X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Fcompiler%2FbasicTypes%2FMkId.lhs;h=cbbe8ec489c5d8d7ed8a2e9e4516709b53aa4788;hb=f714e6b642fd614a9971717045ae47c3d871275e;hp=1f29b86401abb91f859497ed1a10f229abb3620c;hpb=bca9dd54c2b39638cb4638aaccf6015a104a1df5;p=ghc-hetmet.git diff --git a/ghc/compiler/basicTypes/MkId.lhs b/ghc/compiler/basicTypes/MkId.lhs index 1f29b86..cbbe8ec 100644 --- a/ghc/compiler/basicTypes/MkId.lhs +++ b/ghc/compiler/basicTypes/MkId.lhs @@ -13,89 +13,91 @@ have a standard form, namely: \begin{code} module MkId ( - mkSpecPragmaId, mkWorkerId, - mkDictFunId, mkDefaultMethodId, - mkDictSelId, + mkDictSelId, + + mkDataConIds, + mkRecordSelId, + mkPrimOpId, mkFCallId, - mkDataConId, mkDataConWrapId, - mkRecordSelId, - mkPrimOpId, mkCCallOpId, + mkReboxingAlt, mkNewTypeBody, -- And some particular Ids; see below for why they are wired in - wiredInIds, - unsafeCoerceId, realWorldPrimId, - eRROR_ID, rEC_SEL_ERROR_ID, pAT_ERROR_ID, rEC_CON_ERROR_ID, - rEC_UPD_ERROR_ID, iRREFUT_PAT_ERROR_ID, nON_EXHAUSTIVE_GUARDS_ERROR_ID, - nO_METHOD_BINDING_ERROR_ID, aBSENT_ERROR_ID, pAR_ERROR_ID + wiredInIds, ghcPrimIds, + unsafeCoerceId, realWorldPrimId, voidArgId, nullAddrId, seqId, + lazyId, lazyIdUnfolding, lazyIdKey, + + mkRuntimeErrorApp, + rEC_CON_ERROR_ID, iRREFUT_PAT_ERROR_ID, rUNTIME_ERROR_ID, + nON_EXHAUSTIVE_GUARDS_ERROR_ID, nO_METHOD_BINDING_ERROR_ID, + pAT_ERROR_ID, eRROR_ID ) where #include "HsVersions.h" +import BasicTypes ( Arity, StrictnessMark(..), isMarkedUnboxed, isMarkedStrict ) import TysPrim ( openAlphaTyVars, alphaTyVar, alphaTy, - intPrimTy, realWorldStatePrimTy + realWorldStatePrimTy, addrPrimTy ) import TysWiredIn ( charTy, mkListTy ) -import PrelNames ( pREL_ERR, pREL_GHC ) -import PrelRules ( primOpRule ) +import PrelRules ( primOpRules ) import Rules ( addRule ) -import Type ( Type, ThetaType, mkDictTy, mkDictTys, mkTyConApp, mkTyVarTys, - mkFunTys, mkFunTy, mkSigmaTy, +import Type ( TyThing(..) ) +import TcType ( Type, ThetaType, mkDictTy, mkPredTys, mkTyConApp, + mkTyVarTys, mkClassPred, tcEqPred, + mkFunTys, mkFunTy, mkSigmaTy, tcSplitSigmaTy, isUnLiftedType, mkForAllTys, mkTyVarTy, tyVarsOfType, - splitFunTys, splitForAllTys + tcSplitFunTys, tcSplitForAllTys, mkPredTy ) -import Module ( Module ) -import CoreUtils ( exprType, mkInlineMe ) +import CoreUtils ( exprType ) import CoreUnfold ( mkTopUnfolding, mkCompulsoryUnfolding, mkOtherCon ) -import Literal ( Literal(..) ) +import Literal ( Literal(..), nullAddrLit ) import TyCon ( TyCon, isNewTyCon, tyConTyVars, tyConDataCons, - tyConTheta, isProductTyCon, isUnboxedTupleTyCon ) + tyConTheta, isProductTyCon, isDataTyCon, isRecursiveTyCon ) import Class ( Class, classTyCon, classTyVars, classSelIds ) -import Var ( Id, TyVar ) +import Var ( Id, TyVar, Var ) import VarSet ( isEmptyVarSet ) -import Name ( mkDerivedName, mkWiredInName, mkLocalName, - mkWorkerOcc, mkCCallName, - Name, NamedThing(..), - ) -import OccName ( mkVarOcc ) -import PrimOp ( PrimOp(DataToTagOp, CCallOp), - primOpSig, mkPrimOpIdName, - CCall, pprCCallOp - ) -import Demand ( wwStrict, wwPrim, mkStrictnessInfo ) -import DataCon ( DataCon, StrictnessMark(..), - dataConFieldLabels, dataConRepArity, dataConTyCon, - dataConArgTys, dataConRepType, dataConRepStrictness, - dataConInstOrigArgTys, - dataConName, dataConTheta, - dataConSig, dataConStrictMarks, dataConId, - maybeMarkedUnboxed, splitProductType_maybe +import Name ( mkFCallName, mkWiredInName, Name ) +import OccName ( mkOccFS, varName ) +import PrimOp ( PrimOp, primOpSig, primOpOcc, primOpTag ) +import ForeignCall ( ForeignCall ) +import DataCon ( DataCon, DataConIds(..), + dataConFieldLabels, dataConRepArity, + dataConArgTys, dataConRepType, + dataConOrigArgTys, dataConTheta, + dataConSig, dataConStrictMarks, dataConExStricts, + splitProductType ) -import Id ( idType, mkId, - mkVanillaId, mkTemplateLocals, - mkTemplateLocal, idCprInfo +import Id ( idType, mkGlobalId, mkVanillaGlobal, mkSysLocal, + mkTemplateLocals, mkTemplateLocalsNum, mkExportedLocalId, + mkTemplateLocal, idName ) -import IdInfo ( IdInfo, vanillaIdInfo, mkIdInfo, - exactArity, setUnfoldingInfo, setCafInfo, setCprInfo, - setArityInfo, setSpecInfo, setTyGenInfo, - mkStrictnessInfo, setStrictnessInfo, - IdFlavour(..), CafInfo(..), CprInfo(..), TyGenInfo(..) +import IdInfo ( IdInfo, noCafIdInfo, setUnfoldingInfo, + setArityInfo, setSpecInfo, setCafInfo, + setAllStrictnessInfo, vanillaIdInfo, + GlobalIdDetails(..), CafInfo(..) ) -import FieldLabel ( mkFieldLabel, fieldLabelName, - firstFieldLabelTag, allFieldLabelTags, fieldLabelType +import NewDemand ( mkStrictSig, DmdResult(..), + mkTopDmdType, topDmd, evalDmd, lazyDmd, retCPR, + Demand(..), Demands(..) ) +import FieldLabel ( fieldLabelName, firstFieldLabelTag, + allFieldLabelTags, fieldLabelType ) +import DmdAnal ( dmdAnalTopRhs ) import CoreSyn +import Unique ( mkBuiltinUnique, mkPrimOpIdUnique ) import Maybes import PrelNames import Maybe ( isJust ) +import Util ( dropList, isSingleton ) import Outputable +import FastString import ListSetOps ( assoc, assocMaybe ) import UnicodeUtil ( stringToUtf8 ) -import Char ( ord ) +import List ( nubBy ) \end{code} - %************************************************************************ %* * \subsection{Wired in Ids} @@ -110,86 +112,42 @@ wiredInIds -- -- [The interface file format now carry such information, but there's -- no way yet of expressing at the definition site for these - -- error-reporting - -- functions that they have an 'open' result type. -- sof 1/99] - - aBSENT_ERROR_ID - , eRROR_ID - , iRREFUT_PAT_ERROR_ID - , nON_EXHAUSTIVE_GUARDS_ERROR_ID - , nO_METHOD_BINDING_ERROR_ID - , pAR_ERROR_ID - , pAT_ERROR_ID - , rEC_CON_ERROR_ID - , rEC_UPD_ERROR_ID - - -- These two can't be defined in Haskell - , realWorldPrimId - , unsafeCoerceId - , getTagId + -- error-reporting functions that they have an 'open' + -- result type. -- sof 1/99] + + eRROR_ID, -- This one isn't used anywhere else in the compiler + -- But we still need it in wiredInIds so that when GHC + -- compiles a program that mentions 'error' we don't + -- import its type from the interface file; we just get + -- the Id defined here. Which has an 'open-tyvar' type. + + rUNTIME_ERROR_ID, + iRREFUT_PAT_ERROR_ID, + nON_EXHAUSTIVE_GUARDS_ERROR_ID, + nO_METHOD_BINDING_ERROR_ID, + pAT_ERROR_ID, + rEC_CON_ERROR_ID, + + lazyId + ] ++ ghcPrimIds + +-- These Ids are exported from GHC.Prim +ghcPrimIds + = [ -- These can't be defined in Haskell, but they have + -- perfectly reasonable unfoldings in Core + realWorldPrimId, + unsafeCoerceId, + nullAddrId, + seqId ] \end{code} %************************************************************************ %* * -\subsection{Easy ones} -%* * -%************************************************************************ - -\begin{code} -mkSpecPragmaId occ uniq ty loc - = mkId (mkLocalName uniq occ loc) ty (mkIdInfo SpecPragmaId) - -- Maybe a SysLocal? But then we'd lose the location - -mkDefaultMethodId dm_name rec_c ty - = mkId dm_name ty info - where - info = vanillaIdInfo `setTyGenInfo` TyGenNever - -- type is wired-in (see comment at TcClassDcl.tcClassSig), so - -- do not generalise it - -mkWorkerId uniq unwrkr ty - = mkVanillaId (mkDerivedName mkWorkerOcc (getName unwrkr) uniq) ty -\end{code} - -%************************************************************************ -%* * \subsection{Data constructors} %* * %************************************************************************ -\begin{code} -mkDataConId :: Name -> DataCon -> Id - -- Makes the *worker* for the data constructor; that is, the function - -- that takes the reprsentation arguments and builds the constructor. -mkDataConId work_name data_con - = mkId work_name (dataConRepType data_con) info - where - info = mkIdInfo (DataConId data_con) - `setArityInfo` exactArity arity - `setStrictnessInfo` strict_info - `setCprInfo` cpr_info - - arity = dataConRepArity data_con - - strict_info = mkStrictnessInfo (dataConRepStrictness data_con, False) - - cpr_info | isProductTyCon tycon && - not (isUnboxedTupleTyCon tycon) && - arity > 0 = ReturnsCPR - | otherwise = NoCPRInfo - where - tycon = dataConTyCon data_con - -- Newtypes don't have a worker at all - -- - -- If we are a product with 0 args we must be void(like) - -- We can't create an unboxed tuple with 0 args for this - -- and since Void has only one, constant value it should - -- just mean returning a pointer to a pre-existing cell. - -- So we won't really gain from doing anything fancy - -- and we treat this case as Top. -\end{code} - The wrapper for a constructor is an ordinary top-level binding that evaluates any strict args, unboxes any args that are going to be flattened, and calls the worker. @@ -227,99 +185,124 @@ Notice that Making an explicit case expression allows the simplifier to eliminate it in the (common) case where the constructor arg is already evaluated. + \begin{code} -mkDataConWrapId data_con - = wrap_id - where - wrap_id = mkId (dataConName data_con) wrap_ty info - work_id = dataConId data_con - - info = mkIdInfo (DataConWrapId data_con) - `setUnfoldingInfo` mkTopUnfolding (mkInlineMe wrap_rhs) - `setCprInfo` cpr_info - -- The Cpr info can be important inside INLINE rhss, where the - -- wrapper constructor isn't inlined - `setArityInfo` exactArity arity - -- It's important to specify the arity, so that partial - -- applications are treated as values - `setCafInfo` NoCafRefs - -- The wrapper Id ends up in STG code as an argument, - -- sometimes before its definition, so we want to - -- signal that it has no CAFs - `setTyGenInfo` TyGenNever - -- No point generalising its type, since it gets eagerly inlined - -- away anyway - - wrap_ty = mkForAllTys all_tyvars $ - mkFunTys all_arg_tys - result_ty - - cpr_info = idCprInfo work_id - - wrap_rhs | isNewTyCon tycon - = ASSERT( null ex_tyvars && null ex_dict_args && length orig_arg_tys == 1 ) - -- No existentials on a newtype, but it can have a context - -- e.g. newtype Eq a => T a = MkT (...) - - mkLams tyvars $ mkLams dict_args $ Lam id_arg1 $ - Note (Coerce result_ty (head orig_arg_tys)) (Var id_arg1) - - | null dict_args && all not_marked_strict strict_marks - = Var work_id -- The common case. Not only is this efficient, - -- but it also ensures that the wrapper is replaced - -- by the worker even when there are no args. - -- f (:) x - -- becomes - -- f $w: x - -- This is really important in rule matching, - -- (We could match on the wrappers, - -- but that makes it less likely that rules will match - -- when we bring bits of unfoldings together.) - -- - -- NB: because of this special case, (map (:) ys) turns into - -- (map $w: ys), and thence into (map (\x xs. $w: x xs) ys) - -- in core-to-stg. The top-level defn for (:) is never used. - -- This is somewhat of a bore, but I'm currently leaving it - -- as is, so that there still is a top level curried (:) for - -- the interpreter to call. - - | otherwise - = mkLams all_tyvars $ mkLams dict_args $ - mkLams ex_dict_args $ mkLams id_args $ - foldr mk_case con_app - (zip (ex_dict_args++id_args) strict_marks) i3 [] - - con_app i rep_ids = mkApps (Var work_id) - (map varToCoreExpr (all_tyvars ++ reverse rep_ids)) +mkDataConIds :: Name -> Name -> DataCon -> DataConIds + -- Makes the *worker* for the data constructor; that is, the function + -- that takes the reprsentation arguments and builds the constructor. +mkDataConIds wrap_name wkr_name data_con + | isNewTyCon tycon + = NewDC nt_wrap_id - (tyvars, theta, ex_tyvars, ex_theta, orig_arg_tys, tycon) = dataConSig data_con - all_tyvars = tyvars ++ ex_tyvars + | any isMarkedStrict all_strict_marks -- Algebraic, needs wrapper + = AlgDC (Just alg_wrap_id) wrk_id - dict_tys = mkDictTys theta - ex_dict_tys = mkDictTys ex_theta - all_arg_tys = dict_tys ++ ex_dict_tys ++ orig_arg_tys - result_ty = mkTyConApp tycon (mkTyVarTys tyvars) + | otherwise -- Algebraic, no wrapper + = AlgDC Nothing wrk_id + where + (tyvars, _, ex_tyvars, ex_theta, orig_arg_tys, tycon) = dataConSig data_con + all_tyvars = tyvars ++ ex_tyvars - mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n) - where - n = length tys + ex_dict_tys = mkPredTys ex_theta + all_arg_tys = ex_dict_tys ++ orig_arg_tys + result_ty = mkTyConApp tycon (mkTyVarTys tyvars) - (dict_args, i1) = mkLocals 1 dict_tys - (ex_dict_args,i2) = mkLocals i1 ex_dict_tys - (id_args,i3) = mkLocals i2 orig_arg_tys - arity = i3-1 - (id_arg1:_) = id_args -- Used for newtype only + wrap_ty = mkForAllTys all_tyvars (mkFunTys all_arg_tys result_ty) + -- We used to include the stupid theta in the wrapper's args + -- but now we don't. Instead the type checker just injects these + -- extra constraints where necessary. + + ----------- Worker (algebraic data types only) -------------- + wrk_id = mkGlobalId (DataConWorkId data_con) wkr_name + (dataConRepType data_con) wkr_info + + wkr_arity = dataConRepArity data_con + wkr_info = noCafIdInfo + `setArityInfo` wkr_arity + `setAllStrictnessInfo` Just wkr_sig + + wkr_sig = mkStrictSig (mkTopDmdType (replicate wkr_arity topDmd) cpr_info) + -- Notice that we do *not* say the worker is strict + -- even if the data constructor is declared strict + -- e.g. data T = MkT !(Int,Int) + -- Why? Because the *wrapper* is strict (and its unfolding has case + -- expresssions that do the evals) but the *worker* itself is not. + -- If we pretend it is strict then when we see + -- case x of y -> $wMkT y + -- the simplifier thinks that y is "sure to be evaluated" (because + -- $wMkT is strict) and drops the case. No, $wMkT is not strict. + -- + -- When the simplifer sees a pattern + -- case e of MkT x -> ... + -- it uses the dataConRepStrictness of MkT to mark x as evaluated; + -- but that's fine... dataConRepStrictness comes from the data con + -- not from the worker Id. - strict_marks = dataConStrictMarks data_con - not_marked_strict NotMarkedStrict = True - not_marked_strict other = False + cpr_info | isProductTyCon tycon && + isDataTyCon tycon && + wkr_arity > 0 && + wkr_arity <= mAX_CPR_SIZE = retCPR + | otherwise = TopRes + -- RetCPR is only true for products that are real data types; + -- that is, not unboxed tuples or [non-recursive] newtypes + + ----------- Wrappers for newtypes -------------- + nt_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty nt_wrap_info + nt_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo + `setArityInfo` 1 -- Arity 1 + `setUnfoldingInfo` newtype_unf + newtype_unf = ASSERT( null ex_tyvars && null ex_theta && + isSingleton orig_arg_tys ) + -- No existentials on a newtype, but it can have a context + -- e.g. newtype Eq a => T a = MkT (...) + mkTopUnfolding $ Note InlineMe $ + mkLams tyvars $ Lam id_arg1 $ + mkNewTypeBody tycon result_ty (Var id_arg1) + + id_arg1 = mkTemplateLocal 1 (head orig_arg_tys) + + ----------- Wrappers for algebraic data types -------------- + alg_wrap_id = mkGlobalId (DataConWrapId data_con) wrap_name wrap_ty alg_wrap_info + alg_wrap_info = noCafIdInfo -- The NoCaf-ness is set by noCafIdInfo + `setArityInfo` alg_arity + -- It's important to specify the arity, so that partial + -- applications are treated as values + `setUnfoldingInfo` alg_unf + `setAllStrictnessInfo` Just wrap_sig + + all_strict_marks = dataConExStricts data_con ++ dataConStrictMarks data_con + wrap_sig = mkStrictSig (mkTopDmdType arg_dmds cpr_info) + arg_dmds = map mk_dmd all_strict_marks + mk_dmd str | isMarkedStrict str = evalDmd + | otherwise = lazyDmd + -- The Cpr info can be important inside INLINE rhss, where the + -- wrapper constructor isn't inlined. + -- And the argument strictness can be important too; we + -- may not inline a contructor when it is partially applied. + -- For example: + -- data W = C !Int !Int !Int + -- ...(let w = C x in ...(w p q)...)... + -- we want to see that w is strict in its two arguments + + alg_unf = mkTopUnfolding $ Note InlineMe $ + mkLams all_tyvars $ + mkLams ex_dict_args $ mkLams id_args $ + foldr mk_case con_app + (zip (ex_dict_args ++ id_args) all_strict_marks) + i3 [] + + con_app i rep_ids = mkApps (Var wrk_id) + (map varToCoreExpr (all_tyvars ++ reverse rep_ids)) + (ex_dict_args,i2) = mkLocals 1 ex_dict_tys + (id_args,i3) = mkLocals i2 orig_arg_tys + alg_arity = i3-1 mk_case - :: (Id, StrictnessMark) -- arg, strictness - -> (Int -> [Id] -> CoreExpr) -- body - -> Int -- next rep arg id - -> [Id] -- rep args so far + :: (Id, StrictnessMark) -- Arg, strictness + -> (Int -> [Id] -> CoreExpr) -- Body + -> Int -- Next rep arg id + -> [Id] -- Rep args so far, reversed -> CoreExpr mk_case (arg,strict) body i rep_args = case strict of @@ -329,11 +312,28 @@ mkDataConWrapId data_con | otherwise -> Case (Var arg) arg [(DEFAULT,[], body i (arg:rep_args))] - MarkedUnboxed con tys -> - Case (Var arg) arg [(DataAlt con, con_args, - body i' (reverse con_args++rep_args))] - where - (con_args,i') = mkLocals i tys + MarkedUnboxed + -> case splitProductType "do_unbox" (idType arg) of + (tycon, tycon_args, con, tys) -> + Case (Var arg) arg [(DataAlt con, con_args, + body i' (reverse con_args ++ rep_args))] + where + (con_args, i') = mkLocals i tys + +mAX_CPR_SIZE :: Arity +mAX_CPR_SIZE = 10 +-- We do not treat very big tuples as CPR-ish: +-- a) for a start we get into trouble because there aren't +-- "enough" unboxed tuple types (a tiresome restriction, +-- but hard to fix), +-- b) more importantly, big unboxed tuples get returned mainly +-- on the stack, and are often then allocated in the heap +-- by the caller. So doing CPR for them may in fact make +-- things worse. + +mkLocals i tys = (zipWith mkTemplateLocal [i..i+n-1] tys, i+n) + where + n = length tys \end{code} @@ -374,134 +374,195 @@ Then we want (not f :: R -> forall a. a->a, which gives the type inference mechanism problems at call sites) -Similarly for newtypes +Similarly for (recursive) newtypes newtype N = MkN { unN :: forall a. a->a } - unN :: forall a. N -> a -> a - unN = /\a -> \n:N -> coerce (a->a) n + unN :: forall b. N -> b -> b + unN = /\b -> \n:N -> (coerce (forall a. a->a) n) \begin{code} -mkRecordSelId tycon field_label unpack_id unpackUtf8_id +mkRecordSelId tycon field_label -- Assumes that all fields with the same field label have the same type - -- - -- Annoyingly, we have to pass in the unpackCString# Id, because - -- we can't conjure it up out of thin air = sel_id where - sel_id = mkId (fieldLabelName field_label) selector_ty info - + sel_id = mkGlobalId (RecordSelId field_label) (fieldLabelName field_label) selector_ty info field_ty = fieldLabelType field_label data_cons = tyConDataCons tycon tyvars = tyConTyVars tycon -- These scope over the types in -- the FieldLabels of constructors of this type - tycon_theta = tyConTheta tycon -- The context on the data decl - -- eg data (Eq a, Ord b) => T a b = ... - (field_tyvars,field_tau) = splitForAllTys field_ty - data_ty = mkTyConApp tycon tyvar_tys tyvar_tys = mkTyVarTys tyvars -- Very tiresomely, the selectors are (unnecessarily!) overloaded over -- just the dictionaries in the types of the constructors that contain - -- the relevant field. Urgh. + -- the relevant field. [The Report says that pattern matching on a + -- constructor gives the same constraints as applying it.] Urgh. + -- + -- However, not all data cons have all constraints (because of + -- TcTyDecls.thinContext). So we need to find all the data cons + -- involved in the pattern match and take the union of their constraints. + -- -- NB: this code relies on the fact that DataCons are quantified over -- the identical type variables as their parent TyCon - dict_tys = [mkDictTy cls tys | (cls, tys) <- tycon_theta, needed_dict (cls, tys)] - needed_dict pred = or [ pred `elem` (dataConTheta dc) - | (DataAlt dc, _, _) <- the_alts] + tycon_theta = tyConTheta tycon -- The context on the data decl + -- eg data (Eq a, Ord b) => T a b = ... + needed_preds = [pred | (DataAlt dc, _, _) <- the_alts, pred <- dataConTheta dc] + dict_tys = map mkPredTy (nubBy tcEqPred needed_preds) + n_dict_tys = length dict_tys + + (field_tyvars,field_theta,field_tau) = tcSplitSigmaTy field_ty + field_dict_tys = map mkPredTy field_theta + n_field_dict_tys = length field_dict_tys + -- If the field has a universally quantified type we have to + -- be a bit careful. Suppose we have + -- data R = R { op :: forall a. Foo a => a -> a } + -- Then we can't give op the type + -- op :: R -> forall a. Foo a => a -> a + -- because the typechecker doesn't understand foralls to the + -- right of an arrow. The "right" type to give it is + -- op :: forall a. Foo a => R -> a -> a + -- But then we must generate the right unfolding too: + -- op = /\a -> \dfoo -> \ r -> + -- case r of + -- R op -> op a dfoo + -- Note that this is exactly the type we'd infer from a user defn + -- op (R op) = op selector_ty :: Type selector_ty = mkForAllTys tyvars $ mkForAllTys field_tyvars $ - mkFunTys dict_tys $ mkFunTy data_ty field_tau + mkFunTys dict_tys $ mkFunTys field_dict_tys $ + mkFunTy data_ty field_tau - info = mkIdInfo (RecordSelId field_label) - `setArityInfo` exactArity (1 + length dict_tys) - `setUnfoldingInfo` unfolding - `setCafInfo` NoCafRefs - `setTyGenInfo` TyGenNever - -- ToDo: consider adding further IdInfo + arity = 1 + n_dict_tys + n_field_dict_tys + + (strict_sig, rhs_w_str) = dmdAnalTopRhs sel_rhs + -- Use the demand analyser to work out strictness. + -- With all this unpackery it's not easy! + + info = noCafIdInfo + `setCafInfo` caf_info + `setArityInfo` arity + `setUnfoldingInfo` mkTopUnfolding rhs_w_str + `setAllStrictnessInfo` Just strict_sig + + -- Allocate Ids. We do it a funny way round because field_dict_tys is + -- almost always empty. Also note that we use length_tycon_theta + -- rather than n_dict_tys, because the latter gives an infinite loop: + -- n_dict tys depends on the_alts, which depens on arg_ids, which depends + -- on arity, which depends on n_dict tys. Sigh! Mega sigh! + field_dict_base = length tycon_theta + 1 + dict_id_base = field_dict_base + n_field_dict_tys + field_base = dict_id_base + 1 + dict_ids = mkTemplateLocalsNum 1 dict_tys + field_dict_ids = mkTemplateLocalsNum field_dict_base field_dict_tys + data_id = mkTemplateLocal dict_id_base data_ty - unfolding = mkTopUnfolding sel_rhs - - - (data_id:dict_ids) = mkTemplateLocals (data_ty:dict_tys) alts = map mk_maybe_alt data_cons the_alts = catMaybes alts - default_alt | all isJust alts = [] -- No default needed - | otherwise = [(DEFAULT, [], error_expr)] - sel_rhs = mkLams tyvars $ mkLams field_tyvars $ - mkLams dict_ids $ Lam data_id $ - sel_body + no_default = all isJust alts -- No default needed + default_alt | no_default = [] + | otherwise = [(DEFAULT, [], error_expr)] + + -- The default branch may have CAF refs, because it calls recSelError etc. + caf_info | no_default = NoCafRefs + | otherwise = MayHaveCafRefs - sel_body | isNewTyCon tycon = Note (Coerce field_tau data_ty) (Var data_id) - | otherwise = Case (Var data_id) data_id (the_alts ++ default_alt) + sel_rhs = mkLams tyvars $ mkLams field_tyvars $ + mkLams dict_ids $ mkLams field_dict_ids $ + Lam data_id $ sel_body + + sel_body | isNewTyCon tycon = mk_result (mkNewTypeBody tycon field_ty (Var data_id)) + | otherwise = Case (Var data_id) data_id (default_alt ++ the_alts) + + mk_result poly_result = mkVarApps (mkVarApps poly_result field_tyvars) field_dict_ids + -- We pull the field lambdas to the top, so we need to + -- apply them in the body. For example: + -- data T = MkT { foo :: forall a. a->a } + -- + -- foo :: forall a. T -> a -> a + -- foo = /\a. \t:T. case t of { MkT f -> f a } mk_maybe_alt data_con - = case maybe_the_arg_id of + = case maybe_the_arg_id of Nothing -> Nothing - Just the_arg_id -> Just (DataAlt data_con, real_args, expr) - where - body = mkVarApps (Var the_arg_id) field_tyvars - strict_marks = dataConStrictMarks data_con - (expr, real_args) = rebuildConArgs data_con arg_ids strict_marks body - (length arg_ids + 1) + Just the_arg_id -> Just (mkReboxingAlt uniqs data_con arg_ids body) + where + body = mk_result (Var the_arg_id) where - arg_ids = mkTemplateLocals (dataConInstOrigArgTys data_con tyvar_tys) - -- The first one will shadow data_id, but who cares + arg_ids = mkTemplateLocalsNum field_base (dataConOrigArgTys data_con) + -- No need to instantiate; same tyvars in datacon as tycon + -- Records can't be existential, so no existential tyvars or dicts + + unpack_base = field_base + length arg_ids + uniqs = map mkBuiltinUnique [unpack_base..] + + -- arity+1 avoids all shadowing maybe_the_arg_id = assocMaybe (field_lbls `zip` arg_ids) field_label field_lbls = dataConFieldLabels data_con - error_expr = mkApps (Var rEC_SEL_ERROR_ID) [Type field_tau, err_string] - err_string - | all safeChar full_msg - = App (Var unpack_id) (Lit (MachStr (_PK_ full_msg))) - | otherwise - = App (Var unpackUtf8_id) (Lit (MachStr (_PK_ (stringToUtf8 (map ord full_msg))))) - where - safeChar c = c >= '\1' && c <= '\xFF' - -- TODO: Putting this Unicode stuff here is ugly. Find a better - -- generic place to make string literals. This logic is repeated - -- in DsUtils. + error_expr = mkRuntimeErrorApp rEC_SEL_ERROR_ID field_tau full_msg full_msg = showSDoc (sep [text "No match in record selector", ppr sel_id]) --- this rather ugly function converts the unpacked data con arguments back into --- their packed form. It is almost the same as the version in DsUtils, except that --- we use template locals here rather than newDsId (ToDo: merge these). - -rebuildConArgs - :: DataCon -- the con we're matching on - -> [Id] -- the source-level args - -> [StrictnessMark] -- the strictness annotations (per-arg) - -> CoreExpr -- the body - -> Int -- template local - -> (CoreExpr, [Id]) - -rebuildConArgs con [] stricts body i = (body, []) -rebuildConArgs con (arg:args) stricts body i | isTyVar arg - = let (body', args') = rebuildConArgs con args stricts body i - in (body',arg:args') -rebuildConArgs con (arg:args) (str:stricts) body i - = case maybeMarkedUnboxed str of - Just (pack_con1, _) -> - case splitProductType_maybe (idType arg) of - Just (_, tycon_args, pack_con, con_arg_tys) -> - ASSERT( pack_con == pack_con1 ) - let unpacked_args = zipWith mkTemplateLocal [i..] con_arg_tys - (body', real_args) = rebuildConArgs con args stricts body - (i + length con_arg_tys) - in - ( - Let (NonRec arg (mkConApp pack_con - (map Type tycon_args ++ - map Var unpacked_args))) body', - unpacked_args ++ real_args - ) - - _ -> let (body', args') = rebuildConArgs con args stricts body i - in (body', arg:args') +-- (mkReboxingAlt us con xs rhs) basically constructs the case +-- alternative (con, xs, rhs) +-- but it does the reboxing necessary to construct the *source* +-- arguments, xs, from the representation arguments ys. +-- For example: +-- data T = MkT !(Int,Int) Bool +-- +-- mkReboxingAlt MkT [x,b] r +-- = (DataAlt MkT, [y::Int,z::Int,b], let x = (y,z) in r) +-- +-- mkDataAlt should really be in DataCon, but it can't because +-- it manipulates CoreSyn. + +mkReboxingAlt + :: [Unique] -- Uniques for the new Ids + -> DataCon + -> [Var] -- Source-level args + -> CoreExpr -- RHS + -> CoreAlt + +mkReboxingAlt us con args rhs + | not (any isMarkedUnboxed stricts) + = (DataAlt con, args, rhs) + + | otherwise + = let + (binds, args') = go args stricts us + in + (DataAlt con, args', mkLets binds rhs) + + where + stricts = dataConExStricts con ++ dataConStrictMarks con + + go [] stricts us = ([], []) + + -- Type variable case + go (arg:args) stricts us + | isTyVar arg + = let (binds, args') = go args stricts us + in (binds, arg:args') + + -- Term variable case + go (arg:args) (str:stricts) us + | isMarkedUnboxed str + = let + (_, tycon_args, pack_con, con_arg_tys) + = splitProductType "mkReboxingAlt" (idType arg) + + unpacked_args = zipWith (mkSysLocal FSLIT("rb")) us con_arg_tys + (binds, args') = go args stricts (dropList con_arg_tys us) + con_app = mkConApp pack_con (map Type tycon_args ++ map Var unpacked_args) + in + (NonRec arg con_app : binds, unpacked_args ++ args') + + | otherwise + = let (binds, args') = go args stricts us + in (binds, arg:args') \end{code} @@ -514,28 +575,51 @@ rebuildConArgs con (arg:args) (str:stricts) body i Selecting a field for a dictionary. If there is just one field, then there's nothing to do. -ToDo: unify with mkRecordSelId. +Dictionary selectors may get nested forall-types. Thus: + + class Foo a where + op :: forall b. Ord b => a -> b -> b + +Then the top-level type for op is + + op :: forall a. Foo a => + forall b. Ord b => + a -> b -> b + +This is unlike ordinary record selectors, which have all the for-alls +at the outside. When dealing with classes it's very convenient to +recover the original type signature from the class op selector. \begin{code} mkDictSelId :: Name -> Class -> Id mkDictSelId name clas - = sel_id + = mkGlobalId (ClassOpId clas) name sel_ty info where - ty = exprType rhs - sel_id = mkId name ty info - field_lbl = mkFieldLabel name tycon ty tag - tag = assoc "MkId.mkDictSelId" (classSelIds clas `zip` allFieldLabelTags) sel_id - - info = mkIdInfo (RecordSelId field_lbl) - `setArityInfo` exactArity 1 - `setUnfoldingInfo` unfolding - `setCafInfo` NoCafRefs - `setTyGenInfo` TyGenNever - + sel_ty = mkForAllTys tyvars (mkFunTy (idType dict_id) (idType the_arg_id)) + -- We can't just say (exprType rhs), because that would give a type + -- C a -> C a + -- for a single-op class (after all, the selector is the identity) + -- But it's type must expose the representation of the dictionary + -- to gat (say) C a -> (a -> a) + + tag = assoc "MkId.mkDictSelId" (map idName (classSelIds clas) `zip` allFieldLabelTags) name + + info = noCafIdInfo + `setArityInfo` 1 + `setUnfoldingInfo` mkTopUnfolding rhs + `setAllStrictnessInfo` Just strict_sig + -- We no longer use 'must-inline' on record selectors. They'll -- inline like crazy if they scrutinise a constructor - unfolding = mkTopUnfolding rhs + -- The strictness signature is of the form U(AAAVAAAA) -> T + -- where the V depends on which item we are selecting + -- It's worth giving one, so that absence info etc is generated + -- even if the selector isn't inlined + strict_sig = mkStrictSig (mkTopDmdType [arg_dmd] TopRes) + arg_dmd | isNewTyCon tycon = evalDmd + | otherwise = Eval (Prod [ if the_arg_id == id then evalDmd else Abs + | id <- arg_ids ]) tyvars = classTyVars clas @@ -545,14 +629,22 @@ mkDictSelId name clas arg_tys = dataConArgTys data_con tyvar_tys the_arg_id = arg_ids !! (tag - firstFieldLabelTag) - dict_ty = mkDictTy clas tyvar_tys - (dict_id:arg_ids) = mkTemplateLocals (dict_ty : arg_tys) + pred = mkClassPred clas tyvar_tys + (dict_id:arg_ids) = mkTemplateLocals (mkPredTy pred : arg_tys) - rhs | isNewTyCon tycon = mkLams tyvars $ Lam dict_id $ - Note (Coerce (head arg_tys) dict_ty) (Var dict_id) + rhs | isNewTyCon tycon = mkLams tyvars $ Lam dict_id $ + mkNewTypeBody tycon (head arg_tys) (Var dict_id) | otherwise = mkLams tyvars $ Lam dict_id $ Case (Var dict_id) dict_id [(DataAlt data_con, arg_ids, Var the_arg_id)] + +mkNewTypeBody tycon result_ty result_expr + -- Adds a coerce where necessary + -- Used for both wrapping and unwrapping + | isRecursiveTyCon tycon -- Recursive case; use a coerce + = Note (Coerce result_ty (exprType result_expr)) result_expr + | otherwise -- Normal case + = result_expr \end{code} @@ -567,17 +659,19 @@ mkPrimOpId :: PrimOp -> Id mkPrimOpId prim_op = id where - (tyvars,arg_tys,res_ty, arity, strict_info) = primOpSig prim_op + (tyvars,arg_tys,res_ty, arity, strict_sig) = primOpSig prim_op ty = mkForAllTys tyvars (mkFunTys arg_tys res_ty) - name = mkPrimOpIdName prim_op - id = mkId name ty info + name = mkWiredInName gHC_PRIM (primOpOcc prim_op) + (mkPrimOpIdUnique (primOpTag prim_op)) + Nothing (AnId id) + id = mkGlobalId (PrimOpId prim_op) name ty info - info = mkIdInfo (PrimOpId prim_op) + info = noCafIdInfo `setSpecInfo` rules - `setArityInfo` exactArity arity - `setStrictnessInfo` strict_info + `setArityInfo` arity + `setAllStrictnessInfo` Just strict_sig - rules = addRule emptyCoreRules id (primOpRule prim_op) + rules = foldl (addRule id) emptyCoreRules (primOpRules prim_op) -- For each ccall we manufacture a separate CCallOpId, giving it @@ -589,52 +683,76 @@ mkPrimOpId prim_op -- details of the ccall, type and all. This means that the interface -- file reader can reconstruct a suitable Id -mkCCallOpId :: Unique -> CCall -> Type -> Id -mkCCallOpId uniq ccall ty +mkFCallId :: Unique -> ForeignCall -> Type -> Id +mkFCallId uniq fcall ty = ASSERT( isEmptyVarSet (tyVarsOfType ty) ) -- A CCallOpId should have no free type variables; -- when doing substitutions won't substitute over it - mkId name ty info + mkGlobalId (FCallId fcall) name ty info where - occ_str = showSDocIface (braces (pprCCallOp ccall <+> ppr ty)) + occ_str = showSDoc (braces (ppr fcall <+> ppr ty)) -- The "occurrence name" of a ccall is the full info about the -- ccall; it is encoded, but may have embedded spaces etc! - name = mkCCallName uniq occ_str - prim_op = CCallOp ccall + name = mkFCallName uniq occ_str - info = mkIdInfo (PrimOpId prim_op) - `setArityInfo` exactArity arity - `setStrictnessInfo` strict_info + info = noCafIdInfo + `setArityInfo` arity + `setAllStrictnessInfo` Just strict_sig - (_, tau) = splitForAllTys ty - (arg_tys, _) = splitFunTys tau + (_, tau) = tcSplitForAllTys ty + (arg_tys, _) = tcSplitFunTys tau arity = length arg_tys - strict_info = mkStrictnessInfo (take arity (repeat wwPrim), False) + strict_sig = mkStrictSig (mkTopDmdType (replicate arity evalDmd) TopRes) \end{code} %************************************************************************ %* * -\subsection{DictFuns} +\subsection{DictFuns and default methods} %* * %************************************************************************ +Important notes about dict funs and default methods +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Dict funs and default methods are *not* ImplicitIds. Their definition +involves user-written code, so we can't figure out their strictness etc +based on fixed info, as we can for constructors and record selectors (say). + +We build them as GlobalIds, but when in the module where they are +bound, we turn the Id at the *binding site* into an exported LocalId. +This ensures that they are taken to account by free-variable finding +and dependency analysis (e.g. CoreFVs.exprFreeVars). The simplifier +will propagate the LocalId to all occurrence sites. + +Why shouldn't they be bound as GlobalIds? Because, in particular, if +they are globals, the specialiser floats dict uses above their defns, +which prevents good simplifications happening. Also the strictness +analyser treats a occurrence of a GlobalId as imported and assumes it +contains strictness in its IdInfo, which isn't true if the thing is +bound in the same module as the occurrence. + +It's OK for dfuns to be LocalIds, because we form the instance-env to +pass on to the next module (md_insts) in CoreTidy, afer tidying +and globalising the top-level Ids. + +BUT make sure they are *exported* LocalIds (setIdLocalExported) so +that they aren't discarded by the occurrence analyser. + \begin{code} +mkDefaultMethodId dm_name ty = mkExportedLocalId dm_name ty + mkDictFunId :: Name -- Name to use for the dict fun; - -> Class -> [TyVar] - -> [Type] -> ThetaType + -> Class + -> [Type] -> Id -mkDictFunId dfun_name clas inst_tyvars inst_tys dfun_theta - = mkId dfun_name dfun_ty info +mkDictFunId dfun_name inst_tyvars dfun_theta clas inst_tys + = mkExportedLocalId dfun_name dfun_ty where dfun_ty = mkSigmaTy inst_tyvars dfun_theta (mkDictTy clas inst_tys) - info = vanillaIdInfo `setTyGenInfo` TyGenNever - -- type is wired-in (see comment at TcClassDcl.tcClassSig), so - -- do not generalise it {- 1 dec 99: disable the Mark Jones optimisation for the sake of compatibility with Hugs. @@ -672,7 +790,12 @@ mkDictFunId dfun_name clas inst_tyvars inst_tys dfun_theta %* * %************************************************************************ -These two can't be defined in Haskell. +These Ids can't be defined in Haskell. They could be defined in +unfoldings in the wired-in GHC.Prim interface file, but we'd have to +ensure that they were definitely, definitely inlined, because there is +no curried identifier for them. That's what mkCompulsoryUnfolding +does. If we had a way to get a compulsory unfolding from an interface +file, we could do that, but we don't right now. unsafeCoerce# isn't so much a PrimOp as a phantom identifier, that just gets expanded into a type coercion wherever it occurs. Hence we @@ -683,11 +806,34 @@ they can unify with both unlifted and lifted types. Hence we provide another gun with which to shoot yourself in the foot. \begin{code} +mkWiredInIdName mod fs uniq id + = mkWiredInName mod (mkOccFS varName fs) uniq Nothing (AnId id) + +unsafeCoerceName = mkWiredInIdName gHC_PRIM FSLIT("unsafeCoerce#") unsafeCoerceIdKey unsafeCoerceId +nullAddrName = mkWiredInIdName gHC_PRIM FSLIT("nullAddr#") nullAddrIdKey nullAddrId +seqName = mkWiredInIdName gHC_PRIM FSLIT("seq") seqIdKey seqId +realWorldName = mkWiredInIdName gHC_PRIM FSLIT("realWorld#") realWorldPrimIdKey realWorldPrimId +lazyIdName = mkWiredInIdName pREL_BASE FSLIT("lazy") lazyIdKey lazyId + +errorName = mkWiredInIdName pREL_ERR FSLIT("error") errorIdKey eRROR_ID +recSelErrorName = mkWiredInIdName pREL_ERR FSLIT("recSelError") recSelErrorIdKey rEC_SEL_ERROR_ID +runtimeErrorName = mkWiredInIdName pREL_ERR FSLIT("runtimeError") runtimeErrorIdKey rUNTIME_ERROR_ID +irrefutPatErrorName = mkWiredInIdName pREL_ERR FSLIT("irrefutPatError") irrefutPatErrorIdKey iRREFUT_PAT_ERROR_ID +recConErrorName = mkWiredInIdName pREL_ERR FSLIT("recConError") recConErrorIdKey rEC_CON_ERROR_ID +patErrorName = mkWiredInIdName pREL_ERR FSLIT("patError") patErrorIdKey pAT_ERROR_ID +noMethodBindingErrorName = mkWiredInIdName pREL_ERR FSLIT("noMethodBindingError") + noMethodBindingErrorIdKey nO_METHOD_BINDING_ERROR_ID +nonExhaustiveGuardsErrorName + = mkWiredInIdName pREL_ERR FSLIT("nonExhaustiveGuardsError") + nonExhaustiveGuardsErrorIdKey nON_EXHAUSTIVE_GUARDS_ERROR_ID +\end{code} + +\begin{code} +-- unsafeCoerce# :: forall a b. a -> b unsafeCoerceId - = pcMiscPrelId unsafeCoerceIdKey pREL_GHC SLIT("unsafeCoerce#") ty info + = pcMiscPrelId unsafeCoerceName ty info where - info = vanillaIdInfo - `setUnfoldingInfo` mkCompulsoryUnfolding rhs + info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs ty = mkForAllTys [openAlphaTyVar,openBetaTyVar] @@ -695,40 +841,66 @@ unsafeCoerceId [x] = mkTemplateLocals [openAlphaTy] rhs = mkLams [openAlphaTyVar,openBetaTyVar,x] $ Note (Coerce openBetaTy openAlphaTy) (Var x) -\end{code} - -@getTag#@ is another function which can't be defined in Haskell. It needs to -evaluate its argument and call the dataToTag# primitive. +-- nullAddr# :: Addr# +-- The reason is is here is because we don't provide +-- a way to write this literal in Haskell. +nullAddrId + = pcMiscPrelId nullAddrName addrPrimTy info + where + info = noCafIdInfo `setUnfoldingInfo` + mkCompulsoryUnfolding (Lit nullAddrLit) -\begin{code} -getTagId - = pcMiscPrelId getTagIdKey pREL_GHC SLIT("getTag#") ty info +seqId + = pcMiscPrelId seqName ty info where - info = vanillaIdInfo - `setUnfoldingInfo` mkCompulsoryUnfolding rhs - -- We don't provide a defn for this; you must inline it + info = noCafIdInfo `setUnfoldingInfo` mkCompulsoryUnfolding rhs + - ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy intPrimTy) - [x,y] = mkTemplateLocals [alphaTy,alphaTy] - rhs = mkLams [alphaTyVar,x] $ - Case (Var x) y [ (DEFAULT, [], mkApps (Var dataToTagId) [Type alphaTy, Var y]) ] + ty = mkForAllTys [alphaTyVar,openBetaTyVar] + (mkFunTy alphaTy (mkFunTy openBetaTy openBetaTy)) + [x,y] = mkTemplateLocals [alphaTy, openBetaTy] + rhs = mkLams [alphaTyVar,openBetaTyVar,x,y] (Case (Var x) x [(DEFAULT, [], Var y)]) + +-- lazy :: forall a?. a? -> a? (i.e. works for unboxed types too) +-- Used to lazify pseq: pseq a b = a `seq` lazy b +-- No unfolding: it gets "inlined" by the worker/wrapper pass +-- Also, no strictness: by being a built-in Id, it overrides all +-- the info in PrelBase.hi. This is important, because the strictness +-- analyser will spot it as strict! +lazyId + = pcMiscPrelId lazyIdName ty info + where + info = noCafIdInfo + ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy) -dataToTagId = mkPrimOpId DataToTagOp +lazyIdUnfolding :: CoreExpr -- Used to expand LazyOp after strictness anal +lazyIdUnfolding = mkLams [openAlphaTyVar,x] (Var x) + where + [x] = mkTemplateLocals [openAlphaTy] \end{code} @realWorld#@ used to be a magic literal, \tr{void#}. If things get nasty as-is, change it back to a literal (@Literal@). +voidArgId is a Local Id used simply as an argument in functions +where we just want an arg to avoid having a thunk of unlifted type. +E.g. + x = \ void :: State# RealWorld -> (# p, q #) + +This comes up in strictness analysis + \begin{code} realWorldPrimId -- :: State# RealWorld - = pcMiscPrelId realWorldPrimIdKey pREL_GHC SLIT("realWorld#") - realWorldStatePrimTy + = pcMiscPrelId realWorldName realWorldStatePrimTy (noCafIdInfo `setUnfoldingInfo` mkOtherCon []) -- The mkOtherCon makes it look that realWorld# is evaluated -- which in turn makes Simplify.interestingArg return True, -- which in turn makes INLINE things applied to realWorld# likely -- to be inlined + +voidArgId -- :: State# RealWorld + = mkSysLocal FSLIT("void") voidArgIdKey realWorldStatePrimTy \end{code} @@ -754,31 +926,39 @@ not know to be a bottoming Id. It is used in the @_par_@ and @_seq_@ templates, but we don't ever expect to generate code for it. \begin{code} -eRROR_ID - = pc_bottoming_Id errorIdKey pREL_ERR SLIT("error") errorTy -pAT_ERROR_ID - = generic_ERROR_ID patErrorIdKey SLIT("patError") -rEC_SEL_ERROR_ID - = generic_ERROR_ID recSelErrIdKey SLIT("recSelError") -rEC_CON_ERROR_ID - = generic_ERROR_ID recConErrorIdKey SLIT("recConError") -rEC_UPD_ERROR_ID - = generic_ERROR_ID recUpdErrorIdKey SLIT("recUpdError") -iRREFUT_PAT_ERROR_ID - = generic_ERROR_ID irrefutPatErrorIdKey SLIT("irrefutPatError") -nON_EXHAUSTIVE_GUARDS_ERROR_ID - = generic_ERROR_ID nonExhaustiveGuardsErrorIdKey SLIT("nonExhaustiveGuardsError") -nO_METHOD_BINDING_ERROR_ID - = generic_ERROR_ID noMethodBindingErrorIdKey SLIT("noMethodBindingError") - -aBSENT_ERROR_ID - = pc_bottoming_Id absentErrorIdKey pREL_ERR SLIT("absentErr") - (mkSigmaTy [openAlphaTyVar] [] openAlphaTy) - -pAR_ERROR_ID - = pcMiscPrelId parErrorIdKey pREL_ERR SLIT("parError") - (mkSigmaTy [openAlphaTyVar] [] openAlphaTy) noCafIdInfo +mkRuntimeErrorApp + :: Id -- Should be of type (forall a. Addr# -> a) + -- where Addr# points to a UTF8 encoded string + -> Type -- The type to instantiate 'a' + -> String -- The string to print + -> CoreExpr + +mkRuntimeErrorApp err_id res_ty err_msg + = mkApps (Var err_id) [Type res_ty, err_string] + where + err_string = Lit (MachStr (mkFastString (stringToUtf8 err_msg))) + +rEC_SEL_ERROR_ID = mkRuntimeErrorId recSelErrorName +rUNTIME_ERROR_ID = mkRuntimeErrorId runtimeErrorName +iRREFUT_PAT_ERROR_ID = mkRuntimeErrorId irrefutPatErrorName +rEC_CON_ERROR_ID = mkRuntimeErrorId recConErrorName +pAT_ERROR_ID = mkRuntimeErrorId patErrorName +nO_METHOD_BINDING_ERROR_ID = mkRuntimeErrorId noMethodBindingErrorName +nON_EXHAUSTIVE_GUARDS_ERROR_ID = mkRuntimeErrorId nonExhaustiveGuardsErrorName + +-- The runtime error Ids take a UTF8-encoded string as argument +mkRuntimeErrorId name = pc_bottoming_Id name runtimeErrorTy +runtimeErrorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTy addrPrimTy openAlphaTy) +\end{code} + +\begin{code} +eRROR_ID = pc_bottoming_Id errorName errorTy +errorTy :: Type +errorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTys [mkListTy charTy] openAlphaTy) + -- Notice the openAlphaTyVar. It says that "error" can be applied + -- to unboxed as well as boxed types. This is OK because it never + -- returns, so the return type is irrelevant. \end{code} @@ -789,41 +969,33 @@ pAR_ERROR_ID %************************************************************************ \begin{code} -pcMiscPrelId :: Unique{-IdKey-} -> Module -> FAST_STRING -> Type -> IdInfo -> Id -pcMiscPrelId key mod str ty info - = let - name = mkWiredInName mod (mkVarOcc str) key - imp = mkId name ty info -- the usual case... - in - imp +pcMiscPrelId :: Name -> Type -> IdInfo -> Id +pcMiscPrelId name ty info + = mkVanillaGlobal name ty info -- We lie and say the thing is imported; otherwise, we get into -- a mess with dependency analysis; e.g., core2stg may heave in -- random calls to GHCbase.unpackPS__. If GHCbase is the module -- being compiled, then it's just a matter of luck if the definition -- will be in "the right place" to be in scope. -pc_bottoming_Id key mod name ty - = pcMiscPrelId key mod name ty bottoming_info +pc_bottoming_Id name ty + = pcMiscPrelId name ty bottoming_info where - bottoming_info = noCafIdInfo - `setStrictnessInfo` mkStrictnessInfo ([wwStrict], True) - - -- these "bottom" out, no matter what their arguments - -generic_ERROR_ID u n = pc_bottoming_Id u pREL_ERR n errorTy - --- Very useful... -noCafIdInfo = vanillaIdInfo `setCafInfo` NoCafRefs + bottoming_info = vanillaIdInfo `setAllStrictnessInfo` Just strict_sig + -- Do *not* mark them as NoCafRefs, because they can indeed have + -- CAF refs. For example, pAT_ERROR_ID calls GHC.Err.untangle, + -- which has some CAFs + -- In due course we may arrange that these error-y things are + -- regarded by the GC as permanently live, in which case we + -- can give them NoCaf info. As it is, any function that calls + -- any pc_bottoming_Id will itself have CafRefs, which bloats + -- SRTs. + + strict_sig = mkStrictSig (mkTopDmdType [evalDmd] BotRes) + -- These "bottom" out, no matter what their arguments (openAlphaTyVar:openBetaTyVar:_) = openAlphaTyVars openAlphaTy = mkTyVarTy openAlphaTyVar openBetaTy = mkTyVarTy openBetaTyVar - -errorTy :: Type -errorTy = mkSigmaTy [openAlphaTyVar] [] (mkFunTys [mkListTy charTy] - openAlphaTy) - -- Notice the openAlphaTyVar. It says that "error" can be applied - -- to unboxed as well as boxed types. This is OK because it never - -- returns, so the return type is irrelevant. \end{code}