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, mkTickBoxOpId, mkBreakPointOpId,
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"
61 import Var ( Var, TyVar)
69 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 Note [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
174 name of the form :R123Map, and we call it the representation tycon.
175 In contrast, Map is the family tycon (accessible via
176 tyConFamInst_maybe). A coercion allows you to move between
177 representation and family type. It is accessible from :R123Map via
178 tyConFamilyCoercion_maybe and has kind
180 Co123Map a b v :: {Map (a, b) v :=: :R123Map a b v}
182 The wrapper and worker of MapPair get the types
185 $WMapPair :: forall a b v. Map a (Map a b v) -> Map (a, b) v
186 $WMapPair a b v = MapPair a b v `cast` sym (Co123Map a b v)
189 MapPair :: forall a b v. Map a (Map a b v) -> :R123Map a b v
191 This coercion is conditionally applied by wrapFamInstBody.
193 It's a bit more complicated if the data instance is a GADT as well!
195 data instance T [a] where
196 T1 :: forall b. b -> T [Maybe b]
198 Co7T a :: T [a] ~ :R7T a
203 $WT1 :: forall b. b -> T [Maybe b]
204 $WT1 b v = T1 (Maybe b) b (Maybe b) v
205 `cast` sym (Co7T (Maybe b))
208 T1 :: forall c b. (c ~ Maybe b) => b -> :R7T c
211 mkDataConIds :: Name -> Name -> DataCon -> DataConIds
212 mkDataConIds wrap_name wkr_name data_con
213 | isNewTyCon tycon -- Newtype, only has a worker
214 , not (isFamInstTyCon tycon) -- unless it's a family instancex
215 = DCIds Nothing nt_work_id
217 | any isMarkedStrict all_strict_marks -- Algebraic, needs wrapper
218 || not (null eq_spec) -- NB: LoadIface.ifaceDeclSubBndrs
219 || isFamInstTyCon tycon -- depends on this test
220 = DCIds (Just alg_wrap_id) wrk_id
222 | otherwise -- Algebraic, no wrapper
223 = DCIds Nothing wrk_id
225 (univ_tvs, ex_tvs, eq_spec,
226 theta, orig_arg_tys, res_ty) = dataConFullSig data_con
227 tycon = dataConTyCon data_con -- The representation TyCon (not family)
229 ----------- Worker (algebraic data types only) --------------
230 -- The *worker* for the data constructor is the function that
231 -- takes the representation arguments and builds the constructor.
232 wrk_id = mkGlobalId (DataConWorkId data_con) wkr_name
233 (dataConRepType data_con) wkr_info
235 wkr_arity = dataConRepArity data_con
236 wkr_info = noCafIdInfo
237 `setArityInfo` wkr_arity
238 `setAllStrictnessInfo` Just wkr_sig
239 `setUnfoldingInfo` evaldUnfolding -- Record that it's evaluated,
242 wkr_sig = mkStrictSig (mkTopDmdType (replicate wkr_arity topDmd) cpr_info)
243 -- Note [Data-con worker strictness]
244 -- Notice that we do *not* say the worker is strict
245 -- even if the data constructor is declared strict
246 -- e.g. data T = MkT !(Int,Int)
247 -- Why? Because the *wrapper* is strict (and its unfolding has case
248 -- expresssions that do the evals) but the *worker* itself is not.
249 -- If we pretend it is strict then when we see
250 -- case x of y -> $wMkT y
251 -- the simplifier thinks that y is "sure to be evaluated" (because
252 -- $wMkT is strict) and drops the case. No, $wMkT is not strict.
254 -- When the simplifer sees a pattern
255 -- case e of MkT x -> ...
256 -- it uses the dataConRepStrictness of MkT to mark x as evaluated;
257 -- but that's fine... dataConRepStrictness comes from the data con
258 -- not from the worker Id.
260 cpr_info | isProductTyCon tycon &&
263 wkr_arity <= mAX_CPR_SIZE = retCPR
265 -- RetCPR is only true for products that are real data types;
266 -- that is, not unboxed tuples or [non-recursive] newtypes
268 ----------- Workers for newtypes --------------
269 nt_work_id = mkGlobalId (DataConWrapId data_con) wkr_name wrap_ty nt_work_info
270 nt_work_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
271 `setArityInfo` 1 -- Arity 1
272 `setUnfoldingInfo` newtype_unf
273 newtype_unf = ASSERT( isVanillaDataCon data_con &&
274 isSingleton orig_arg_tys )
275 -- No existentials on a newtype, but it can have a context
276 -- e.g. newtype Eq a => T a = MkT (...)
277 mkCompulsoryUnfolding $
278 mkLams wrap_tvs $ Lam id_arg1 $
279 wrapNewTypeBody tycon res_ty_args
282 id_arg1 = mkTemplateLocal 1 (head orig_arg_tys)
284 ----------- Wrapper --------------
285 -- We used to include the stupid theta in the wrapper's args
286 -- but now we don't. Instead the type checker just injects these
287 -- extra constraints where necessary.
288 wrap_tvs = (univ_tvs `minusList` map fst eq_spec) ++ ex_tvs
289 res_ty_args = substTyVars (mkTopTvSubst eq_spec) univ_tvs
290 dict_tys = mkPredTys theta
291 wrap_ty = mkForAllTys wrap_tvs $ mkFunTys dict_tys $
292 mkFunTys orig_arg_tys $ res_ty
293 -- NB: watch out here if you allow user-written equality
294 -- constraints in data constructor signatures
296 ----------- Wrappers for algebraic data types --------------
297 alg_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty alg_wrap_info
298 alg_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo
299 `setArityInfo` wrap_arity
300 -- It's important to specify the arity, so that partial
301 -- applications are treated as values
302 `setUnfoldingInfo` wrap_unf
303 `setAllStrictnessInfo` Just wrap_sig
305 all_strict_marks = dataConExStricts data_con ++ dataConStrictMarks data_con
306 wrap_sig = mkStrictSig (mkTopDmdType arg_dmds cpr_info)
307 arg_dmds = map mk_dmd all_strict_marks
308 mk_dmd str | isMarkedStrict str = evalDmd
309 | otherwise = lazyDmd
310 -- The Cpr info can be important inside INLINE rhss, where the
311 -- wrapper constructor isn't inlined.
312 -- And the argument strictness can be important too; we
313 -- may not inline a contructor when it is partially applied.
315 -- data W = C !Int !Int !Int
316 -- ...(let w = C x in ...(w p q)...)...
317 -- we want to see that w is strict in its two arguments
319 wrap_unf = mkTopUnfolding $ Note InlineMe $
321 mkLams dict_args $ mkLams id_args $
322 foldr mk_case con_app
323 (zip (dict_args ++ id_args) all_strict_marks)
326 con_app _ rep_ids = wrapFamInstBody tycon res_ty_args $
327 Var wrk_id `mkTyApps` res_ty_args
329 `mkTyApps` map snd eq_spec -- Equality evidence
330 `mkVarApps` reverse rep_ids
332 (dict_args,i2) = mkLocals 1 dict_tys
333 (id_args,i3) = mkLocals i2 orig_arg_tys
337 :: (Id, StrictnessMark) -- Arg, strictness
338 -> (Int -> [Id] -> CoreExpr) -- Body
339 -> Int -- Next rep arg id
340 -> [Id] -- Rep args so far, reversed
342 mk_case (arg,strict) body i rep_args
344 NotMarkedStrict -> body i (arg:rep_args)
346 | isUnLiftedType (idType arg) -> body i (arg:rep_args)
348 Case (Var arg) arg res_ty [(DEFAULT,[], body i (arg:rep_args))]
351 -> unboxProduct i (Var arg) (idType arg) the_body
353 the_body i con_args = body i (reverse con_args ++ rep_args)
355 mAX_CPR_SIZE :: Arity
357 -- We do not treat very big tuples as CPR-ish:
358 -- a) for a start we get into trouble because there aren't
359 -- "enough" unboxed tuple types (a tiresome restriction,
361 -- b) more importantly, big unboxed tuples get returned mainly
362 -- on the stack, and are often then allocated in the heap
363 -- by the caller. So doing CPR for them may in fact make
366 mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n)
372 %************************************************************************
374 \subsection{Record selectors}
376 %************************************************************************
378 We're going to build a record selector unfolding that looks like this:
380 data T a b c = T1 { ..., op :: a, ...}
381 | T2 { ..., op :: a, ...}
384 sel = /\ a b c -> \ d -> case d of
389 Similarly for newtypes
391 newtype N a = MkN { unN :: a->a }
394 unN n = coerce (a->a) n
396 We need to take a little care if the field has a polymorphic type:
398 data R = R { f :: forall a. a->a }
402 f :: forall a. R -> a -> a
403 f = /\ a \ r = case r of
406 (not f :: R -> forall a. a->a, which gives the type inference mechanism
407 problems at call sites)
409 Similarly for (recursive) newtypes
411 newtype N = MkN { unN :: forall a. a->a }
413 unN :: forall b. N -> b -> b
414 unN = /\b -> \n:N -> (coerce (forall a. a->a) n)
417 Note [Naughty record selectors]
418 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
419 A "naughty" field is one for which we can't define a record
420 selector, because an existential type variable would escape. For example:
421 data T = forall a. MkT { x,y::a }
422 We obviously can't define
424 Nevertheless we *do* put a RecordSelId into the type environment
425 so that if the user tries to use 'x' as a selector we can bleat
426 helpfully, rather than saying unhelpfully that 'x' is not in scope.
427 Hence the sel_naughty flag, to identify record selectors that don't really exist.
429 In general, a field is naughty if its type mentions a type variable that
430 isn't in the result type of the constructor.
432 Note [GADT record selectors]
433 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
434 For GADTs, we require that all constructors with a common field 'f' have the same
435 result type (modulo alpha conversion). [Checked in TcTyClsDecls.checkValidTyCon]
438 T1 { f :: a } :: T [a]
439 T2 { f :: a, y :: b } :: T [a]
440 and now the selector takes that type as its argument:
441 f :: forall a. T [a] -> a
445 Note the forall'd tyvars of the selector are just the free tyvars
446 of the result type; there may be other tyvars in the constructor's
447 type (e.g. 'b' in T2).
449 Note [Selector running example]
450 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
451 It's OK to combine GADTs and type families. Here's a running example:
453 data instance T [a] where
454 T1 { fld :: b } :: T [Maybe b]
456 The representation type looks like this
458 T1 { fld :: b } :: :R7T (Maybe b)
460 and there's coercion from the family type to the representation type
461 :CoR7T a :: T [a] ~ :R7T a
463 The selector we want for fld looks like this:
465 fld :: forall b. T [Maybe b] -> b
466 fld = /\b. \(d::T [Maybe b]).
467 case d `cast` :CoR7T (Maybe b) of
470 The scrutinee of the case has type :R7T (Maybe b), which can be
471 gotten by appying the eq_spec to the univ_tvs of the data con.
474 mkRecordSelId :: TyCon -> FieldLabel -> Id
475 mkRecordSelId tycon field_label
476 -- Assumes that all fields with the same field label have the same type
477 | is_naughty = naughty_id
480 is_naughty = not (tyVarsOfType field_ty `subVarSet` data_tv_set)
481 sel_id_details = RecordSelId tycon field_label is_naughty
483 -- Escapist case here for naughty constructors
484 -- We give it no IdInfo, and a type of forall a.a (never looked at)
485 naughty_id = mkGlobalId sel_id_details field_label forall_a_a noCafIdInfo
486 forall_a_a = mkForAllTy alphaTyVar (mkTyVarTy alphaTyVar)
488 -- Normal case starts here
489 sel_id = mkGlobalId sel_id_details field_label selector_ty info
490 data_cons = tyConDataCons tycon
491 data_cons_w_field = filter has_field data_cons -- Can't be empty!
492 has_field con = field_label `elem` dataConFieldLabels con
494 con1 = head data_cons_w_field
495 (univ_tvs, _, eq_spec, _, _, data_ty) = dataConFullSig con1
496 data_tv_set = tyVarsOfType data_ty
497 data_tvs = varSetElems data_tv_set
498 field_ty = dataConFieldType con1 field_label
500 -- *Very* tiresomely, the selectors are (unnecessarily!) overloaded over
501 -- just the dictionaries in the types of the constructors that contain
502 -- the relevant field. [The Report says that pattern matching on a
503 -- constructor gives the same constraints as applying it.] Urgh.
505 -- However, not all data cons have all constraints (because of
506 -- BuildTyCl.mkDataConStupidTheta). So we need to find all the data cons
507 -- involved in the pattern match and take the union of their constraints.
508 stupid_dict_tys = mkPredTys (dataConsStupidTheta data_cons_w_field)
509 n_stupid_dicts = length stupid_dict_tys
511 (field_tyvars,pre_field_theta,field_tau) = tcSplitSigmaTy field_ty
512 field_theta = filter (not . isEqPred) pre_field_theta
513 field_dict_tys = mkPredTys field_theta
514 n_field_dict_tys = length field_dict_tys
515 -- If the field has a universally quantified type we have to
516 -- be a bit careful. Suppose we have
517 -- data R = R { op :: forall a. Foo a => a -> a }
518 -- Then we can't give op the type
519 -- op :: R -> forall a. Foo a => a -> a
520 -- because the typechecker doesn't understand foralls to the
521 -- right of an arrow. The "right" type to give it is
522 -- op :: forall a. Foo a => R -> a -> a
523 -- But then we must generate the right unfolding too:
524 -- op = /\a -> \dfoo -> \ r ->
527 -- Note that this is exactly the type we'd infer from a user defn
531 selector_ty = mkForAllTys data_tvs $ mkForAllTys field_tyvars $
532 mkFunTys stupid_dict_tys $ mkFunTys field_dict_tys $
533 mkFunTy data_ty field_tau
535 arity = 1 + n_stupid_dicts + n_field_dict_tys
537 (strict_sig, rhs_w_str) = dmdAnalTopRhs sel_rhs
538 -- Use the demand analyser to work out strictness.
539 -- With all this unpackery it's not easy!
542 `setCafInfo` caf_info
544 `setUnfoldingInfo` mkTopUnfolding rhs_w_str
545 `setAllStrictnessInfo` Just strict_sig
547 -- Allocate Ids. We do it a funny way round because field_dict_tys is
548 -- almost always empty. Also note that we use max_dict_tys
549 -- rather than n_dict_tys, because the latter gives an infinite loop:
550 -- n_dict tys depends on the_alts, which depens on arg_ids, which depends
551 -- on arity, which depends on n_dict tys. Sigh! Mega sigh!
552 stupid_dict_ids = mkTemplateLocalsNum 1 stupid_dict_tys
553 max_stupid_dicts = length (tyConStupidTheta tycon)
554 field_dict_base = max_stupid_dicts + 1
555 field_dict_ids = mkTemplateLocalsNum field_dict_base field_dict_tys
556 dict_id_base = field_dict_base + n_field_dict_tys
557 data_id = mkTemplateLocal dict_id_base data_ty
558 scrut_id = mkTemplateLocal (dict_id_base+1) scrut_ty
559 arg_base = dict_id_base + 2
561 the_alts :: [CoreAlt]
562 the_alts = map mk_alt data_cons_w_field -- Already sorted by data-con
563 no_default = length data_cons == length data_cons_w_field -- No default needed
565 default_alt | no_default = []
566 | otherwise = [(DEFAULT, [], error_expr)]
568 -- The default branch may have CAF refs, because it calls recSelError etc.
569 caf_info | no_default = NoCafRefs
570 | otherwise = MayHaveCafRefs
572 sel_rhs = mkLams data_tvs $ mkLams field_tyvars $
573 mkLams stupid_dict_ids $ mkLams field_dict_ids $
574 Lam data_id $ mk_result sel_body
576 scrut_ty_args = substTyVars (mkTopTvSubst eq_spec) univ_tvs
577 scrut_ty = mkTyConApp tycon scrut_ty_args
578 scrut = unwrapFamInstScrut tycon scrut_ty_args (Var data_id)
579 -- First coerce from the type family to the representation type
581 -- NB: A newtype always has a vanilla DataCon; no existentials etc
582 -- data_tys will simply be the dataConUnivTyVars
583 sel_body | isNewTyCon tycon = unwrapNewTypeBody tycon scrut_ty_args scrut
584 | otherwise = Case scrut scrut_id field_ty (default_alt ++ the_alts)
586 mk_result poly_result = mkVarApps (mkVarApps poly_result field_tyvars) field_dict_ids
587 -- We pull the field lambdas to the top, so we need to
588 -- apply them in the body. For example:
589 -- data T = MkT { foo :: forall a. a->a }
591 -- foo :: forall a. T -> a -> a
592 -- foo = /\a. \t:T. case t of { MkT f -> f a }
595 = ASSERT2( data_ty `tcEqType` field_ty, ppr data_con $$ ppr data_ty $$ ppr field_ty )
596 mkReboxingAlt rebox_uniqs data_con (ex_tvs ++ co_tvs ++ arg_vs) rhs
598 -- get pattern binders with types appropriately instantiated
599 arg_uniqs = map mkBuiltinUnique [arg_base..]
600 (ex_tvs, co_tvs, arg_vs) = dataConOrigInstPat arg_uniqs data_con scrut_ty_args
602 rebox_base = arg_base + length ex_tvs + length co_tvs + length arg_vs
603 rebox_uniqs = map mkBuiltinUnique [rebox_base..]
605 -- data T :: *->* where T1 { fld :: Maybe b } -> T [b]
606 -- Hence T1 :: forall a b. (a=[b]) => b -> T a
607 -- fld :: forall b. T [b] -> Maybe b
608 -- fld = /\b.\(t:T[b]). case t of
609 -- T1 b' (c : [b]=[b']) (x:Maybe b')
610 -- -> x `cast` Maybe (sym (right c))
613 -- Generate the refinement for b'=b,
614 -- and apply to (Maybe b'), to get (Maybe b)
615 Succeeded refinement = gadtRefine emptyRefinement ex_tvs co_tvs
616 the_arg_id_ty = idType the_arg_id
617 (rhs, data_ty) = case refineType refinement the_arg_id_ty of
618 Just (co, data_ty) -> (Cast (Var the_arg_id) co, data_ty)
619 Nothing -> (Var the_arg_id, the_arg_id_ty)
621 field_vs = filter (not . isPredTy . idType) arg_vs
622 the_arg_id = assoc "mkRecordSelId:mk_alt" (field_lbls `zip` field_vs) field_label
623 field_lbls = dataConFieldLabels data_con
625 error_expr = mkRuntimeErrorApp rEC_SEL_ERROR_ID field_ty full_msg
626 full_msg = showSDoc (sep [text "No match in record selector", ppr sel_id])
628 -- unbox a product type...
629 -- we will recurse into newtypes, casting along the way, and unbox at the
630 -- first product data constructor we find. e.g.
632 -- data PairInt = PairInt Int Int
633 -- newtype S = MkS PairInt
636 -- If we have e = MkT (MkS (PairInt 0 1)) and some body expecting a list of
637 -- ids, we get (modulo int passing)
639 -- case (e `cast` CoT) `cast` CoS of
640 -- PairInt a b -> body [a,b]
642 -- The Ints passed around are just for creating fresh locals
643 unboxProduct :: Int -> CoreExpr -> Type -> (Int -> [Id] -> CoreExpr) -> CoreExpr
644 unboxProduct i arg arg_ty body
647 result = mkUnpackCase the_id arg con_args boxing_con rhs
648 (_tycon, _tycon_args, boxing_con, tys) = deepSplitProductType "unboxProduct" arg_ty
649 ([the_id], i') = mkLocals i [arg_ty]
650 (con_args, i'') = mkLocals i' tys
651 rhs = body i'' con_args
653 mkUnpackCase :: Id -> CoreExpr -> [Id] -> DataCon -> CoreExpr -> CoreExpr
654 -- (mkUnpackCase x e args Con body)
656 -- case (e `cast` ...) of bndr { Con args -> body }
658 -- the type of the bndr passed in is irrelevent
659 mkUnpackCase bndr arg unpk_args boxing_con body
660 = Case cast_arg (setIdType bndr bndr_ty) (exprType body) [(DataAlt boxing_con, unpk_args, body)]
662 (cast_arg, bndr_ty) = go (idType bndr) arg
664 | (tycon, tycon_args, _, _) <- splitProductType "mkUnpackCase" ty
665 , isNewTyCon tycon && not (isRecursiveTyCon tycon)
666 = go (newTyConInstRhs tycon tycon_args)
667 (unwrapNewTypeBody tycon tycon_args arg)
668 | otherwise = (arg, ty)
671 reboxProduct :: [Unique] -- uniques to create new local binders
672 -> Type -- type of product to box
673 -> ([Unique], -- remaining uniques
674 CoreExpr, -- boxed product
675 [Id]) -- Ids being boxed into product
678 (_tycon, _tycon_args, _pack_con, con_arg_tys) = deepSplitProductType "reboxProduct" ty
680 us' = dropList con_arg_tys us
682 arg_ids = zipWith (mkSysLocal FSLIT("rb")) us con_arg_tys
684 bind_rhs = mkProductBox arg_ids ty
687 (us', bind_rhs, arg_ids)
689 mkProductBox :: [Id] -> Type -> CoreExpr
690 mkProductBox arg_ids ty
693 (tycon, tycon_args, pack_con, _con_arg_tys) = splitProductType "mkProductBox" ty
696 | isNewTyCon tycon && not (isRecursiveTyCon tycon)
697 = wrap (mkProductBox arg_ids (newTyConInstRhs tycon tycon_args))
698 | otherwise = mkConApp pack_con (map Type tycon_args ++ map Var arg_ids)
700 wrap expr = wrapNewTypeBody tycon tycon_args expr
703 -- (mkReboxingAlt us con xs rhs) basically constructs the case
704 -- alternative (con, xs, rhs)
705 -- but it does the reboxing necessary to construct the *source*
706 -- arguments, xs, from the representation arguments ys.
708 -- data T = MkT !(Int,Int) Bool
710 -- mkReboxingAlt MkT [x,b] r
711 -- = (DataAlt MkT, [y::Int,z::Int,b], let x = (y,z) in r)
713 -- mkDataAlt should really be in DataCon, but it can't because
714 -- it manipulates CoreSyn.
717 :: [Unique] -- Uniques for the new Ids
719 -> [Var] -- Source-level args, including existential dicts
723 mkReboxingAlt us con args rhs
724 | not (any isMarkedUnboxed stricts)
725 = (DataAlt con, args, rhs)
729 (binds, args') = go args stricts us
731 (DataAlt con, args', mkLets binds rhs)
734 stricts = dataConExStricts con ++ dataConStrictMarks con
736 go [] _stricts _us = ([], [])
738 -- Type variable case
739 go (arg:args) stricts us
741 = let (binds, args') = go args stricts us
742 in (binds, arg:args')
744 -- Term variable case
745 go (arg:args) (str:stricts) us
746 | isMarkedUnboxed str
748 let (binds, unpacked_args') = go args stricts us'
749 (us', bind_rhs, unpacked_args) = reboxProduct us (idType arg)
751 (NonRec arg bind_rhs : binds, unpacked_args ++ unpacked_args')
753 = let (binds, args') = go args stricts us
754 in (binds, arg:args')
758 %************************************************************************
760 \subsection{Dictionary selectors}
762 %************************************************************************
764 Selecting a field for a dictionary. If there is just one field, then
765 there's nothing to do.
767 Dictionary selectors may get nested forall-types. Thus:
770 op :: forall b. Ord b => a -> b -> b
772 Then the top-level type for op is
774 op :: forall a. Foo a =>
778 This is unlike ordinary record selectors, which have all the for-alls
779 at the outside. When dealing with classes it's very convenient to
780 recover the original type signature from the class op selector.
783 mkDictSelId :: Name -> Class -> Id
784 mkDictSelId name clas
785 = mkGlobalId (ClassOpId clas) name sel_ty info
787 sel_ty = mkForAllTys tyvars (mkFunTy (idType dict_id) (idType the_arg_id))
788 -- We can't just say (exprType rhs), because that would give a type
790 -- for a single-op class (after all, the selector is the identity)
791 -- But it's type must expose the representation of the dictionary
792 -- to gat (say) C a -> (a -> a)
796 `setUnfoldingInfo` mkTopUnfolding rhs
797 `setAllStrictnessInfo` Just strict_sig
799 -- We no longer use 'must-inline' on record selectors. They'll
800 -- inline like crazy if they scrutinise a constructor
802 -- The strictness signature is of the form U(AAAVAAAA) -> T
803 -- where the V depends on which item we are selecting
804 -- It's worth giving one, so that absence info etc is generated
805 -- even if the selector isn't inlined
806 strict_sig = mkStrictSig (mkTopDmdType [arg_dmd] TopRes)
807 arg_dmd | isNewTyCon tycon = evalDmd
808 | otherwise = Eval (Prod [ if the_arg_id == id then evalDmd else Abs
811 tycon = classTyCon clas
812 [data_con] = tyConDataCons tycon
813 tyvars = dataConUnivTyVars data_con
814 arg_tys = ASSERT( isVanillaDataCon data_con ) dataConRepArgTys data_con
815 the_arg_id = assoc "MkId.mkDictSelId" (map idName (classSelIds clas) `zip` arg_ids) name
817 pred = mkClassPred clas (mkTyVarTys tyvars)
818 (dict_id:arg_ids) = mkTemplateLocals (mkPredTy pred : arg_tys)
820 rhs = mkLams tyvars (Lam dict_id rhs_body)
821 rhs_body | isNewTyCon tycon = unwrapNewTypeBody tycon (map mkTyVarTy tyvars) (Var dict_id)
822 | otherwise = Case (Var dict_id) dict_id (idType the_arg_id)
823 [(DataAlt data_con, arg_ids, Var the_arg_id)]
827 %************************************************************************
829 Wrapping and unwrapping newtypes and type families
831 %************************************************************************
834 wrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
835 -- The wrapper for the data constructor for a newtype looks like this:
836 -- newtype T a = MkT (a,Int)
837 -- MkT :: forall a. (a,Int) -> T a
838 -- MkT = /\a. \(x:(a,Int)). x `cast` sym (CoT a)
839 -- where CoT is the coercion TyCon assoicated with the newtype
841 -- The call (wrapNewTypeBody T [a] e) returns the
842 -- body of the wrapper, namely
843 -- e `cast` (CoT [a])
845 -- If a coercion constructor is provided in the newtype, then we use
846 -- it, otherwise the wrap/unwrap are both no-ops
848 -- If the we are dealing with a newtype *instance*, we have a second coercion
849 -- identifying the family instance with the constructor of the newtype
850 -- instance. This coercion is applied in any case (ie, composed with the
851 -- coercion constructor of the newtype or applied by itself).
853 wrapNewTypeBody tycon args result_expr
854 = wrapFamInstBody tycon args inner
857 | Just co_con <- newTyConCo_maybe tycon
858 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) result_expr
862 -- When unwrapping, we do *not* apply any family coercion, because this will
863 -- be done via a CoPat by the type checker. We have to do it this way as
864 -- computing the right type arguments for the coercion requires more than just
865 -- a spliting operation (cf, TcPat.tcConPat).
867 unwrapNewTypeBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
868 unwrapNewTypeBody tycon args result_expr
869 | Just co_con <- newTyConCo_maybe tycon
870 = mkCoerce (mkTyConApp co_con args) result_expr
874 -- If the type constructor is a representation type of a data instance, wrap
875 -- the expression into a cast adjusting the expression type, which is an
876 -- instance of the representation type, to the corresponding instance of the
877 -- family instance type.
878 -- See Note [Wrappers for data instance tycons]
879 wrapFamInstBody :: TyCon -> [Type] -> CoreExpr -> CoreExpr
880 wrapFamInstBody tycon args body
881 | Just co_con <- tyConFamilyCoercion_maybe tycon
882 = mkCoerce (mkSymCoercion (mkTyConApp co_con args)) body
886 unwrapFamInstScrut :: TyCon -> [Type] -> CoreExpr -> CoreExpr
887 unwrapFamInstScrut tycon args scrut
888 | Just co_con <- tyConFamilyCoercion_maybe tycon
889 = mkCoerce (mkTyConApp co_con args) scrut
895 %************************************************************************
897 \subsection{Primitive operations
899 %************************************************************************
902 mkPrimOpId :: PrimOp -> Id
906 (tyvars,arg_tys,res_ty, arity, strict_sig) = primOpSig prim_op
907 ty = mkForAllTys tyvars (mkFunTys arg_tys res_ty)
908 name = mkWiredInName gHC_PRIM (primOpOcc prim_op)
909 (mkPrimOpIdUnique (primOpTag prim_op))
911 id = mkGlobalId (PrimOpId prim_op) name ty info
914 `setSpecInfo` mkSpecInfo (primOpRules prim_op name)
916 `setAllStrictnessInfo` Just strict_sig
918 -- For each ccall we manufacture a separate CCallOpId, giving it
919 -- a fresh unique, a type that is correct for this particular ccall,
920 -- and a CCall structure that gives the correct details about calling
923 -- The *name* of this Id is a local name whose OccName gives the full
924 -- details of the ccall, type and all. This means that the interface
925 -- file reader can reconstruct a suitable Id
927 mkFCallId :: Unique -> ForeignCall -> Type -> Id
928 mkFCallId uniq fcall ty
929 = ASSERT( isEmptyVarSet (tyVarsOfType ty) )
930 -- A CCallOpId should have no free type variables;
931 -- when doing substitutions won't substitute over it
932 mkGlobalId (FCallId fcall) name ty info
934 occ_str = showSDoc (braces (ppr fcall <+> ppr ty))
935 -- The "occurrence name" of a ccall is the full info about the
936 -- ccall; it is encoded, but may have embedded spaces etc!
938 name = mkFCallName uniq occ_str
942 `setAllStrictnessInfo` Just strict_sig
944 (_, tau) = tcSplitForAllTys ty
945 (arg_tys, _) = tcSplitFunTys tau
946 arity = length arg_tys
947 strict_sig = mkStrictSig (mkTopDmdType (replicate arity evalDmd) TopRes)
949 -- Tick boxes and breakpoints are both represented as TickBoxOpIds,
950 -- except for the type:
952 -- a plain HPC tick box has type (State# RealWorld)
953 -- a breakpoint Id has type forall a.a
955 -- The breakpoint Id will be applied to a list of arbitrary free variables,
956 -- which is why it needs a polymorphic type.
958 mkTickBoxOpId :: Unique -> Module -> TickBoxId -> Id
959 mkTickBoxOpId uniq mod ix = mkTickBox' uniq mod ix realWorldStatePrimTy
961 mkBreakPointOpId :: Unique -> Module -> TickBoxId -> Id
962 mkBreakPointOpId uniq mod ix = mkTickBox' uniq mod ix ty
963 where ty = mkSigmaTy [openAlphaTyVar] [] openAlphaTy
965 mkTickBox' uniq mod ix ty = mkGlobalId (TickBoxOpId tickbox) name ty info
967 tickbox = TickBox mod ix
968 occ_str = showSDoc (braces (ppr tickbox))
969 name = mkTickBoxOpName uniq occ_str
974 %************************************************************************
976 \subsection{DictFuns and default methods}
978 %************************************************************************
980 Important notes about dict funs and default methods
981 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
982 Dict funs and default methods are *not* ImplicitIds. Their definition
983 involves user-written code, so we can't figure out their strictness etc
984 based on fixed info, as we can for constructors and record selectors (say).
986 We build them as LocalIds, but with External Names. This ensures that
987 they are taken to account by free-variable finding and dependency
988 analysis (e.g. CoreFVs.exprFreeVars).
990 Why shouldn't they be bound as GlobalIds? Because, in particular, if
991 they are globals, the specialiser floats dict uses above their defns,
992 which prevents good simplifications happening. Also the strictness
993 analyser treats a occurrence of a GlobalId as imported and assumes it
994 contains strictness in its IdInfo, which isn't true if the thing is
995 bound in the same module as the occurrence.
997 It's OK for dfuns to be LocalIds, because we form the instance-env to
998 pass on to the next module (md_insts) in CoreTidy, afer tidying
999 and globalising the top-level Ids.
1001 BUT make sure they are *exported* LocalIds (mkExportedLocalId) so
1002 that they aren't discarded by the occurrence analyser.
1005 mkDefaultMethodId dm_name ty = mkExportedLocalId dm_name ty
1007 mkDictFunId :: Name -- Name to use for the dict fun;
1014 mkDictFunId dfun_name inst_tyvars dfun_theta clas inst_tys
1015 = mkExportedLocalId dfun_name dfun_ty
1017 dfun_ty = mkSigmaTy inst_tyvars dfun_theta (mkDictTy clas inst_tys)
1019 {- 1 dec 99: disable the Mark Jones optimisation for the sake
1020 of compatibility with Hugs.
1021 See `types/InstEnv' for a discussion related to this.
1023 (class_tyvars, sc_theta, _, _) = classBigSig clas
1024 not_const (clas, tys) = not (isEmptyVarSet (tyVarsOfTypes tys))
1025 sc_theta' = substClasses (zipTopTvSubst class_tyvars inst_tys) sc_theta
1026 dfun_theta = case inst_decl_theta of
1027 [] -> [] -- If inst_decl_theta is empty, then we don't
1028 -- want to have any dict arguments, so that we can
1029 -- expose the constant methods.
1031 other -> nub (inst_decl_theta ++ filter not_const sc_theta')
1032 -- Otherwise we pass the superclass dictionaries to
1033 -- the dictionary function; the Mark Jones optimisation.
1035 -- NOTE the "nub". I got caught by this one:
1036 -- class Monad m => MonadT t m where ...
1037 -- instance Monad m => MonadT (EnvT env) m where ...
1038 -- Here, the inst_decl_theta has (Monad m); but so
1039 -- does the sc_theta'!
1041 -- NOTE the "not_const". I got caught by this one too:
1042 -- class Foo a => Baz a b where ...
1043 -- instance Wob b => Baz T b where..
1044 -- Now sc_theta' has Foo T
1049 %************************************************************************
1051 \subsection{Un-definable}
1053 %************************************************************************
1055 These Ids can't be defined in Haskell. They could be defined in
1056 unfoldings in the wired-in GHC.Prim interface file, but we'd have to
1057 ensure that they were definitely, definitely inlined, because there is
1058 no curried identifier for them. That's what mkCompulsoryUnfolding
1059 does. If we had a way to get a compulsory unfolding from an interface
1060 file, we could do that, but we don't right now.
1062 unsafeCoerce# isn't so much a PrimOp as a phantom identifier, that
1063 just gets expanded into a type coercion wherever it occurs. Hence we
1064 add it as a built-in Id with an unfolding here.
1066 The type variables we use here are "open" type variables: this means
1067 they can unify with both unlifted and lifted types. Hence we provide
1068 another gun with which to shoot yourself in the foot.
1071 mkWiredInIdName mod fs uniq id
1072 = mkWiredInName mod (mkOccNameFS varName fs) uniq (AnId id) UserSyntax
1074 unsafeCoerceName = mkWiredInIdName gHC_PRIM FSLIT("unsafeCoerce#") unsafeCoerceIdKey unsafeCoerceId
1075 nullAddrName = mkWiredInIdName gHC_PRIM FSLIT("nullAddr#") nullAddrIdKey nullAddrId
1076 seqName = mkWiredInIdName gHC_PRIM FSLIT("seq") seqIdKey seqId
1077 realWorldName = mkWiredInIdName gHC_PRIM FSLIT("realWorld#") realWorldPrimIdKey realWorldPrimId
1078 lazyIdName = mkWiredInIdName gHC_BASE FSLIT("lazy") lazyIdKey lazyId
1080 errorName = mkWiredInIdName gHC_ERR FSLIT("error") errorIdKey eRROR_ID
1081 recSelErrorName = mkWiredInIdName gHC_ERR FSLIT("recSelError") recSelErrorIdKey rEC_SEL_ERROR_ID
1082 runtimeErrorName = mkWiredInIdName gHC_ERR FSLIT("runtimeError") runtimeErrorIdKey rUNTIME_ERROR_ID
1083 irrefutPatErrorName = mkWiredInIdName gHC_ERR FSLIT("irrefutPatError") irrefutPatErrorIdKey iRREFUT_PAT_ERROR_ID
1084 recConErrorName = mkWiredInIdName gHC_ERR FSLIT("recConError") recConErrorIdKey rEC_CON_ERROR_ID
1085 patErrorName = mkWiredInIdName gHC_ERR FSLIT("patError") patErrorIdKey pAT_ERROR_ID
1086 noMethodBindingErrorName = mkWiredInIdName gHC_ERR FSLIT("noMethodBindingError")
1087 noMethodBindingErrorIdKey nO_METHOD_BINDING_ERROR_ID
1088 nonExhaustiveGuardsErrorName
1089 = mkWiredInIdName gHC_ERR FSLIT("nonExhaustiveGuardsError")
1090 nonExhaustiveGuardsErrorIdKey nON_EXHAUSTIVE_GUARDS_ERROR_ID
1094 -- unsafeCoerce# :: forall a b. a -> b
1096 = pcMiscPrelId unsafeCoerceName ty info
1098 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
1101 ty = mkForAllTys [openAlphaTyVar,openBetaTyVar]
1102 (mkFunTy openAlphaTy openBetaTy)
1103 [x] = mkTemplateLocals [openAlphaTy]
1104 rhs = mkLams [openAlphaTyVar,openBetaTyVar,x] $
1105 Cast (Var x) (mkUnsafeCoercion openAlphaTy openBetaTy)
1107 -- nullAddr# :: Addr#
1108 -- The reason is is here is because we don't provide
1109 -- a way to write this literal in Haskell.
1111 = pcMiscPrelId nullAddrName addrPrimTy info
1113 info = noCafIdInfo `setUnfoldingInfo`
1114 mkCompulsoryUnfolding (Lit nullAddrLit)
1117 = pcMiscPrelId seqName ty info
1119 info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs
1122 ty = mkForAllTys [alphaTyVar,openBetaTyVar]
1123 (mkFunTy alphaTy (mkFunTy openBetaTy openBetaTy))
1124 [x,y] = mkTemplateLocals [alphaTy, openBetaTy]
1125 rhs = mkLams [alphaTyVar,openBetaTyVar,x,y] (Case (Var x) x openBetaTy [(DEFAULT, [], Var y)])
1127 -- lazy :: forall a?. a? -> a? (i.e. works for unboxed types too)
1128 -- Used to lazify pseq: pseq a b = a `seq` lazy b
1130 -- Also, no strictness: by being a built-in Id, all the info about lazyId comes from here,
1131 -- not from GHC.Base.hi. This is important, because the strictness
1132 -- analyser will spot it as strict!
1134 -- Also no unfolding in lazyId: it gets "inlined" by a HACK in the worker/wrapperpass
1135 -- (see WorkWrap.wwExpr)
1136 -- We could use inline phases to do this, but that would be vulnerable to changes in
1137 -- phase numbering....we must inline precisely after strictness analysis.
1139 = pcMiscPrelId lazyIdName ty info
1142 ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy)
1144 lazyIdUnfolding :: CoreExpr -- Used to expand 'lazyId' after strictness anal
1145 lazyIdUnfolding = mkLams [openAlphaTyVar,x] (Var x)
1147 [x] = mkTemplateLocals [openAlphaTy]
1150 @realWorld#@ used to be a magic literal, \tr{void#}. If things get
1151 nasty as-is, change it back to a literal (@Literal@).
1153 voidArgId is a Local Id used simply as an argument in functions
1154 where we just want an arg to avoid having a thunk of unlifted type.
1156 x = \ void :: State# RealWorld -> (# p, q #)
1158 This comes up in strictness analysis
1161 realWorldPrimId -- :: State# RealWorld
1162 = pcMiscPrelId realWorldName realWorldStatePrimTy
1163 (noCafIdInfo `setUnfoldingInfo` evaldUnfolding)
1164 -- The evaldUnfolding makes it look that realWorld# is evaluated
1165 -- which in turn makes Simplify.interestingArg return True,
1166 -- which in turn makes INLINE things applied to realWorld# likely
1169 voidArgId -- :: State# RealWorld
1170 = mkSysLocal FSLIT("void") voidArgIdKey realWorldStatePrimTy
1174 %************************************************************************
1176 \subsection[PrelVals-error-related]{@error@ and friends; @trace@}
1178 %************************************************************************
1180 GHC randomly injects these into the code.
1182 @patError@ is just a version of @error@ for pattern-matching
1183 failures. It knows various ``codes'' which expand to longer
1184 strings---this saves space!
1186 @absentErr@ is a thing we put in for ``absent'' arguments. They jolly
1187 well shouldn't be yanked on, but if one is, then you will get a
1188 friendly message from @absentErr@ (rather than a totally random
1191 @parError@ is a special version of @error@ which the compiler does
1192 not know to be a bottoming Id. It is used in the @_par_@ and @_seq_@
1193 templates, but we don't ever expect to generate code for it.
1197 :: Id -- Should be of type (forall a. Addr# -> a)
1198 -- where Addr# points to a UTF8 encoded string
1199 -> Type -- The type to instantiate 'a'
1200 -> String -- The string to print
1203 mkRuntimeErrorApp err_id res_ty err_msg
1204 = mkApps (Var err_id) [Type res_ty, err_string]
1206 err_string = Lit (mkStringLit err_msg)
1208 rEC_SEL_ERROR_ID = mkRuntimeErrorId recSelErrorName
1209 rUNTIME_ERROR_ID = mkRuntimeErrorId runtimeErrorName
1210 iRREFUT_PAT_ERROR_ID = mkRuntimeErrorId irrefutPatErrorName
1211 rEC_CON_ERROR_ID = mkRuntimeErrorId recConErrorName
1212 pAT_ERROR_ID = mkRuntimeErrorId patErrorName
1213 nO_METHOD_BINDING_ERROR_ID = mkRuntimeErrorId noMethodBindingErrorName
1214 nON_EXHAUSTIVE_GUARDS_ERROR_ID = mkRuntimeErrorId nonExhaustiveGuardsErrorName
1216 -- The runtime error Ids take a UTF8-encoded string as argument
1217 mkRuntimeErrorId name = pc_bottoming_Id name runtimeErrorTy
1218 runtimeErrorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTy addrPrimTy openAlphaTy)
1222 eRROR_ID = pc_bottoming_Id errorName errorTy
1225 errorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTys [mkListTy charTy] openAlphaTy)
1226 -- Notice the openAlphaTyVar. It says that "error" can be applied
1227 -- to unboxed as well as boxed types. This is OK because it never
1228 -- returns, so the return type is irrelevant.
1232 %************************************************************************
1234 \subsection{Utilities}
1236 %************************************************************************
1239 pcMiscPrelId :: Name -> Type -> IdInfo -> Id
1240 pcMiscPrelId name ty info
1241 = mkVanillaGlobal name ty info
1242 -- We lie and say the thing is imported; otherwise, we get into
1243 -- a mess with dependency analysis; e.g., core2stg may heave in
1244 -- random calls to GHCbase.unpackPS__. If GHCbase is the module
1245 -- being compiled, then it's just a matter of luck if the definition
1246 -- will be in "the right place" to be in scope.
1248 pc_bottoming_Id name ty
1249 = pcMiscPrelId name ty bottoming_info
1251 bottoming_info = vanillaIdInfo `setAllStrictnessInfo` Just strict_sig
1252 -- Do *not* mark them as NoCafRefs, because they can indeed have
1253 -- CAF refs. For example, pAT_ERROR_ID calls GHC.Err.untangle,
1254 -- which has some CAFs
1255 -- In due course we may arrange that these error-y things are
1256 -- regarded by the GC as permanently live, in which case we
1257 -- can give them NoCaf info. As it is, any function that calls
1258 -- any pc_bottoming_Id will itself have CafRefs, which bloats
1261 strict_sig = mkStrictSig (mkTopDmdType [evalDmd] BotRes)
1262 -- These "bottom" out, no matter what their arguments