819 mkDefaultMethodId sel_id dm_name = mkExportedLocalId dm_name (idType sel_id)
821 mkDictFunId :: Name -- Name to use for the dict fun;
827 -- Implements the DFun Superclass Invariant (see TcInstDcls)
829 mkDictFunId dfun_name tvs theta clas tys
830 = mkExportedLocalVar (DFunId n_silent is_nt)
835 is_nt = isNewTyCon (classTyCon clas)
836 (n_silent, dfun_ty) = mkDictFunTy tvs theta clas tys
838 mkDictFunTy :: [TyVar] -> ThetaType -> Class -> [Type] -> (Int, Type)
839 mkDictFunTy tvs theta clas tys
840 = (length silent_theta, dfun_ty)
842 dfun_ty = mkSigmaTy tvs (silent_theta ++ theta) (mkDictTy clas tys)
843 silent_theta = filterOut discard $
844 substTheta (zipTopTvSubst (classTyVars clas) tys)
846 -- See Note [Silent Superclass Arguments]
847 discard pred = isEmptyVarSet (tyVarsOfPred pred)
848 || any (`eqPred` pred) theta
849 -- See the DFun Superclass Invariant in TcInstDcls
853 %************************************************************************
855 \subsection{Un-definable}
857 %************************************************************************
859 These Ids can't be defined in Haskell. They could be defined in
860 unfoldings in the wired-in GHC.Prim interface file, but we'd have to
861 ensure that they were definitely, definitely inlined, because there is
862 no curried identifier for them. That's what mkCompulsoryUnfolding
863 does. If we had a way to get a compulsory unfolding from an interface
864 file, we could do that, but we don't right now.
866 unsafeCoerce# isn't so much a PrimOp as a phantom identifier, that
867 just gets expanded into a type coercion wherever it occurs. Hence we
868 add it as a built-in Id with an unfolding here.
870 The type variables we use here are "open" type variables: this means
871 they can unify with both unlifted and lifted types. Hence we provide
872 another gun with which to shoot yourself in the foot.
875 lazyIdName, unsafeCoerceName, nullAddrName, seqName, realWorldName, coercionTokenName :: Name
876 unsafeCoerceName = mkWiredInIdName gHC_PRIM (fsLit "unsafeCoerce#") unsafeCoerceIdKey unsafeCoerceId
877 nullAddrName = mkWiredInIdName gHC_PRIM (fsLit "nullAddr#") nullAddrIdKey nullAddrId
878 seqName = mkWiredInIdName gHC_PRIM (fsLit "seq") seqIdKey seqId
879 realWorldName = mkWiredInIdName gHC_PRIM (fsLit "realWorld#") realWorldPrimIdKey realWorldPrimId
880 lazyIdName = mkWiredInIdName gHC_BASE (fsLit "lazy") lazyIdKey lazyId
881 coercionTokenName = mkWiredInIdName gHC_PRIM (fsLit "coercionToken#") coercionTokenIdKey coercionTokenId
885 ------------------------------------------------
886 -- unsafeCoerce# :: forall a b. a -> b
889 = pcMiscPrelId unsafeCoerceName ty info
891 info = noCafIdInfo `setInlinePragInfo` alwaysInlinePragma
892 `setUnfoldingInfo` mkCompulsoryUnfolding rhs
895 ty = mkForAllTys [argAlphaTyVar,openBetaTyVar]
896 (mkFunTy argAlphaTy openBetaTy)
897 [x] = mkTemplateLocals [argAlphaTy]
898 rhs = mkLams [argAlphaTyVar,openBetaTyVar,x] $
899 Cast (Var x) (mkUnsafeCo argAlphaTy openBetaTy)
901 ------------------------------------------------
903 -- nullAddr# :: Addr#
904 -- The reason is is here is because we don't provide
905 -- a way to write this literal in Haskell.
906 nullAddrId = pcMiscPrelId nullAddrName addrPrimTy info
908 info = noCafIdInfo `setInlinePragInfo` alwaysInlinePragma
909 `setUnfoldingInfo` mkCompulsoryUnfolding (Lit nullAddrLit)
911 ------------------------------------------------
912 seqId :: Id -- See Note [seqId magic]
913 seqId = pcMiscPrelId seqName ty info
915 info = noCafIdInfo `setInlinePragInfo` alwaysInlinePragma
916 `setUnfoldingInfo` mkCompulsoryUnfolding rhs
917 `setSpecInfo` mkSpecInfo [seq_cast_rule]
920 ty = mkForAllTys [alphaTyVar,argBetaTyVar]
921 (mkFunTy alphaTy (mkFunTy argBetaTy argBetaTy))
922 [x,y] = mkTemplateLocals [alphaTy, argBetaTy]
923 rhs = mkLams [alphaTyVar,argBetaTyVar,x,y] (Case (Var x) x argBetaTy [(DEFAULT, [], Var y)])
925 -- See Note [Built-in RULES for seq]
926 seq_cast_rule = BuiltinRule { ru_name = fsLit "seq of cast"
929 , ru_try = match_seq_of_cast
932 match_seq_of_cast :: IdUnfoldingFun -> [CoreExpr] -> Maybe CoreExpr
933 -- See Note [Built-in RULES for seq]
934 match_seq_of_cast _ [Type _, Type res_ty, Cast scrut co, expr]
935 = Just (Var seqId `mkApps` [Type (pFst (coercionKind co)), Type res_ty,
937 match_seq_of_cast _ _ = Nothing
939 ------------------------------------------------
940 lazyId :: Id -- See Note [lazyId magic]
941 lazyId = pcMiscPrelId lazyIdName ty info
944 ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy)
949 'GHC.Prim.seq' is special in several ways.
951 a) Its second arg can have an unboxed type
954 b) Its fixity is set in LoadIface.ghcPrimIface
956 c) It has quite a bit of desugaring magic.
957 See DsUtils.lhs Note [Desugaring seq (1)] and (2) and (3)
959 d) There is some special rule handing: Note [User-defined RULES for seq]
961 e) See Note [Typing rule for seq] in TcExpr.
963 Note [User-defined RULES for seq]
964 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
965 Roman found situations where he had
967 where he knew that f (which was strict in n) would terminate if n did.
968 Notice that the result of (f n) is discarded. So it makes sense to
972 Rather than attempt some general analysis to support this, I've added
973 enough support that you can do this using a rewrite rule:
975 RULE "f/seq" forall n. seq (f n) e = seq n e
977 You write that rule. When GHC sees a case expression that discards
978 its result, it mentally transforms it to a call to 'seq' and looks for
979 a RULE. (This is done in Simplify.rebuildCase.) As usual, the
980 correctness of the rule is up to you.
982 To make this work, we need to be careful that the magical desugaring
983 done in Note [seqId magic] item (c) is *not* done on the LHS of a rule.
984 Or rather, we arrange to un-do it, in DsBinds.decomposeRuleLhs.
986 Note [Built-in RULES for seq]
987 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
988 We also have the following built-in rule for seq
990 seq (x `cast` co) y = seq x y
992 This eliminates unnecessary casts and also allows other seq rules to
993 match more often. Notably,
995 seq (f x `cast` co) y --> seq (f x) y
997 and now a user-defined rule for seq (see Note [User-defined RULES for seq])
1003 lazy :: forall a?. a? -> a? (i.e. works for unboxed types too)
1005 Used to lazify pseq: pseq a b = a `seq` lazy b
1007 Also, no strictness: by being a built-in Id, all the info about lazyId comes from here,
1008 not from GHC.Base.hi. This is important, because the strictness
1009 analyser will spot it as strict!
1011 Also no unfolding in lazyId: it gets "inlined" by a HACK in CorePrep.
1012 It's very important to do this inlining *after* unfoldings are exposed
1013 in the interface file. Otherwise, the unfolding for (say) pseq in the
1014 interface file will not mention 'lazy', so if we inline 'pseq' we'll totally
1015 miss the very thing that 'lazy' was there for in the first place.
1016 See Trac #3259 for a real world example.
1018 lazyId is defined in GHC.Base, so we don't *have* to inline it. If it
1019 appears un-applied, we'll end up just calling it.
1021 -------------------------------------------------------------
1022 @realWorld#@ used to be a magic literal, \tr{void#}. If things get
1023 nasty as-is, change it back to a literal (@Literal@).
1025 voidArgId is a Local Id used simply as an argument in functions
1026 where we just want an arg to avoid having a thunk of unlifted type.
1028 x = \ void :: State# RealWorld -> (# p, q #)
1030 This comes up in strictness analysis
1033 realWorldPrimId :: Id
1034 realWorldPrimId -- :: State# RealWorld
1035 = pcMiscPrelId realWorldName realWorldStatePrimTy
1036 (noCafIdInfo `setUnfoldingInfo` evaldUnfolding)
1037 -- The evaldUnfolding makes it look that realWorld# is evaluated
1038 -- which in turn makes Simplify.interestingArg return True,
1039 -- which in turn makes INLINE things applied to realWorld# likely
1043 voidArgId -- :: State# RealWorld
1044 = mkSysLocal (fsLit "void") voidArgIdKey realWorldStatePrimTy
1046 coercionTokenId :: Id -- :: () ~ ()
1047 coercionTokenId -- Used to replace Coercion terms when we go to STG
1048 = pcMiscPrelId coercionTokenName
1049 (mkTyConApp eqPredPrimTyCon [unitTy, unitTy])
1055 pcMiscPrelId :: Name -> Type -> IdInfo -> Id
1056 pcMiscPrelId name ty info
1057 = mkVanillaGlobalWithInfo name ty info
1058 -- We lie and say the thing is imported; otherwise, we get into
1059 -- a mess with dependency analysis; e.g., core2stg may heave in
1060 -- random calls to GHCbase.unpackPS__. If GHCbase is the module
1061 -- being compiled, then it's just a matter of luck if the definition
1062 -- will be in "the right place" to be in scope.