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,
21 mkPrimOpId, mkFCallId,
23 mkReboxingAlt, wrapNewTypeBody, unwrapNewTypeBody,
24 mkUnpackCase, mkProductBox,
26 -- And some particular Ids; see below for why they are wired in
27 wiredInIds, ghcPrimIds,
28 unsafeCoerceId, realWorldPrimId, voidArgId, nullAddrId, seqId,
29 lazyId, lazyIdUnfolding, lazyIdKey,
32 rEC_CON_ERROR_ID, iRREFUT_PAT_ERROR_ID, rUNTIME_ERROR_ID,
33 nON_EXHAUSTIVE_GUARDS_ERROR_ID, nO_METHOD_BINDING_ERROR_ID,
34 pAT_ERROR_ID, eRROR_ID,
39 #include "HsVersions.h"
62 import Var ( Var, TyVar)
70 import BasicTypes hiding ( SuccessFlag(..) )
77 %************************************************************************
79 \subsection{Wired in Ids}
81 %************************************************************************
85 = [ -- These error-y things are wired in because we don't yet have
86 -- a way to express in an interface file that the result type variable
87 -- is 'open'; that is can be unified with an unboxed type
89 -- [The interface file format now carry such information, but there's
90 -- no way yet of expressing at the definition site for these
91 -- error-reporting functions that they have an 'open'
92 -- result type. -- sof 1/99]
94 eRROR_ID, -- This one isn't used anywhere else in the compiler
95 -- But we still need it in wiredInIds so that when GHC
96 -- compiles a program that mentions 'error' we don't
97 -- import its type from the interface file; we just get
98 -- the Id defined here. Which has an 'open-tyvar' type.
101 iRREFUT_PAT_ERROR_ID,
102 nON_EXHAUSTIVE_GUARDS_ERROR_ID,
103 nO_METHOD_BINDING_ERROR_ID,
110 -- These Ids are exported from GHC.Prim
112 = [ -- These can't be defined in Haskell, but they have
113 -- perfectly reasonable unfoldings in Core
121 %************************************************************************
123 \subsection{Data constructors}
125 %************************************************************************
127 The wrapper for a constructor is an ordinary top-level binding that evaluates
128 any strict args, unboxes any args that are going to be flattened, and calls
131 We're going to build a constructor that looks like:
133 data (Data a, C b) => T a b = T1 !a !Int b
136 \d1::Data a, d2::C b ->
137 \p q r -> case p of { p ->
139 Con T1 [a,b] [p,q,r]}}
143 * d2 is thrown away --- a context in a data decl is used to make sure
144 one *could* construct dictionaries at the site the constructor
145 is used, but the dictionary isn't actually used.
147 * We have to check that we can construct Data dictionaries for
148 the types a and Int. Once we've done that we can throw d1 away too.
150 * We use (case p of q -> ...) to evaluate p, rather than "seq" because
151 all that matters is that the arguments are evaluated. "seq" is
152 very careful to preserve evaluation order, which we don't need
155 You might think that we could simply give constructors some strictness
156 info, like PrimOps, and let CoreToStg do the let-to-case transformation.
157 But we don't do that because in the case of primops and functions strictness
158 is a *property* not a *requirement*. In the case of constructors we need to
159 do something active to evaluate the argument.
161 Making an explicit case expression allows the simplifier to eliminate
162 it in the (common) case where the constructor arg is already evaluated.
164 [Wrappers for data instance tycons]
165 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
166 In the case of data instances, the wrapper also applies the coercion turning
167 the representation type into the family instance type to cast the result of
168 the wrapper. For example, consider the declarations
170 data family Map k :: * -> *
171 data instance Map (a, b) v = MapPair (Map a (Pair b v))
173 The tycon to which the datacon MapPair belongs gets a unique internal name of
174 the form :R123Map, and we call it the representation tycon. In contrast, Map
175 is the family tycon (accessible via tyConFamInst_maybe). The wrapper and work
176 of MapPair get the types
178 $WMapPair :: forall a b v. Map a (Map a b v) -> Map (a, b) v
179 $wMapPair :: forall a b v. Map a (Map a b v) -> :R123Map a b v
181 which implies that the wrapper code will have to apply the coercion moving
182 between representation and family type. It is accessible via
183 tyConFamilyCoercion_maybe and has kind
185 Co123Map a b v :: {Map (a, b) v :=: :R123Map a b v}
187 This coercion is conditionally applied by wrapFamInstBody.
190 mkDataConIds :: Name -> Name -> DataCon -> DataConIds
191 mkDataConIds wrap_name wkr_name data_con
193 = DCIds Nothing nt_work_id -- Newtype, only has a worker
195 | any isMarkedStrict all_strict_marks -- Algebraic, needs wrapper
196 || not (null eq_spec) -- NB: LoadIface.ifaceDeclSubBndrs
197 || isFamInstTyCon tycon -- depends on this test
198 = DCIds (Just alg_wrap_id) wrk_id
200 | otherwise -- Algebraic, no wrapper
201 = DCIds Nothing wrk_id
203 (univ_tvs, ex_tvs, eq_spec,
204 theta, orig_arg_tys) = dataConFullSig data_con
205 tycon = dataConTyCon data_con
207 ----------- Wrapper --------------
208 -- We used to include the stupid theta in the wrapper's args
209 -- but now we don't. Instead the type checker just injects these
210 -- extra constraints where necessary.
211 wrap_tvs = (univ_tvs `minusList` map fst eq_spec) ++ ex_tvs
212 subst = mkTopTvSubst eq_spec
213 famSubst = ASSERT( length (tyConTyVars tycon ) ==
214 length (mkTyVarTys univ_tvs) )
215 zipTopTvSubst (tyConTyVars tycon) (mkTyVarTys univ_tvs)
216 -- substitution mapping the type constructor's type
217 -- arguments to the universals of the data constructor
218 -- (crucial when type checking interfaces)
219 dict_tys = mkPredTys theta
220 result_ty_args = map (substTyVar subst) univ_tvs
221 result_ty = case tyConFamInst_maybe tycon of
222 -- ordinary constructor
223 Nothing -> mkTyConApp tycon result_ty_args
224 -- family instance constructor
227 mkTyConApp familyTyCon ( substTys subst
230 wrap_ty = mkForAllTys wrap_tvs $ mkFunTys dict_tys $
231 mkFunTys orig_arg_tys $ result_ty
232 -- NB: watch out here if you allow user-written equality
233 -- constraints in data constructor signatures
235 ----------- Worker (algebraic data types only) --------------
236 -- The *worker* for the data constructor is the function that
237 -- takes the representation arguments and builds the constructor.
238 wrk_id = mkGlobalId (DataConWorkId data_con) wkr_name
239 (dataConRepType data_con) wkr_info
241 wkr_arity = dataConRepArity data_con
242 wkr_info = noCafIdInfo
243 `setArityInfo` wkr_arity
244 `setAllStrictnessInfo` Just wkr_sig
245 `setUnfoldingInfo` evaldUnfolding -- Record that it's evaluated,
248 wkr_sig = mkStrictSig (mkTopDmdType (replicate wkr_arity topDmd) cpr_info)
249 -- Notice that we do *not* say the worker is strict
250 -- even if the data constructor is declared strict
251 -- e.g. data T = MkT !(Int,Int)
252 -- Why? Because the *wrapper* is strict (and its unfolding has case
253 -- expresssions that do the evals) but the *worker* itself is not.
254 -- If we pretend it is strict then when we see
255 -- case x of y -> $wMkT y
256 -- the simplifier thinks that y is "sure to be evaluated" (because
257 -- $wMkT is strict) and drops the case. No, $wMkT is not strict.
259 -- When the simplifer sees a pattern
260 -- case e of MkT x -> ...
261 -- it uses the dataConRepStrictness of MkT to mark x as evaluated;
262 -- but that's fine... dataConRepStrictness comes from the data con
263 -- not from the worker Id.
265 cpr_info | isProductTyCon tycon &&
268 wkr_arity <= mAX_CPR_SIZE = retCPR
270 -- RetCPR is only true for products that are real data types;
271 -- that is, not unboxed tuples or [non-recursive] newtypes
273 ----------- Workers for newtypes --------------
274 nt_work_id = mkGlobalId (DataConWrapId data_con) wkr_name wrap_ty nt_work_info
275 nt_work_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
276 `setArityInfo` 1 -- Arity 1
277 `setUnfoldingInfo` newtype_unf
278 newtype_unf = ASSERT( isVanillaDataCon data_con &&
279 isSingleton orig_arg_tys )
280 -- No existentials on a newtype, but it can have a context
281 -- e.g. newtype Eq a => T a = MkT (...)
282 mkCompulsoryUnfolding $
283 mkLams wrap_tvs $ Lam id_arg1 $
284 wrapNewTypeBody tycon result_ty_args
287 id_arg1 = mkTemplateLocal 1 (head orig_arg_tys)
289 ----------- Wrappers for algebraic data types --------------
290 alg_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty alg_wrap_info
291 alg_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
292 `setArityInfo` alg_arity
293 -- It's important to specify the arity, so that partial
294 -- applications are treated as values
295 `setUnfoldingInfo` alg_unf
296 `setAllStrictnessInfo` Just wrap_sig
298 all_strict_marks = dataConExStricts data_con ++ dataConStrictMarks data_con
299 wrap_sig = mkStrictSig (mkTopDmdType arg_dmds cpr_info)
300 arg_dmds = map mk_dmd all_strict_marks
301 mk_dmd str | isMarkedStrict str = evalDmd
302 | otherwise = lazyDmd
303 -- The Cpr info can be important inside INLINE rhss, where the
304 -- wrapper constructor isn't inlined.
305 -- And the argument strictness can be important too; we
306 -- may not inline a contructor when it is partially applied.
308 -- data W = C !Int !Int !Int
309 -- ...(let w = C x in ...(w p q)...)...
310 -- we want to see that w is strict in its two arguments
312 alg_unf = mkTopUnfolding $ Note InlineMe $
314 mkLams dict_args $ mkLams id_args $
315 foldr mk_case con_app
316 (zip (dict_args ++ id_args) all_strict_marks)
319 con_app _ rep_ids = wrapFamInstBody tycon result_ty_args $
320 Var wrk_id `mkTyApps` result_ty_args
322 `mkTyApps` map snd eq_spec
323 `mkVarApps` reverse rep_ids
325 (dict_args,i2) = mkLocals 1 dict_tys
326 (id_args,i3) = mkLocals i2 orig_arg_tys
330 :: (Id, StrictnessMark) -- Arg, strictness
331 -> (Int -> [Id] -> CoreExpr) -- Body
332 -> Int -- Next rep arg id
333 -> [Id] -- Rep args so far, reversed
335 mk_case (arg,strict) body i rep_args
337 NotMarkedStrict -> body i (arg:rep_args)
339 | isUnLiftedType (idType arg) -> body i (arg:rep_args)
341 Case (Var arg) arg result_ty [(DEFAULT,[], body i (arg:rep_args))]
344 -> unboxProduct i (Var arg) (idType arg) the_body
346 the_body i con_args = body i (reverse con_args ++ rep_args)
348 mAX_CPR_SIZE :: Arity
350 -- We do not treat very big tuples as CPR-ish:
351 -- a) for a start we get into trouble because there aren't
352 -- "enough" unboxed tuple types (a tiresome restriction,
354 -- b) more importantly, big unboxed tuples get returned mainly
355 -- on the stack, and are often then allocated in the heap
356 -- by the caller. So doing CPR for them may in fact make
359 mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n)
363 -- If the type constructor is a representation type of a data instance, wrap
364 -- the expression into a cast adjusting the expression type, which is an
365 -- instance of the representation type, to the corresponding instance of the
366 -- family instance type.
368 wrapFamInstBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
369 wrapFamInstBody tycon args result_expr
370 | Just co_con <- tyConFamilyCoercion_maybe tycon
371 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) result_expr
377 %************************************************************************
379 \subsection{Record selectors}
381 %************************************************************************
383 We're going to build a record selector unfolding that looks like this:
385 data T a b c = T1 { ..., op :: a, ...}
386 | T2 { ..., op :: a, ...}
389 sel = /\ a b c -> \ d -> case d of
394 Similarly for newtypes
396 newtype N a = MkN { unN :: a->a }
399 unN n = coerce (a->a) n
401 We need to take a little care if the field has a polymorphic type:
403 data R = R { f :: forall a. a->a }
407 f :: forall a. R -> a -> a
408 f = /\ a \ r = case r of
411 (not f :: R -> forall a. a->a, which gives the type inference mechanism
412 problems at call sites)
414 Similarly for (recursive) newtypes
416 newtype N = MkN { unN :: forall a. a->a }
418 unN :: forall b. N -> b -> b
419 unN = /\b -> \n:N -> (coerce (forall a. a->a) n)
422 Note [Naughty record selectors]
423 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
424 A "naughty" field is one for which we can't define a record
425 selector, because an existential type variable would escape. For example:
426 data T = forall a. MkT { x,y::a }
427 We obviously can't define
429 Nevertheless we *do* put a RecordSelId into the type environment
430 so that if the user tries to use 'x' as a selector we can bleat
431 helpfully, rather than saying unhelpfully that 'x' is not in scope.
432 Hence the sel_naughty flag, to identify record selectors that don't really exist.
434 In general, a field is naughty if its type mentions a type variable that
435 isn't in the result type of the constructor.
437 Note [GADT record selectors]
438 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
439 For GADTs, we require that all constructors with a common field 'f' have the same
440 result type (modulo alpha conversion). [Checked in TcTyClsDecls.checkValidTyCon]
443 T1 { f :: a } :: T [a]
444 T2 { f :: a, y :: b } :: T [a]
445 and now the selector takes that type as its argument:
446 f :: forall a. T [a] -> a
450 Note the forall'd tyvars of the selector are just the free tyvars
451 of the result type; there may be other tyvars in the constructor's
452 type (e.g. 'b' in T2).
456 -- Steps for handling "naughty" vs "non-naughty" selectors:
457 -- 1. Determine naughtiness by comparing field type vs result type
458 -- 2. Install naughty ones with selector_ty of type _|_ and fill in mzero for info
459 -- 3. If it's not naughty, do the normal plan.
461 mkRecordSelId :: TyCon -> FieldLabel -> Id
462 mkRecordSelId tycon field_label
463 -- Assumes that all fields with the same field label have the same type
464 | is_naughty = naughty_id
467 is_naughty = not (tyVarsOfType field_ty `subVarSet` res_tv_set)
468 sel_id_details = RecordSelId tycon field_label is_naughty
470 -- Escapist case here for naughty construcotrs
471 -- We give it no IdInfo, and a type of forall a.a (never looked at)
472 naughty_id = mkGlobalId sel_id_details field_label forall_a_a noCafIdInfo
473 forall_a_a = mkForAllTy alphaTyVar (mkTyVarTy alphaTyVar)
475 -- Normal case starts here
476 sel_id = mkGlobalId sel_id_details field_label selector_ty info
477 data_cons = tyConDataCons tycon
478 data_cons_w_field = filter has_field data_cons -- Can't be empty!
479 has_field con = field_label `elem` dataConFieldLabels con
481 con1 = head data_cons_w_field
482 res_tys = dataConResTys con1
483 res_tv_set = tyVarsOfTypes res_tys
484 res_tvs = varSetElems res_tv_set
485 data_ty = mkTyConApp tycon res_tys
486 field_ty = dataConFieldType con1 field_label
488 -- *Very* tiresomely, the selectors are (unnecessarily!) overloaded over
489 -- just the dictionaries in the types of the constructors that contain
490 -- the relevant field. [The Report says that pattern matching on a
491 -- constructor gives the same constraints as applying it.] Urgh.
493 -- However, not all data cons have all constraints (because of
494 -- BuildTyCl.mkDataConStupidTheta). So we need to find all the data cons
495 -- involved in the pattern match and take the union of their constraints.
496 stupid_dict_tys = mkPredTys (dataConsStupidTheta data_cons_w_field)
497 n_stupid_dicts = length stupid_dict_tys
499 (field_tyvars,pre_field_theta,field_tau) = tcSplitSigmaTy field_ty
501 field_theta = filter (not . isEqPred) pre_field_theta
502 field_dict_tys = mkPredTys field_theta
503 n_field_dict_tys = length field_dict_tys
504 -- If the field has a universally quantified type we have to
505 -- be a bit careful. Suppose we have
506 -- data R = R { op :: forall a. Foo a => a -> a }
507 -- Then we can't give op the type
508 -- op :: R -> forall a. Foo a => a -> a
509 -- because the typechecker doesn't understand foralls to the
510 -- right of an arrow. The "right" type to give it is
511 -- op :: forall a. Foo a => R -> a -> a
512 -- But then we must generate the right unfolding too:
513 -- op = /\a -> \dfoo -> \ r ->
516 -- Note that this is exactly the type we'd infer from a user defn
520 selector_ty = mkForAllTys res_tvs $ mkForAllTys field_tyvars $
521 mkFunTys stupid_dict_tys $ mkFunTys field_dict_tys $
522 mkFunTy data_ty field_tau
524 arity = 1 + n_stupid_dicts + n_field_dict_tys
526 (strict_sig, rhs_w_str) = dmdAnalTopRhs sel_rhs
527 -- Use the demand analyser to work out strictness.
528 -- With all this unpackery it's not easy!
531 `setCafInfo` caf_info
533 `setUnfoldingInfo` mkTopUnfolding rhs_w_str
534 `setAllStrictnessInfo` Just strict_sig
536 -- Allocate Ids. We do it a funny way round because field_dict_tys is
537 -- almost always empty. Also note that we use max_dict_tys
538 -- rather than n_dict_tys, because the latter gives an infinite loop:
539 -- n_dict tys depends on the_alts, which depens on arg_ids, which depends
540 -- on arity, which depends on n_dict tys. Sigh! Mega sigh!
541 stupid_dict_ids = mkTemplateLocalsNum 1 stupid_dict_tys
542 max_stupid_dicts = length (tyConStupidTheta tycon)
543 field_dict_base = max_stupid_dicts + 1
544 field_dict_ids = mkTemplateLocalsNum field_dict_base field_dict_tys
545 dict_id_base = field_dict_base + n_field_dict_tys
546 data_id = mkTemplateLocal dict_id_base data_ty
547 arg_base = dict_id_base + 1
549 the_alts :: [CoreAlt]
550 the_alts = map mk_alt data_cons_w_field -- Already sorted by data-con
551 no_default = length data_cons == length data_cons_w_field -- No default needed
553 default_alt | no_default = []
554 | otherwise = [(DEFAULT, [], error_expr)]
556 -- The default branch may have CAF refs, because it calls recSelError etc.
557 caf_info | no_default = NoCafRefs
558 | otherwise = MayHaveCafRefs
560 sel_rhs = mkLams res_tvs $ mkLams field_tyvars $
561 mkLams stupid_dict_ids $ mkLams field_dict_ids $
562 Lam data_id $ mk_result sel_body
564 -- NB: A newtype always has a vanilla DataCon; no existentials etc
565 -- res_tys will simply be the dataConUnivTyVars
566 sel_body | isNewTyCon tycon = unwrapNewTypeBody tycon res_tys (Var data_id)
567 | otherwise = Case (Var data_id) data_id field_ty (default_alt ++ the_alts)
569 mk_result poly_result = mkVarApps (mkVarApps poly_result field_tyvars) field_dict_ids
570 -- We pull the field lambdas to the top, so we need to
571 -- apply them in the body. For example:
572 -- data T = MkT { foo :: forall a. a->a }
574 -- foo :: forall a. T -> a -> a
575 -- foo = /\a. \t:T. case t of { MkT f -> f a }
578 = ASSERT2( res_ty `tcEqType` field_ty, ppr data_con $$ ppr res_ty $$ ppr field_ty )
579 mkReboxingAlt rebox_uniqs data_con (ex_tvs ++ co_tvs ++ arg_vs) rhs
581 -- get pattern binders with types appropriately instantiated
582 arg_uniqs = map mkBuiltinUnique [arg_base..]
583 (ex_tvs, co_tvs, arg_vs) = dataConOrigInstPat arg_uniqs data_con res_tys
585 rebox_base = arg_base + length ex_tvs + length co_tvs + length arg_vs
586 rebox_uniqs = map mkBuiltinUnique [rebox_base..]
588 -- data T :: *->* where T1 { fld :: Maybe b } -> T [b]
589 -- Hence T1 :: forall a b. (a=[b]) => b -> T a
590 -- fld :: forall b. T [b] -> Maybe b
591 -- fld = /\b.\(t:T[b]). case t of
592 -- T1 b' (c : [b]=[b']) (x:Maybe b')
593 -- -> x `cast` Maybe (sym (right c))
596 -- Generate the refinement for b'=b,
597 -- and apply to (Maybe b'), to get (Maybe b)
598 Succeeded refinement = gadtRefine emptyRefinement ex_tvs co_tvs
599 the_arg_id_ty = idType the_arg_id
600 (rhs, res_ty) = case refineType refinement the_arg_id_ty of
601 Just (co, res_ty) -> (Cast (Var the_arg_id) co, res_ty)
602 Nothing -> (Var the_arg_id, the_arg_id_ty)
604 field_vs = filter (not . isPredTy . idType) arg_vs
605 the_arg_id = assoc "mkRecordSelId:mk_alt" (field_lbls `zip` field_vs) field_label
606 field_lbls = dataConFieldLabels data_con
608 error_expr = mkRuntimeErrorApp rEC_SEL_ERROR_ID field_ty full_msg
609 full_msg = showSDoc (sep [text "No match in record selector", ppr sel_id])
611 -- unbox a product type...
612 -- we will recurse into newtypes, casting along the way, and unbox at the
613 -- first product data constructor we find. e.g.
615 -- data PairInt = PairInt Int Int
616 -- newtype S = MkS PairInt
619 -- If we have e = MkT (MkS (PairInt 0 1)) and some body expecting a list of
620 -- ids, we get (modulo int passing)
622 -- case (e `cast` CoT) `cast` CoS of
623 -- PairInt a b -> body [a,b]
625 -- The Ints passed around are just for creating fresh locals
626 unboxProduct :: Int -> CoreExpr -> Type -> (Int -> [Id] -> CoreExpr) -> CoreExpr
627 unboxProduct i arg arg_ty body
630 result = mkUnpackCase the_id arg con_args boxing_con rhs
631 (_tycon, _tycon_args, boxing_con, tys) = deepSplitProductType "unboxProduct" arg_ty
632 ([the_id], i') = mkLocals i [arg_ty]
633 (con_args, i'') = mkLocals i' tys
634 rhs = body i'' con_args
636 mkUnpackCase :: Id -> CoreExpr -> [Id] -> DataCon -> CoreExpr -> CoreExpr
637 -- (mkUnpackCase x e args Con body)
639 -- case (e `cast` ...) of bndr { Con args -> body }
641 -- the type of the bndr passed in is irrelevent
642 mkUnpackCase bndr arg unpk_args boxing_con body
643 = Case cast_arg (setIdType bndr bndr_ty) (exprType body) [(DataAlt boxing_con, unpk_args, body)]
645 (cast_arg, bndr_ty) = go (idType bndr) arg
647 | (tycon, tycon_args, _, _) <- splitProductType "mkUnpackCase" ty
648 , isNewTyCon tycon && not (isRecursiveTyCon tycon)
649 = go (newTyConInstRhs tycon tycon_args)
650 (unwrapNewTypeBody tycon tycon_args arg)
651 | otherwise = (arg, ty)
654 reboxProduct :: [Unique] -- uniques to create new local binders
655 -> Type -- type of product to box
656 -> ([Unique], -- remaining uniques
657 CoreExpr, -- boxed product
658 [Id]) -- Ids being boxed into product
661 (_tycon, _tycon_args, _pack_con, con_arg_tys) = deepSplitProductType "reboxProduct" ty
663 us' = dropList con_arg_tys us
665 arg_ids = zipWith (mkSysLocal FSLIT("rb")) us con_arg_tys
667 bind_rhs = mkProductBox arg_ids ty
670 (us', bind_rhs, arg_ids)
672 mkProductBox :: [Id] -> Type -> CoreExpr
673 mkProductBox arg_ids ty
676 (tycon, tycon_args, pack_con, _con_arg_tys) = splitProductType "mkProductBox" ty
679 | isNewTyCon tycon && not (isRecursiveTyCon tycon)
680 = wrap (mkProductBox arg_ids (newTyConInstRhs tycon tycon_args))
681 | otherwise = mkConApp pack_con (map Type tycon_args ++ map Var arg_ids)
683 wrap expr = wrapNewTypeBody tycon tycon_args expr
686 -- (mkReboxingAlt us con xs rhs) basically constructs the case
687 -- alternative (con, xs, rhs)
688 -- but it does the reboxing necessary to construct the *source*
689 -- arguments, xs, from the representation arguments ys.
691 -- data T = MkT !(Int,Int) Bool
693 -- mkReboxingAlt MkT [x,b] r
694 -- = (DataAlt MkT, [y::Int,z::Int,b], let x = (y,z) in r)
696 -- mkDataAlt should really be in DataCon, but it can't because
697 -- it manipulates CoreSyn.
700 :: [Unique] -- Uniques for the new Ids
702 -> [Var] -- Source-level args, including existential dicts
706 mkReboxingAlt us con args rhs
707 | not (any isMarkedUnboxed stricts)
708 = (DataAlt con, args, rhs)
712 (binds, args') = go args stricts us
714 (DataAlt con, args', mkLets binds rhs)
717 stricts = dataConExStricts con ++ dataConStrictMarks con
719 go [] _stricts _us = ([], [])
721 -- Type variable case
722 go (arg:args) stricts us
724 = let (binds, args') = go args stricts us
725 in (binds, arg:args')
727 -- Term variable case
728 go (arg:args) (str:stricts) us
729 | isMarkedUnboxed str
731 let (binds, unpacked_args') = go args stricts us'
732 (us', bind_rhs, unpacked_args) = reboxProduct us (idType arg)
734 (NonRec arg bind_rhs : binds, unpacked_args ++ unpacked_args')
736 = let (binds, args') = go args stricts us
737 in (binds, arg:args')
741 %************************************************************************
743 \subsection{Dictionary selectors}
745 %************************************************************************
747 Selecting a field for a dictionary. If there is just one field, then
748 there's nothing to do.
750 Dictionary selectors may get nested forall-types. Thus:
753 op :: forall b. Ord b => a -> b -> b
755 Then the top-level type for op is
757 op :: forall a. Foo a =>
761 This is unlike ordinary record selectors, which have all the for-alls
762 at the outside. When dealing with classes it's very convenient to
763 recover the original type signature from the class op selector.
766 mkDictSelId :: Name -> Class -> Id
767 mkDictSelId name clas
768 = mkGlobalId (ClassOpId clas) name sel_ty info
770 sel_ty = mkForAllTys tyvars (mkFunTy (idType dict_id) (idType the_arg_id))
771 -- We can't just say (exprType rhs), because that would give a type
773 -- for a single-op class (after all, the selector is the identity)
774 -- But it's type must expose the representation of the dictionary
775 -- to gat (say) C a -> (a -> a)
779 `setUnfoldingInfo` mkTopUnfolding rhs
780 `setAllStrictnessInfo` Just strict_sig
782 -- We no longer use 'must-inline' on record selectors. They'll
783 -- inline like crazy if they scrutinise a constructor
785 -- The strictness signature is of the form U(AAAVAAAA) -> T
786 -- where the V depends on which item we are selecting
787 -- It's worth giving one, so that absence info etc is generated
788 -- even if the selector isn't inlined
789 strict_sig = mkStrictSig (mkTopDmdType [arg_dmd] TopRes)
790 arg_dmd | isNewTyCon tycon = evalDmd
791 | otherwise = Eval (Prod [ if the_arg_id == id then evalDmd else Abs
794 tycon = classTyCon clas
795 [data_con] = tyConDataCons tycon
796 tyvars = dataConUnivTyVars data_con
797 arg_tys = ASSERT( isVanillaDataCon data_con ) dataConRepArgTys data_con
798 the_arg_id = assoc "MkId.mkDictSelId" (map idName (classSelIds clas) `zip` arg_ids) name
800 pred = mkClassPred clas (mkTyVarTys tyvars)
801 (dict_id:arg_ids) = mkTemplateLocals (mkPredTy pred : arg_tys)
803 rhs = mkLams tyvars (Lam dict_id rhs_body)
804 rhs_body | isNewTyCon tycon = unwrapNewTypeBody tycon (map mkTyVarTy tyvars) (Var dict_id)
805 | otherwise = Case (Var dict_id) dict_id (idType the_arg_id)
806 [(DataAlt data_con, arg_ids, Var the_arg_id)]
808 wrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
809 -- The wrapper for the data constructor for a newtype looks like this:
810 -- newtype T a = MkT (a,Int)
811 -- MkT :: forall a. (a,Int) -> T a
812 -- MkT = /\a. \(x:(a,Int)). x `cast` sym (CoT a)
813 -- where CoT is the coercion TyCon assoicated with the newtype
815 -- The call (wrapNewTypeBody T [a] e) returns the
816 -- body of the wrapper, namely
817 -- e `cast` (CoT [a])
819 -- If a coercion constructor is prodivided in the newtype, then we use
820 -- it, otherwise the wrap/unwrap are both no-ops
822 -- If the we are dealing with a newtype instance, we have a second coercion
823 -- identifying the family instance with the constructor of the newtype
824 -- instance. This coercion is applied in any case (ie, composed with the
825 -- coercion constructor of the newtype or applied by itself).
827 wrapNewTypeBody tycon args result_expr
828 = wrapFamInstBody tycon args inner
831 | Just co_con <- newTyConCo_maybe tycon
832 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) result_expr
836 -- When unwrapping, we do *not* apply any family coercion, because this will
837 -- be done via a CoPat by the type checker. We have to do it this way as
838 -- computing the right type arguments for the coercion requires more than just
839 -- a spliting operation (cf, TcPat.tcConPat).
841 unwrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
842 unwrapNewTypeBody tycon args result_expr
843 | Just co_con <- newTyConCo_maybe tycon
844 = mkCoerce (mkTyConApp co_con args) result_expr
852 %************************************************************************
854 \subsection{Primitive operations
856 %************************************************************************
859 mkPrimOpId :: PrimOp -> Id
863 (tyvars,arg_tys,res_ty, arity, strict_sig) = primOpSig prim_op
864 ty = mkForAllTys tyvars (mkFunTys arg_tys res_ty)
865 name = mkWiredInName gHC_PRIM (primOpOcc prim_op)
866 (mkPrimOpIdUnique (primOpTag prim_op))
868 id = mkGlobalId (PrimOpId prim_op) name ty info
871 `setSpecInfo` mkSpecInfo (primOpRules prim_op name)
873 `setAllStrictnessInfo` Just strict_sig
875 -- For each ccall we manufacture a separate CCallOpId, giving it
876 -- a fresh unique, a type that is correct for this particular ccall,
877 -- and a CCall structure that gives the correct details about calling
880 -- The *name* of this Id is a local name whose OccName gives the full
881 -- details of the ccall, type and all. This means that the interface
882 -- file reader can reconstruct a suitable Id
884 mkFCallId :: Unique -> ForeignCall -> Type -> Id
885 mkFCallId uniq fcall ty
886 = ASSERT( isEmptyVarSet (tyVarsOfType ty) )
887 -- A CCallOpId should have no free type variables;
888 -- when doing substitutions won't substitute over it
889 mkGlobalId (FCallId fcall) name ty info
891 occ_str = showSDoc (braces (ppr fcall <+> ppr ty))
892 -- The "occurrence name" of a ccall is the full info about the
893 -- ccall; it is encoded, but may have embedded spaces etc!
895 name = mkFCallName uniq occ_str
899 `setAllStrictnessInfo` Just strict_sig
901 (_, tau) = tcSplitForAllTys ty
902 (arg_tys, _) = tcSplitFunTys tau
903 arity = length arg_tys
904 strict_sig = mkStrictSig (mkTopDmdType (replicate arity evalDmd) TopRes)
908 %************************************************************************
910 \subsection{DictFuns and default methods}
912 %************************************************************************
914 Important notes about dict funs and default methods
915 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
916 Dict funs and default methods are *not* ImplicitIds. Their definition
917 involves user-written code, so we can't figure out their strictness etc
918 based on fixed info, as we can for constructors and record selectors (say).
920 We build them as LocalIds, but with External Names. This ensures that
921 they are taken to account by free-variable finding and dependency
922 analysis (e.g. CoreFVs.exprFreeVars).
924 Why shouldn't they be bound as GlobalIds? Because, in particular, if
925 they are globals, the specialiser floats dict uses above their defns,
926 which prevents good simplifications happening. Also the strictness
927 analyser treats a occurrence of a GlobalId as imported and assumes it
928 contains strictness in its IdInfo, which isn't true if the thing is
929 bound in the same module as the occurrence.
931 It's OK for dfuns to be LocalIds, because we form the instance-env to
932 pass on to the next module (md_insts) in CoreTidy, afer tidying
933 and globalising the top-level Ids.
935 BUT make sure they are *exported* LocalIds (mkExportedLocalId) so
936 that they aren't discarded by the occurrence analyser.
939 mkDefaultMethodId dm_name ty = mkExportedLocalId dm_name ty
941 mkDictFunId :: Name -- Name to use for the dict fun;
948 mkDictFunId dfun_name inst_tyvars dfun_theta clas inst_tys
949 = mkExportedLocalId dfun_name dfun_ty
951 dfun_ty = mkSigmaTy inst_tyvars dfun_theta (mkDictTy clas inst_tys)
953 {- 1 dec 99: disable the Mark Jones optimisation for the sake
954 of compatibility with Hugs.
955 See `types/InstEnv' for a discussion related to this.
957 (class_tyvars, sc_theta, _, _) = classBigSig clas
958 not_const (clas, tys) = not (isEmptyVarSet (tyVarsOfTypes tys))
959 sc_theta' = substClasses (zipTopTvSubst class_tyvars inst_tys) sc_theta
960 dfun_theta = case inst_decl_theta of
961 [] -> [] -- If inst_decl_theta is empty, then we don't
962 -- want to have any dict arguments, so that we can
963 -- expose the constant methods.
965 other -> nub (inst_decl_theta ++ filter not_const sc_theta')
966 -- Otherwise we pass the superclass dictionaries to
967 -- the dictionary function; the Mark Jones optimisation.
969 -- NOTE the "nub". I got caught by this one:
970 -- class Monad m => MonadT t m where ...
971 -- instance Monad m => MonadT (EnvT env) m where ...
972 -- Here, the inst_decl_theta has (Monad m); but so
973 -- does the sc_theta'!
975 -- NOTE the "not_const". I got caught by this one too:
976 -- class Foo a => Baz a b where ...
977 -- instance Wob b => Baz T b where..
978 -- Now sc_theta' has Foo T
983 %************************************************************************
985 \subsection{Un-definable}
987 %************************************************************************
989 These Ids can't be defined in Haskell. They could be defined in
990 unfoldings in the wired-in GHC.Prim interface file, but we'd have to
991 ensure that they were definitely, definitely inlined, because there is
992 no curried identifier for them. That's what mkCompulsoryUnfolding
993 does. If we had a way to get a compulsory unfolding from an interface
994 file, we could do that, but we don't right now.
996 unsafeCoerce# isn't so much a PrimOp as a phantom identifier, that
997 just gets expanded into a type coercion wherever it occurs. Hence we
998 add it as a built-in Id with an unfolding here.
1000 The type variables we use here are "open" type variables: this means
1001 they can unify with both unlifted and lifted types. Hence we provide
1002 another gun with which to shoot yourself in the foot.
1005 mkWiredInIdName mod fs uniq id
1006 = mkWiredInName mod (mkOccNameFS varName fs) uniq (AnId id) UserSyntax
1008 unsafeCoerceName = mkWiredInIdName gHC_PRIM FSLIT("unsafeCoerce#") unsafeCoerceIdKey unsafeCoerceId
1009 nullAddrName = mkWiredInIdName gHC_PRIM FSLIT("nullAddr#") nullAddrIdKey nullAddrId
1010 seqName = mkWiredInIdName gHC_PRIM FSLIT("seq") seqIdKey seqId
1011 realWorldName = mkWiredInIdName gHC_PRIM FSLIT("realWorld#") realWorldPrimIdKey realWorldPrimId
1012 lazyIdName = mkWiredInIdName gHC_BASE FSLIT("lazy") lazyIdKey lazyId
1014 errorName = mkWiredInIdName gHC_ERR FSLIT("error") errorIdKey eRROR_ID
1015 recSelErrorName = mkWiredInIdName gHC_ERR FSLIT("recSelError") recSelErrorIdKey rEC_SEL_ERROR_ID
1016 runtimeErrorName = mkWiredInIdName gHC_ERR FSLIT("runtimeError") runtimeErrorIdKey rUNTIME_ERROR_ID
1017 irrefutPatErrorName = mkWiredInIdName gHC_ERR FSLIT("irrefutPatError") irrefutPatErrorIdKey iRREFUT_PAT_ERROR_ID
1018 recConErrorName = mkWiredInIdName gHC_ERR FSLIT("recConError") recConErrorIdKey rEC_CON_ERROR_ID
1019 patErrorName = mkWiredInIdName gHC_ERR FSLIT("patError") patErrorIdKey pAT_ERROR_ID
1020 noMethodBindingErrorName = mkWiredInIdName gHC_ERR FSLIT("noMethodBindingError")
1021 noMethodBindingErrorIdKey nO_METHOD_BINDING_ERROR_ID
1022 nonExhaustiveGuardsErrorName
1023 = mkWiredInIdName gHC_ERR FSLIT("nonExhaustiveGuardsError")
1024 nonExhaustiveGuardsErrorIdKey nON_EXHAUSTIVE_GUARDS_ERROR_ID
1028 -- unsafeCoerce# :: forall a b. a -> b
1030 = pcMiscPrelId unsafeCoerceName ty info
1032 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
1035 ty = mkForAllTys [openAlphaTyVar,openBetaTyVar]
1036 (mkFunTy openAlphaTy openBetaTy)
1037 [x] = mkTemplateLocals [openAlphaTy]
1038 rhs = mkLams [openAlphaTyVar,openBetaTyVar,x] $
1039 -- Note (Coerce openBetaTy openAlphaTy) (Var x)
1040 Cast (Var x) (mkUnsafeCoercion openAlphaTy openBetaTy)
1042 -- nullAddr# :: Addr#
1043 -- The reason is is here is because we don't provide
1044 -- a way to write this literal in Haskell.
1046 = pcMiscPrelId nullAddrName addrPrimTy info
1048 info = noCafIdInfo `setUnfoldingInfo`
1049 mkCompulsoryUnfolding (Lit nullAddrLit)
1052 = pcMiscPrelId seqName ty info
1054 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
1057 ty = mkForAllTys [alphaTyVar,openBetaTyVar]
1058 (mkFunTy alphaTy (mkFunTy openBetaTy openBetaTy))
1059 [x,y] = mkTemplateLocals [alphaTy, openBetaTy]
1060 rhs = mkLams [alphaTyVar,openBetaTyVar,x,y] (Case (Var x) x openBetaTy [(DEFAULT, [], Var y)])
1062 -- lazy :: forall a?. a? -> a? (i.e. works for unboxed types too)
1063 -- Used to lazify pseq: pseq a b = a `seq` lazy b
1065 -- Also, no strictness: by being a built-in Id, all the info about lazyId comes from here,
1066 -- not from GHC.Base.hi. This is important, because the strictness
1067 -- analyser will spot it as strict!
1069 -- Also no unfolding in lazyId: it gets "inlined" by a HACK in the worker/wrapper pass
1070 -- (see WorkWrap.wwExpr)
1071 -- We could use inline phases to do this, but that would be vulnerable to changes in
1072 -- phase numbering....we must inline precisely after strictness analysis.
1074 = pcMiscPrelId lazyIdName ty info
1077 ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy)
1079 lazyIdUnfolding :: CoreExpr -- Used to expand 'lazyId' after strictness anal
1080 lazyIdUnfolding = mkLams [openAlphaTyVar,x] (Var x)
1082 [x] = mkTemplateLocals [openAlphaTy]
1085 @realWorld#@ used to be a magic literal, \tr{void#}. If things get
1086 nasty as-is, change it back to a literal (@Literal@).
1088 voidArgId is a Local Id used simply as an argument in functions
1089 where we just want an arg to avoid having a thunk of unlifted type.
1091 x = \ void :: State# RealWorld -> (# p, q #)
1093 This comes up in strictness analysis
1096 realWorldPrimId -- :: State# RealWorld
1097 = pcMiscPrelId realWorldName realWorldStatePrimTy
1098 (noCafIdInfo `setUnfoldingInfo` evaldUnfolding)
1099 -- The evaldUnfolding makes it look that realWorld# is evaluated
1100 -- which in turn makes Simplify.interestingArg return True,
1101 -- which in turn makes INLINE things applied to realWorld# likely
1104 voidArgId -- :: State# RealWorld
1105 = mkSysLocal FSLIT("void") voidArgIdKey realWorldStatePrimTy
1109 %************************************************************************
1111 \subsection[PrelVals-error-related]{@error@ and friends; @trace@}
1113 %************************************************************************
1115 GHC randomly injects these into the code.
1117 @patError@ is just a version of @error@ for pattern-matching
1118 failures. It knows various ``codes'' which expand to longer
1119 strings---this saves space!
1121 @absentErr@ is a thing we put in for ``absent'' arguments. They jolly
1122 well shouldn't be yanked on, but if one is, then you will get a
1123 friendly message from @absentErr@ (rather than a totally random
1126 @parError@ is a special version of @error@ which the compiler does
1127 not know to be a bottoming Id. It is used in the @_par_@ and @_seq_@
1128 templates, but we don't ever expect to generate code for it.
1132 :: Id -- Should be of type (forall a. Addr# -> a)
1133 -- where Addr# points to a UTF8 encoded string
1134 -> Type -- The type to instantiate 'a'
1135 -> String -- The string to print
1138 mkRuntimeErrorApp err_id res_ty err_msg
1139 = mkApps (Var err_id) [Type res_ty, err_string]
1141 err_string = Lit (mkStringLit err_msg)
1143 rEC_SEL_ERROR_ID = mkRuntimeErrorId recSelErrorName
1144 rUNTIME_ERROR_ID = mkRuntimeErrorId runtimeErrorName
1145 iRREFUT_PAT_ERROR_ID = mkRuntimeErrorId irrefutPatErrorName
1146 rEC_CON_ERROR_ID = mkRuntimeErrorId recConErrorName
1147 pAT_ERROR_ID = mkRuntimeErrorId patErrorName
1148 nO_METHOD_BINDING_ERROR_ID = mkRuntimeErrorId noMethodBindingErrorName
1149 nON_EXHAUSTIVE_GUARDS_ERROR_ID = mkRuntimeErrorId nonExhaustiveGuardsErrorName
1151 -- The runtime error Ids take a UTF8-encoded string as argument
1152 mkRuntimeErrorId name = pc_bottoming_Id name runtimeErrorTy
1153 runtimeErrorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTy addrPrimTy openAlphaTy)
1157 eRROR_ID = pc_bottoming_Id errorName errorTy
1160 errorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTys [mkListTy charTy] openAlphaTy)
1161 -- Notice the openAlphaTyVar. It says that "error" can be applied
1162 -- to unboxed as well as boxed types. This is OK because it never
1163 -- returns, so the return type is irrelevant.
1167 %************************************************************************
1169 \subsection{Utilities}
1171 %************************************************************************
1174 pcMiscPrelId :: Name -> Type -> IdInfo -> Id
1175 pcMiscPrelId name ty info
1176 = mkVanillaGlobal name ty info
1177 -- We lie and say the thing is imported; otherwise, we get into
1178 -- a mess with dependency analysis; e.g., core2stg may heave in
1179 -- random calls to GHCbase.unpackPS__. If GHCbase is the module
1180 -- being compiled, then it's just a matter of luck if the definition
1181 -- will be in "the right place" to be in scope.
1183 pc_bottoming_Id name ty
1184 = pcMiscPrelId name ty bottoming_info
1186 bottoming_info = vanillaIdInfo `setAllStrictnessInfo` Just strict_sig
1187 -- Do *not* mark them as NoCafRefs, because they can indeed have
1188 -- CAF refs. For example, pAT_ERROR_ID calls GHC.Err.untangle,
1189 -- which has some CAFs
1190 -- In due course we may arrange that these error-y things are
1191 -- regarded by the GC as permanently live, in which case we
1192 -- can give them NoCaf info. As it is, any function that calls
1193 -- any pc_bottoming_Id will itself have CafRefs, which bloats
1196 strict_sig = mkStrictSig (mkTopDmdType [evalDmd] BotRes)
1197 -- These "bottom" out, no matter what their arguments