{- (c) The GRASP/AQUA Project, Glasgow University, 1992-1998 \section{Tidying up Core} -} {-# LANGUAGE CPP, ViewPatterns #-} module TidyPgm ( mkBootModDetailsTc, tidyProgram, globaliseAndTidyId ) where #include "HsVersions.h" import GhcPrelude import TcRnTypes import DynFlags import CoreSyn import CoreUnfold import CoreFVs import CoreTidy import CoreMonad import CorePrep import CoreUtils (rhsIsStatic) import CoreStats (coreBindsStats, CoreStats(..)) import CoreSeq (seqBinds) import CoreLint import Literal import Rules import PatSyn import ConLike import CoreArity ( exprArity, exprBotStrictness_maybe ) import StaticPtrTable import VarEnv import VarSet import Var import Id import MkId ( mkDictSelRhs ) import IdInfo import InstEnv import FamInstEnv import Type ( tidyTopType ) import Demand ( appIsBottom, isTopSig, isBottomingSig ) import BasicTypes import Name hiding (varName) import NameSet import NameEnv import NameCache import Avail import IfaceEnv import TcEnv import TcRnMonad import DataCon import TyCon import Class import Module import Packages( isDllName ) import HscTypes import Maybes import UniqSupply import ErrUtils (Severity(..)) import Outputable import SrcLoc import qualified ErrUtils as Err import Control.Monad import Data.Function import Data.List ( sortBy ) import Data.IORef ( atomicModifyIORef' ) {- Constructing the TypeEnv, Instances, Rules from which the ModIface is constructed, and which goes on to subsequent modules in --make mode. Most of the interface file is obtained simply by serialising the TypeEnv. One important consequence is that if the *interface file* has pragma info if and only if the final TypeEnv does. This is not so important for *this* module, but it's essential for ghc --make: subsequent compilations must not see (e.g.) the arity if the interface file does not contain arity If they do, they'll exploit the arity; then the arity might change, but the iface file doesn't change => recompilation does not happen => disaster. For data types, the final TypeEnv will have a TyThing for the TyCon, plus one for each DataCon; the interface file will contain just one data type declaration, but it is de-serialised back into a collection of TyThings. ************************************************************************ * * Plan A: simpleTidyPgm * * ************************************************************************ Plan A: mkBootModDetails: omit pragmas, make interfaces small ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * Ignore the bindings * Drop all WiredIn things from the TypeEnv (we never want them in interface files) * Retain all TyCons and Classes in the TypeEnv, to avoid having to find which ones are mentioned in the types of exported Ids * Trim off the constructors of non-exported TyCons, both from the TyCon and from the TypeEnv * Drop non-exported Ids from the TypeEnv * Tidy the types of the DFunIds of Instances, make them into GlobalIds, (they already have External Names) and add them to the TypeEnv * Tidy the types of the (exported) Ids in the TypeEnv, make them into GlobalIds (they already have External Names) * Drop rules altogether * Tidy the bindings, to ensure that the Caf and Arity information is correct for each top-level binder; the code generator needs it. And to ensure that local names have distinct OccNames in case of object-file splitting * If this an hsig file, drop the instances altogether too (they'll get pulled in by the implicit module import. -} -- This is Plan A: make a small type env when typechecking only, -- or when compiling a hs-boot file, or simply when not using -O -- -- We don't look at the bindings at all -- there aren't any -- for hs-boot files mkBootModDetailsTc :: HscEnv -> TcGblEnv -> IO ModDetails mkBootModDetailsTc hsc_env TcGblEnv{ tcg_exports = exports, tcg_type_env = type_env, -- just for the Ids tcg_tcs = tcs, tcg_patsyns = pat_syns, tcg_insts = insts, tcg_fam_insts = fam_insts, tcg_mod = this_mod } = -- This timing isn't terribly useful since the result isn't forced, but -- the message is useful to locating oneself in the compilation process. Err.withTiming (pure dflags) (text "CoreTidy"<+>brackets (ppr this_mod)) (const ()) $ do { let { insts' = map (tidyClsInstDFun globaliseAndTidyId) insts ; pat_syns' = map (tidyPatSynIds globaliseAndTidyId) pat_syns ; type_env1 = mkBootTypeEnv (availsToNameSet exports) (typeEnvIds type_env) tcs fam_insts ; type_env2 = extendTypeEnvWithPatSyns pat_syns' type_env1 ; dfun_ids = map instanceDFunId insts' ; type_env' = extendTypeEnvWithIds type_env2 dfun_ids } ; return (ModDetails { md_types = type_env' , md_insts = insts' , md_fam_insts = fam_insts , md_rules = [] , md_anns = [] , md_exports = exports , md_complete_sigs = [] }) } where dflags = hsc_dflags hsc_env mkBootTypeEnv :: NameSet -> [Id] -> [TyCon] -> [FamInst] -> TypeEnv mkBootTypeEnv exports ids tcs fam_insts = tidyTypeEnv True $ typeEnvFromEntities final_ids tcs fam_insts where -- Find the LocalIds in the type env that are exported -- Make them into GlobalIds, and tidy their types -- -- It's very important to remove the non-exported ones -- because we don't tidy the OccNames, and if we don't remove -- the non-exported ones we'll get many things with the -- same name in the interface file, giving chaos. -- -- Do make sure that we keep Ids that are already Global. -- When typechecking an .hs-boot file, the Ids come through as -- GlobalIds. final_ids = [ (if isLocalId id then globaliseAndTidyId id else id) `setIdUnfolding` BootUnfolding | id <- ids , keep_it id ] -- default methods have their export flag set, but everything -- else doesn't (yet), because this is pre-desugaring, so we -- must test both. keep_it id = isExportedId id || idName id `elemNameSet` exports globaliseAndTidyId :: Id -> Id -- Takes a LocalId with an External Name, -- makes it into a GlobalId -- * unchanged Name (might be Internal or External) -- * unchanged details -- * VanillaIdInfo (makes a conservative assumption about Caf-hood) globaliseAndTidyId id = Id.setIdType (globaliseId id) tidy_type where tidy_type = tidyTopType (idType id) {- ************************************************************************ * * Plan B: tidy bindings, make TypeEnv full of IdInfo * * ************************************************************************ Plan B: include pragmas, make interfaces ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * Step 1: Figure out which Ids are externally visible See Note [Choosing external Ids] * Step 2: Gather the externally visible rules, separately from the top-level bindings. See Note [Finding external rules] * Step 3: Tidy the bindings, externalising appropriate Ids See Note [Tidy the top-level bindings] * Drop all Ids from the TypeEnv, and add all the External Ids from the bindings. (This adds their IdInfo to the TypeEnv; and adds floated-out Ids that weren't even in the TypeEnv before.) Note [Choosing external Ids] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ See also the section "Interface stability" in the RecompilationAvoidance commentary: http://ghc.haskell.org/trac/ghc/wiki/Commentary/Compiler/RecompilationAvoidance First we figure out which Ids are "external" Ids. An "external" Id is one that is visible from outside the compilation unit. These are a) the user exported ones b) the ones bound to static forms c) ones mentioned in the unfoldings, workers, or rules of externally-visible ones While figuring out which Ids are external, we pick a "tidy" OccName for each one. That is, we make its OccName distinct from the other external OccNames in this module, so that in interface files and object code we can refer to it unambiguously by its OccName. The OccName for each binder is prefixed by the name of the exported Id that references it; e.g. if "f" references "x" in its unfolding, then "x" is renamed to "f_x". This helps distinguish the different "x"s from each other, and means that if "f" is later removed, things that depend on the other "x"s will not need to be recompiled. Of course, if there are multiple "f_x"s, then we have to disambiguate somehow; we use "f_x0", "f_x1" etc. As far as possible we should assign names in a deterministic fashion. Each time this module is compiled with the same options, we should end up with the same set of external names with the same types. That is, the ABI hash in the interface should not change. This turns out to be quite tricky, since the order of the bindings going into the tidy phase is already non-deterministic, as it is based on the ordering of Uniques, which are assigned unpredictably. To name things in a stable way, we do a depth-first-search of the bindings, starting from the exports sorted by name. This way, as long as the bindings themselves are deterministic (they sometimes aren't!), the order in which they are presented to the tidying phase does not affect the names we assign. Note [Tidy the top-level bindings] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Next we traverse the bindings top to bottom. For each *top-level* binder 1. Make it into a GlobalId; its IdDetails becomes VanillaGlobal, reflecting the fact that from now on we regard it as a global, not local, Id 2. Give it a system-wide Unique. [Even non-exported things need system-wide Uniques because the byte-code generator builds a single Name->BCO symbol table.] We use the NameCache kept in the HscEnv as the source of such system-wide uniques. For external Ids, use the original-name cache in the NameCache to ensure that the unique assigned is the same as the Id had in any previous compilation run. 3. Rename top-level Ids according to the names we chose in step 1. If it's an external Id, make it have a External Name, otherwise make it have an Internal Name. This is used by the code generator to decide whether to make the label externally visible 4. Give it its UTTERLY FINAL IdInfo; in ptic, * its unfolding, if it should have one * its arity, computed from the number of visible lambdas * its CAF info, computed from what is free in its RHS Finally, substitute these new top-level binders consistently throughout, including in unfoldings. We also tidy binders in RHSs, so that they print nicely in interfaces. -} tidyProgram :: HscEnv -> ModGuts -> IO (CgGuts, ModDetails) tidyProgram hsc_env (ModGuts { mg_module = mod , mg_exports = exports , mg_rdr_env = rdr_env , mg_tcs = tcs , mg_insts = cls_insts , mg_fam_insts = fam_insts , mg_binds = binds , mg_patsyns = patsyns , mg_rules = imp_rules , mg_anns = anns , mg_complete_sigs = complete_sigs , mg_deps = deps , mg_foreign = foreign_stubs , mg_foreign_files = foreign_files , mg_hpc_info = hpc_info , mg_modBreaks = modBreaks }) = Err.withTiming (pure dflags) (text "CoreTidy"<+>brackets (ppr mod)) (const ()) $ do { let { omit_prags = gopt Opt_OmitInterfacePragmas dflags ; expose_all = gopt Opt_ExposeAllUnfoldings dflags ; print_unqual = mkPrintUnqualified dflags rdr_env } ; let { type_env = typeEnvFromEntities [] tcs fam_insts ; implicit_binds = concatMap getClassImplicitBinds (typeEnvClasses type_env) ++ concatMap getTyConImplicitBinds (typeEnvTyCons type_env) } ; (unfold_env, tidy_occ_env) <- chooseExternalIds hsc_env mod omit_prags expose_all binds implicit_binds imp_rules ; let { (trimmed_binds, trimmed_rules) = findExternalRules omit_prags binds imp_rules unfold_env } ; (tidy_env, tidy_binds) <- tidyTopBinds hsc_env mod unfold_env tidy_occ_env trimmed_binds ; let { final_ids = [ id | id <- bindersOfBinds tidy_binds, isExternalName (idName id)] ; type_env1 = extendTypeEnvWithIds type_env final_ids ; tidy_cls_insts = map (tidyClsInstDFun (tidyVarOcc tidy_env)) cls_insts -- A DFunId will have a binding in tidy_binds, and so will now be in -- tidy_type_env, replete with IdInfo. Its name will be unchanged since -- it was born, but we want Global, IdInfo-rich (or not) DFunId in the -- tidy_cls_insts. Similarly the Ids inside a PatSyn. ; tidy_rules = tidyRules tidy_env trimmed_rules -- You might worry that the tidy_env contains IdInfo-rich stuff -- and indeed it does, but if omit_prags is on, ext_rules is -- empty -- Tidy the Ids inside each PatSyn, very similarly to DFunIds -- and then override the PatSyns in the type_env with the new tidy ones -- This is really the only reason we keep mg_patsyns at all; otherwise -- they could just stay in type_env ; tidy_patsyns = map (tidyPatSynIds (tidyVarOcc tidy_env)) patsyns ; type_env2 = extendTypeEnvWithPatSyns tidy_patsyns type_env1 ; tidy_type_env = tidyTypeEnv omit_prags type_env2 } -- See Note [Grand plan for static forms] in StaticPtrTable. ; (spt_entries, tidy_binds') <- sptCreateStaticBinds hsc_env mod tidy_binds ; let { spt_init_code = sptModuleInitCode mod spt_entries ; add_spt_init_code = case hscTarget dflags of -- If we are compiling for the interpreter we will insert -- any necessary SPT entries dynamically HscInterpreted -> id -- otherwise add a C stub to do so _ -> (`appendStubC` spt_init_code) } ; let { -- See Note [Injecting implicit bindings] all_tidy_binds = implicit_binds ++ tidy_binds' -- Get the TyCons to generate code for. Careful! We must use -- the untidied TypeEnv here, because we need -- (a) implicit TyCons arising from types and classes defined -- in this module -- (b) wired-in TyCons, which are normally removed from the -- TypeEnv we put in the ModDetails -- (c) Constructors even if they are not exported (the -- tidied TypeEnv has trimmed these away) ; alg_tycons = filter isAlgTyCon (typeEnvTyCons type_env) } ; endPassIO hsc_env print_unqual CoreTidy all_tidy_binds tidy_rules -- If the endPass didn't print the rules, but ddump-rules is -- on, print now ; unless (dopt Opt_D_dump_simpl dflags) $ Err.dumpIfSet_dyn dflags Opt_D_dump_rules (showSDoc dflags (ppr CoreTidy <+> text "rules")) (pprRulesForUser dflags tidy_rules) -- Print one-line size info ; let cs = coreBindsStats tidy_binds ; when (dopt Opt_D_dump_core_stats dflags) (putLogMsg dflags NoReason SevDump noSrcSpan (defaultDumpStyle dflags) (text "Tidy size (terms,types,coercions)" <+> ppr (moduleName mod) <> colon <+> int (cs_tm cs) <+> int (cs_ty cs) <+> int (cs_co cs) )) ; return (CgGuts { cg_module = mod, cg_tycons = alg_tycons, cg_binds = all_tidy_binds, cg_foreign = add_spt_init_code foreign_stubs, cg_foreign_files = foreign_files, cg_dep_pkgs = map fst $ dep_pkgs deps, cg_hpc_info = hpc_info, cg_modBreaks = modBreaks, cg_spt_entries = spt_entries }, ModDetails { md_types = tidy_type_env, md_rules = tidy_rules, md_insts = tidy_cls_insts, md_fam_insts = fam_insts, md_exports = exports, md_anns = anns, -- are already tidy md_complete_sigs = complete_sigs }) } where dflags = hsc_dflags hsc_env tidyTypeEnv :: Bool -- Compiling without -O, so omit prags -> TypeEnv -> TypeEnv -- The competed type environment is gotten from -- a) the types and classes defined here (plus implicit things) -- b) adding Ids with correct IdInfo, including unfoldings, -- gotten from the bindings -- From (b) we keep only those Ids with External names; -- the CoreTidy pass makes sure these are all and only -- the externally-accessible ones -- This truncates the type environment to include only the -- exported Ids and things needed from them, which saves space -- -- See Note [Don't attempt to trim data types] tidyTypeEnv omit_prags type_env = let type_env1 = filterNameEnv (not . isWiredInName . getName) type_env -- (1) remove wired-in things type_env2 | omit_prags = mapNameEnv trimThing type_env1 | otherwise = type_env1 -- (2) trimmed if necessary in type_env2 -------------------------- trimThing :: TyThing -> TyThing -- Trim off inessentials, for boot files and no -O trimThing (AnId id) | not (isImplicitId id) = AnId (id `setIdInfo` vanillaIdInfo) trimThing other_thing = other_thing extendTypeEnvWithPatSyns :: [PatSyn] -> TypeEnv -> TypeEnv extendTypeEnvWithPatSyns tidy_patsyns type_env = extendTypeEnvList type_env [AConLike (PatSynCon ps) | ps <- tidy_patsyns ] {- Note [Don't attempt to trim data types] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ For some time GHC tried to avoid exporting the data constructors of a data type if it wasn't strictly necessary to do so; see Trac #835. But "strictly necessary" accumulated a longer and longer list of exceptions, and finally I gave up the battle: commit 9a20e540754fc2af74c2e7392f2786a81d8d5f11 Author: Simon Peyton Jones Date: Thu Dec 6 16:03:16 2012 +0000 Stop attempting to "trim" data types in interface files Without -O, we previously tried to make interface files smaller by not including the data constructors of data types. But there are a lot of exceptions, notably when Template Haskell is involved or, more recently, DataKinds. However Trac #7445 shows that even without TemplateHaskell, using the Data class and invoking Language.Haskell.TH.Quote.dataToExpQ is enough to require us to expose the data constructors. So I've given up on this "optimisation" -- it's probably not important anyway. Now I'm simply not attempting to trim off the data constructors. The gain in simplicity is worth the modest cost in interface file growth, which is limited to the bits reqd to describe those data constructors. ************************************************************************ * * Implicit bindings * * ************************************************************************ Note [Injecting implicit bindings] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We inject the implicit bindings right at the end, in CoreTidy. Some of these bindings, notably record selectors, are not constructed in an optimised form. E.g. record selector for data T = MkT { x :: {-# UNPACK #-} !Int } Then the unfolding looks like x = \t. case t of MkT x1 -> let x = I# x1 in x This generates bad code unless it's first simplified a bit. That is why CoreUnfold.mkImplicitUnfolding uses simpleOptExpr to do a bit of optimisation first. (Only matters when the selector is used curried; eg map x ys.) See Trac #2070. [Oct 09: in fact, record selectors are no longer implicit Ids at all, because we really do want to optimise them properly. They are treated much like any other Id. But doing "light" optimisation on an implicit Id still makes sense.] At one time I tried injecting the implicit bindings *early*, at the beginning of SimplCore. But that gave rise to real difficulty, because GlobalIds are supposed to have *fixed* IdInfo, but the simplifier and other core-to-core passes mess with IdInfo all the time. The straw that broke the camels back was when a class selector got the wrong arity -- ie the simplifier gave it arity 2, whereas importing modules were expecting it to have arity 1 (Trac #2844). It's much safer just to inject them right at the end, after tidying. Oh: two other reasons for injecting them late: - If implicit Ids are already in the bindings when we start TidyPgm, we'd have to be careful not to treat them as external Ids (in the sense of chooseExternalIds); else the Ids mentioned in *their* RHSs will be treated as external and you get an interface file saying a18 = but nothing referring to a18 (because the implicit Id is the one that does, and implicit Ids don't appear in interface files). - More seriously, the tidied type-envt will include the implicit Id replete with a18 in its unfolding; but we won't take account of a18 when computing a fingerprint for the class; result chaos. There is one sort of implicit binding that is injected still later, namely those for data constructor workers. Reason (I think): it's really just a code generation trick.... binding itself makes no sense. See Note [Data constructor workers] in CorePrep. -} getTyConImplicitBinds :: TyCon -> [CoreBind] getTyConImplicitBinds tc = map get_defn (mapMaybe dataConWrapId_maybe (tyConDataCons tc)) getClassImplicitBinds :: Class -> [CoreBind] getClassImplicitBinds cls = [ NonRec op (mkDictSelRhs cls val_index) | (op, val_index) <- classAllSelIds cls `zip` [0..] ] get_defn :: Id -> CoreBind get_defn id = NonRec id (unfoldingTemplate (realIdUnfolding id)) {- ************************************************************************ * * \subsection{Step 1: finding externals} * * ************************************************************************ See Note [Choosing external Ids]. -} type UnfoldEnv = IdEnv (Name{-new name-}, Bool {-show unfolding-}) -- Maps each top-level Id to its new Name (the Id is tidied in step 2) -- The Unique is unchanged. If the new Name is external, it will be -- visible in the interface file. -- -- Bool => expose unfolding or not. chooseExternalIds :: HscEnv -> Module -> Bool -> Bool -> [CoreBind] -> [CoreBind] -> [CoreRule] -> IO (UnfoldEnv, TidyOccEnv) -- Step 1 from the notes above chooseExternalIds hsc_env mod omit_prags expose_all binds implicit_binds imp_id_rules = do { (unfold_env1,occ_env1) <- search init_work_list emptyVarEnv init_occ_env ; let internal_ids = filter (not . (`elemVarEnv` unfold_env1)) binders ; tidy_internal internal_ids unfold_env1 occ_env1 } where nc_var = hsc_NC hsc_env -- init_ext_ids is the initial list of Ids that should be -- externalised. It serves as the starting point for finding a -- deterministic, tidy, renaming for all external Ids in this -- module. -- -- It is sorted, so that it has adeterministic order (i.e. it's the -- same list every time this module is compiled), in contrast to the -- bindings, which are ordered non-deterministically. init_work_list = zip init_ext_ids init_ext_ids init_ext_ids = sortBy (compare `on` getOccName) $ filter is_external binders -- An Id should be external if either (a) it is exported, -- (b) it appears in the RHS of a local rule for an imported Id, or -- See Note [Which rules to expose] is_external id = isExportedId id || id `elemVarSet` rule_rhs_vars rule_rhs_vars = mapUnionVarSet ruleRhsFreeVars imp_id_rules binders = map fst $ flattenBinds binds implicit_binders = bindersOfBinds implicit_binds binder_set = mkVarSet binders avoids = [getOccName name | bndr <- binders ++ implicit_binders, let name = idName bndr, isExternalName name ] -- In computing our "avoids" list, we must include -- all implicit Ids -- all things with global names (assigned once and for -- all by the renamer) -- since their names are "taken". -- The type environment is a convenient source of such things. -- In particular, the set of binders doesn't include -- implicit Ids at this stage. -- We also make sure to avoid any exported binders. Consider -- f{-u1-} = 1 -- Local decl -- ... -- f{-u2-} = 2 -- Exported decl -- -- The second exported decl must 'get' the name 'f', so we -- have to put 'f' in the avoids list before we get to the first -- decl. tidyTopId then does a no-op on exported binders. init_occ_env = initTidyOccEnv avoids search :: [(Id,Id)] -- The work-list: (external id, referring id) -- Make a tidy, external Name for the external id, -- add it to the UnfoldEnv, and do the same for the -- transitive closure of Ids it refers to -- The referring id is used to generate a tidy --- name for the external id -> UnfoldEnv -- id -> (new Name, show_unfold) -> TidyOccEnv -- occ env for choosing new Names -> IO (UnfoldEnv, TidyOccEnv) search [] unfold_env occ_env = return (unfold_env, occ_env) search ((idocc,referrer) : rest) unfold_env occ_env | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env | otherwise = do (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc let (new_ids, show_unfold) | omit_prags = ([], False) | otherwise = addExternal expose_all refined_id -- 'idocc' is an *occurrence*, but we need to see the -- unfolding in the *definition*; so look up in binder_set refined_id = case lookupVarSet binder_set idocc of Just id -> id Nothing -> WARN( True, ppr idocc ) idocc unfold_env' = extendVarEnv unfold_env idocc (name',show_unfold) referrer' | isExportedId refined_id = refined_id | otherwise = referrer -- search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env' tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv -> IO (UnfoldEnv, TidyOccEnv) tidy_internal [] unfold_env occ_env = return (unfold_env,occ_env) tidy_internal (id:ids) unfold_env occ_env = do (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id let unfold_env' = extendVarEnv unfold_env id (name',False) tidy_internal ids unfold_env' occ_env' addExternal :: Bool -> Id -> ([Id], Bool) addExternal expose_all id = (new_needed_ids, show_unfold) where new_needed_ids = bndrFvsInOrder show_unfold id idinfo = idInfo id show_unfold = show_unfolding (unfoldingInfo idinfo) never_active = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo)) loop_breaker = isStrongLoopBreaker (occInfo idinfo) bottoming_fn = isBottomingSig (strictnessInfo idinfo) -- Stuff to do with the Id's unfolding -- We leave the unfolding there even if there is a worker -- In GHCi the unfolding is used by importers show_unfolding (CoreUnfolding { uf_src = src, uf_guidance = guidance }) = expose_all -- 'expose_all' says to expose all -- unfoldings willy-nilly || isStableSource src -- Always expose things whose -- source is an inline rule || not (bottoming_fn -- No need to inline bottom functions || never_active -- Or ones that say not to || loop_breaker -- Or that are loop breakers || neverUnfoldGuidance guidance) show_unfolding (DFunUnfolding {}) = True show_unfolding _ = False {- ************************************************************************ * * Deterministic free variables * * ************************************************************************ We want a deterministic free-variable list. exprFreeVars gives us a VarSet, which is in a non-deterministic order when converted to a list. Hence, here we define a free-variable finder that returns the free variables in the order that they are encountered. See Note [Choosing external Ids] -} bndrFvsInOrder :: Bool -> Id -> [Id] bndrFvsInOrder show_unfold id = run (dffvLetBndr show_unfold id) run :: DFFV () -> [Id] run (DFFV m) = case m emptyVarSet (emptyVarSet, []) of ((_,ids),_) -> ids newtype DFFV a = DFFV (VarSet -- Envt: non-top-level things that are in scope -- we don't want to record these as free vars -> (VarSet, [Var]) -- Input State: (set, list) of free vars so far -> ((VarSet,[Var]),a)) -- Output state instance Functor DFFV where fmap = liftM instance Applicative DFFV where pure a = DFFV $ \_ st -> (st, a) (<*>) = ap instance Monad DFFV where (DFFV m) >>= k = DFFV $ \env st -> case m env st of (st',a) -> case k a of DFFV f -> f env st' extendScope :: Var -> DFFV a -> DFFV a extendScope v (DFFV f) = DFFV (\env st -> f (extendVarSet env v) st) extendScopeList :: [Var] -> DFFV a -> DFFV a extendScopeList vs (DFFV f) = DFFV (\env st -> f (extendVarSetList env vs) st) insert :: Var -> DFFV () insert v = DFFV $ \ env (set, ids) -> let keep_me = isLocalId v && not (v `elemVarSet` env) && not (v `elemVarSet` set) in if keep_me then ((extendVarSet set v, v:ids), ()) else ((set, ids), ()) dffvExpr :: CoreExpr -> DFFV () dffvExpr (Var v) = insert v dffvExpr (App e1 e2) = dffvExpr e1 >> dffvExpr e2 dffvExpr (Lam v e) = extendScope v (dffvExpr e) dffvExpr (Tick (Breakpoint _ ids) e) = mapM_ insert ids >> dffvExpr e dffvExpr (Tick _other e) = dffvExpr e dffvExpr (Cast e _) = dffvExpr e dffvExpr (Let (NonRec x r) e) = dffvBind (x,r) >> extendScope x (dffvExpr e) dffvExpr (Let (Rec prs) e) = extendScopeList (map fst prs) $ (mapM_ dffvBind prs >> dffvExpr e) dffvExpr (Case e b _ as) = dffvExpr e >> extendScope b (mapM_ dffvAlt as) dffvExpr _other = return () dffvAlt :: (t, [Var], CoreExpr) -> DFFV () dffvAlt (_,xs,r) = extendScopeList xs (dffvExpr r) dffvBind :: (Id, CoreExpr) -> DFFV () dffvBind(x,r) | not (isId x) = dffvExpr r | otherwise = dffvLetBndr False x >> dffvExpr r -- Pass False because we are doing the RHS right here -- If you say True you'll get *exponential* behaviour! dffvLetBndr :: Bool -> Id -> DFFV () -- Gather the free vars of the RULES and unfolding of a binder -- We always get the free vars of a *stable* unfolding, but -- for a *vanilla* one (InlineRhs), the flag controls what happens: -- True <=> get fvs of even a *vanilla* unfolding -- False <=> ignore an InlineRhs -- For nested bindings (call from dffvBind) we always say "False" because -- we are taking the fvs of the RHS anyway -- For top-level bindings (call from addExternal, via bndrFvsInOrder) -- we say "True" if we are exposing that unfolding dffvLetBndr vanilla_unfold id = do { go_unf (unfoldingInfo idinfo) ; mapM_ go_rule (ruleInfoRules (ruleInfo idinfo)) } where idinfo = idInfo id go_unf (CoreUnfolding { uf_tmpl = rhs, uf_src = src }) = case src of InlineRhs | vanilla_unfold -> dffvExpr rhs | otherwise -> return () _ -> dffvExpr rhs go_unf (DFunUnfolding { df_bndrs = bndrs, df_args = args }) = extendScopeList bndrs $ mapM_ dffvExpr args go_unf _ = return () go_rule (BuiltinRule {}) = return () go_rule (Rule { ru_bndrs = bndrs, ru_rhs = rhs }) = extendScopeList bndrs (dffvExpr rhs) {- ************************************************************************ * * findExternalRules * * ************************************************************************ Note [Finding external rules] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The complete rules are gotten by combining a) local rules for imported Ids b) rules embedded in the top-level Ids There are two complications: * Note [Which rules to expose] * Note [Trimming auto-rules] Note [Which rules to expose] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The function 'expose_rule' filters out rules that mention, on the LHS, Ids that aren't externally visible; these rules can't fire in a client module. The externally-visible binders are computed (by chooseExternalIds) assuming that all orphan rules are externalised (see init_ext_ids in function 'search'). So in fact it's a bit conservative and we may export more than we need. (It's a sort of mutual recursion.) Note [Trimming auto-rules] ~~~~~~~~~~~~~~~~~~~~~~~~~~~ Second, with auto-specialisation we may specialise local or imported dfuns or INLINE functions, and then later inline them. That may leave behind something like RULE "foo" forall d. f @ Int d = f_spec where f is either local or imported, and there is no remaining reference to f_spec except from the RULE. Now that RULE *might* be useful to an importing module, but that is purely speculative, and meanwhile the code is taking up space and codegen time. I found that binary sizes jumped by 6-10% when I started to specialise INLINE functions (again, Note [Inline specialisations] in Specialise). So it seems better to drop the binding for f_spec, and the rule itself, if the auto-generated rule is the *only* reason that it is being kept alive. (The RULE still might have been useful in the past; that is, it was the right thing to have generated it in the first place. See Note [Inline specialisations] in Specialise. But now it has served its purpose, and can be discarded.) So findExternalRules does this: * Remove all bindings that are kept alive *only* by isAutoRule rules (this is done in trim_binds) * Remove all auto rules that mention bindings that have been removed (this is done by filtering by keep_rule) NB: if a binding is kept alive for some *other* reason (e.g. f_spec is called in the final code), we keep the rule too. This stuff is the only reason for the ru_auto field in a Rule. -} findExternalRules :: Bool -- Omit pragmas -> [CoreBind] -> [CoreRule] -- Local rules for imported fns -> UnfoldEnv -- Ids that are exported, so we need their rules -> ([CoreBind], [CoreRule]) -- See Note [Finding external rules] findExternalRules omit_prags binds imp_id_rules unfold_env = (trimmed_binds, filter keep_rule all_rules) where imp_rules = filter expose_rule imp_id_rules imp_user_rule_fvs = mapUnionVarSet user_rule_rhs_fvs imp_rules user_rule_rhs_fvs rule | isAutoRule rule = emptyVarSet | otherwise = ruleRhsFreeVars rule (trimmed_binds, local_bndrs, _, all_rules) = trim_binds binds keep_rule rule = ruleFreeVars rule `subVarSet` local_bndrs -- Remove rules that make no sense, because they mention a -- local binder (on LHS or RHS) that we have now discarded. -- (NB: ruleFreeVars only includes LocalIds) -- -- LHS: we have already filtered out rules that mention internal Ids -- on LHS but that isn't enough because we might have by now -- discarded a binding with an external Id. (How? -- chooseExternalIds is a bit conservative.) -- -- RHS: the auto rules that might mention a binder that has -- been discarded; see Note [Trimming auto-rules] expose_rule rule | omit_prags = False | otherwise = all is_external_id (ruleLhsFreeIdsList rule) -- Don't expose a rule whose LHS mentions a locally-defined -- Id that is completely internal (i.e. not visible to an -- importing module). NB: ruleLhsFreeIds only returns LocalIds. -- See Note [Which rules to expose] is_external_id id = case lookupVarEnv unfold_env id of Just (name, _) -> isExternalName name Nothing -> False trim_binds :: [CoreBind] -> ( [CoreBind] -- Trimmed bindings , VarSet -- Binders of those bindings , VarSet -- Free vars of those bindings + rhs of user rules -- (we don't bother to delete the binders) , [CoreRule]) -- All rules, imported + from the bindings -- This function removes unnecessary bindings, and gathers up rules from -- the bindings we keep. See Note [Trimming auto-rules] trim_binds [] -- Base case, start with imp_user_rule_fvs = ([], emptyVarSet, imp_user_rule_fvs, imp_rules) trim_binds (bind:binds) | any needed bndrs -- Keep binding = ( bind : binds', bndr_set', needed_fvs', local_rules ++ rules ) | otherwise -- Discard binding altogether = stuff where stuff@(binds', bndr_set, needed_fvs, rules) = trim_binds binds needed bndr = isExportedId bndr || bndr `elemVarSet` needed_fvs bndrs = bindersOf bind rhss = rhssOfBind bind bndr_set' = bndr_set `extendVarSetList` bndrs needed_fvs' = needed_fvs `unionVarSet` mapUnionVarSet idUnfoldingVars bndrs `unionVarSet` -- Ignore type variables in the type of bndrs mapUnionVarSet exprFreeVars rhss `unionVarSet` mapUnionVarSet user_rule_rhs_fvs local_rules -- In needed_fvs', we don't bother to delete binders from the fv set local_rules = [ rule | id <- bndrs , is_external_id id -- Only collect rules for external Ids , rule <- idCoreRules id , expose_rule rule ] -- and ones that can fire in a client {- ************************************************************************ * * tidyTopName * * ************************************************************************ This is where we set names to local/global based on whether they really are externally visible (see comment at the top of this module). If the name was previously local, we have to give it a unique occurrence name if we intend to externalise it. -} tidyTopName :: Module -> IORef NameCache -> Maybe Id -> TidyOccEnv -> Id -> IO (TidyOccEnv, Name) tidyTopName mod nc_var maybe_ref occ_env id | global && internal = return (occ_env, localiseName name) | global && external = return (occ_env, name) -- Global names are assumed to have been allocated by the renamer, -- so they already have the "right" unique -- And it's a system-wide unique too -- Now we get to the real reason that all this is in the IO Monad: -- we have to update the name cache in a nice atomic fashion | local && internal = do { new_local_name <- atomicModifyIORef' nc_var mk_new_local ; return (occ_env', new_local_name) } -- Even local, internal names must get a unique occurrence, because -- if we do -split-objs we externalise the name later, in the code generator -- -- Similarly, we must make sure it has a system-wide Unique, because -- the byte-code generator builds a system-wide Name->BCO symbol table | local && external = do { new_external_name <- atomicModifyIORef' nc_var mk_new_external ; return (occ_env', new_external_name) } | otherwise = panic "tidyTopName" where name = idName id external = isJust maybe_ref global = isExternalName name local = not global internal = not external loc = nameSrcSpan name old_occ = nameOccName name new_occ | Just ref <- maybe_ref , ref /= id = mkOccName (occNameSpace old_occ) $ let ref_str = occNameString (getOccName ref) occ_str = occNameString old_occ in case occ_str of '$':'w':_ -> occ_str -- workers: the worker for a function already -- includes the occname for its parent, so there's -- no need to prepend the referrer. _other | isSystemName name -> ref_str | otherwise -> ref_str ++ '_' : occ_str -- If this name was system-generated, then don't bother -- to retain its OccName, just use the referrer. These -- system-generated names will become "f1", "f2", etc. for -- a referrer "f". | otherwise = old_occ (occ_env', occ') = tidyOccName occ_env new_occ mk_new_local nc = (nc { nsUniqs = us }, mkInternalName uniq occ' loc) where (uniq, us) = takeUniqFromSupply (nsUniqs nc) mk_new_external nc = allocateGlobalBinder nc mod occ' loc -- If we want to externalise a currently-local name, check -- whether we have already assigned a unique for it. -- If so, use it; if not, extend the table. -- All this is done by allcoateGlobalBinder. -- This is needed when *re*-compiling a module in GHCi; we must -- use the same name for externally-visible things as we did before. {- ************************************************************************ * * \subsection{Step 2: top-level tidying} * * ************************************************************************ -} -- TopTidyEnv: when tidying we need to know -- * nc_var: The NameCache, containing a unique supply and any pre-ordained Names. -- These may have arisen because the -- renamer read in an interface file mentioning M.$wf, say, -- and assigned it unique r77. If, on this compilation, we've -- invented an Id whose name is $wf (but with a different unique) -- we want to rename it to have unique r77, so that we can do easy -- comparisons with stuff from the interface file -- -- * occ_env: The TidyOccEnv, which tells us which local occurrences -- are 'used' -- -- * subst_env: A Var->Var mapping that substitutes the new Var for the old tidyTopBinds :: HscEnv -> Module -> UnfoldEnv -> TidyOccEnv -> CoreProgram -> IO (TidyEnv, CoreProgram) tidyTopBinds hsc_env this_mod unfold_env init_occ_env binds = do mkIntegerId <- lookupMkIntegerName dflags hsc_env mkNaturalId <- lookupMkNaturalName dflags hsc_env integerSDataCon <- lookupIntegerSDataConName dflags hsc_env naturalSDataCon <- lookupNaturalSDataConName dflags hsc_env let cvt_literal nt i = case nt of LitNumInteger -> Just (cvtLitInteger dflags mkIntegerId integerSDataCon i) LitNumNatural -> Just (cvtLitNatural dflags mkNaturalId naturalSDataCon i) _ -> Nothing result = tidy cvt_literal init_env binds seqBinds (snd result) `seq` return result -- This seqBinds avoids a spike in space usage (see #13564) where dflags = hsc_dflags hsc_env init_env = (init_occ_env, emptyVarEnv) tidy _ env [] = (env, []) tidy cvt_literal env (b:bs) = let (env1, b') = tidyTopBind dflags this_mod cvt_literal unfold_env env b (env2, bs') = tidy cvt_literal env1 bs in (env2, b':bs') ------------------------ tidyTopBind :: DynFlags -> Module -> (LitNumType -> Integer -> Maybe CoreExpr) -> UnfoldEnv -> TidyEnv -> CoreBind -> (TidyEnv, CoreBind) tidyTopBind dflags this_mod cvt_literal unfold_env (occ_env,subst1) (NonRec bndr rhs) = (tidy_env2, NonRec bndr' rhs') where Just (name',show_unfold) = lookupVarEnv unfold_env bndr caf_info = hasCafRefs dflags this_mod (subst1, cvt_literal) (idArity bndr) rhs (bndr', rhs') = tidyTopPair dflags show_unfold tidy_env2 caf_info name' (bndr, rhs) subst2 = extendVarEnv subst1 bndr bndr' tidy_env2 = (occ_env, subst2) tidyTopBind dflags this_mod cvt_literal unfold_env (occ_env, subst1) (Rec prs) = (tidy_env2, Rec prs') where prs' = [ tidyTopPair dflags show_unfold tidy_env2 caf_info name' (id,rhs) | (id,rhs) <- prs, let (name',show_unfold) = expectJust "tidyTopBind" $ lookupVarEnv unfold_env id ] subst2 = extendVarEnvList subst1 (bndrs `zip` map fst prs') tidy_env2 = (occ_env, subst2) bndrs = map fst prs -- the CafInfo for a recursive group says whether *any* rhs in -- the group may refer indirectly to a CAF (because then, they all do). caf_info | or [ mayHaveCafRefs (hasCafRefs dflags this_mod (subst1, cvt_literal) (idArity bndr) rhs) | (bndr,rhs) <- prs ] = MayHaveCafRefs | otherwise = NoCafRefs ----------------------------------------------------------- tidyTopPair :: DynFlags -> Bool -- show unfolding -> TidyEnv -- The TidyEnv is used to tidy the IdInfo -- It is knot-tied: don't look at it! -> CafInfo -> Name -- New name -> (Id, CoreExpr) -- Binder and RHS before tidying -> (Id, CoreExpr) -- This function is the heart of Step 2 -- The rec_tidy_env is the one to use for the IdInfo -- It's necessary because when we are dealing with a recursive -- group, a variable late in the group might be mentioned -- in the IdInfo of one early in the group tidyTopPair dflags show_unfold rhs_tidy_env caf_info name' (bndr, rhs) = (bndr1, rhs1) where bndr1 = mkGlobalId details name' ty' idinfo' details = idDetails bndr -- Preserve the IdDetails ty' = tidyTopType (idType bndr) rhs1 = tidyExpr rhs_tidy_env rhs idinfo' = tidyTopIdInfo dflags rhs_tidy_env name' rhs rhs1 (idInfo bndr) show_unfold caf_info -- tidyTopIdInfo creates the final IdInfo for top-level -- binders. There are two delicate pieces: -- -- * Arity. After CoreTidy, this arity must not change any more. -- Indeed, CorePrep must eta expand where necessary to make -- the manifest arity equal to the claimed arity. -- -- * CAF info. This must also remain valid through to code generation. -- We add the info here so that it propagates to all -- occurrences of the binders in RHSs, and hence to occurrences in -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES. -- CoreToStg makes use of this when constructing SRTs. tidyTopIdInfo :: DynFlags -> TidyEnv -> Name -> CoreExpr -> CoreExpr -> IdInfo -> Bool -> CafInfo -> IdInfo tidyTopIdInfo dflags rhs_tidy_env name orig_rhs tidy_rhs idinfo show_unfold caf_info | not is_external -- For internal Ids (not externally visible) = vanillaIdInfo -- we only need enough info for code generation -- Arity and strictness info are enough; -- c.f. CoreTidy.tidyLetBndr `setCafInfo` caf_info `setArityInfo` arity `setStrictnessInfo` final_sig `setUnfoldingInfo` minimal_unfold_info -- See note [Preserve evaluatedness] -- in CoreTidy | otherwise -- Externally-visible Ids get the whole lot = vanillaIdInfo `setCafInfo` caf_info `setArityInfo` arity `setStrictnessInfo` final_sig `setOccInfo` robust_occ_info `setInlinePragInfo` (inlinePragInfo idinfo) `setUnfoldingInfo` unfold_info -- NB: we throw away the Rules -- They have already been extracted by findExternalRules where is_external = isExternalName name --------- OccInfo ------------ robust_occ_info = zapFragileOcc (occInfo idinfo) -- It's important to keep loop-breaker information -- when we are doing -fexpose-all-unfoldings --------- Strictness ------------ mb_bot_str = exprBotStrictness_maybe orig_rhs sig = strictnessInfo idinfo final_sig | not $ isTopSig sig = WARN( _bottom_hidden sig , ppr name ) sig -- try a cheap-and-cheerful bottom analyser | Just (_, nsig) <- mb_bot_str = nsig | otherwise = sig _bottom_hidden id_sig = case mb_bot_str of Nothing -> False Just (arity, _) -> not (appIsBottom id_sig arity) --------- Unfolding ------------ unf_info = unfoldingInfo idinfo unfold_info | show_unfold = tidyUnfolding rhs_tidy_env unf_info unf_from_rhs | otherwise = minimal_unfold_info minimal_unfold_info = zapUnfolding unf_info unf_from_rhs = mkTopUnfolding dflags is_bot tidy_rhs is_bot = isBottomingSig final_sig -- NB: do *not* expose the worker if show_unfold is off, -- because that means this thing is a loop breaker or -- marked NOINLINE or something like that -- This is important: if you expose the worker for a loop-breaker -- then you can make the simplifier go into an infinite loop, because -- in effect the unfolding is exposed. See Trac #1709 -- -- You might think that if show_unfold is False, then the thing should -- not be w/w'd in the first place. But a legitimate reason is this: -- the function returns bottom -- In this case, show_unfold will be false (we don't expose unfoldings -- for bottoming functions), but we might still have a worker/wrapper -- split (see Note [Worker-wrapper for bottoming functions] in WorkWrap.hs --------- Arity ------------ -- Usually the Id will have an accurate arity on it, because -- the simplifier has just run, but not always. -- One case I found was when the last thing the simplifier -- did was to let-bind a non-atomic argument and then float -- it to the top level. So it seems more robust just to -- fix it here. arity = exprArity orig_rhs {- ************************************************************************ * * Figuring out CafInfo for an expression * * ************************************************************************ hasCafRefs decides whether a top-level closure can point into the dynamic heap. We mark such things as `MayHaveCafRefs' because this information is used to decide whether a particular closure needs to be referenced in an SRT or not. There are two reasons for setting MayHaveCafRefs: a) The RHS is a CAF: a top-level updatable thunk. b) The RHS refers to something that MayHaveCafRefs Possible improvement: In an effort to keep the number of CAFs (and hence the size of the SRTs) down, we could also look at the expression and decide whether it requires a small bounded amount of heap, so we can ignore it as a CAF. In these cases however, we would need to use an additional CAF list to keep track of non-collectable CAFs. Note [Disgusting computation of CafRefs] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We compute hasCafRefs here, because IdInfo is supposed to be finalised after TidyPgm. But CorePrep does some transformations that affect CAF-hood. So we have to *predict* the result here, which is revolting. In particular CorePrep expands Integer and Natural literals. So in the prediction code here we resort to applying the same expansion (cvt_literal). Ugh! -} type CafRefEnv = (VarEnv Id, LitNumType -> Integer -> Maybe CoreExpr) -- The env finds the Caf-ness of the Id -- The LitNumType -> Integer -> CoreExpr is the desugaring functions for -- Integer and Natural literals -- See Note [Disgusting computation of CafRefs] hasCafRefs :: DynFlags -> Module -> CafRefEnv -> Arity -> CoreExpr -> CafInfo hasCafRefs dflags this_mod (subst, cvt_literal) arity expr | is_caf || mentions_cafs = MayHaveCafRefs | otherwise = NoCafRefs where mentions_cafs = cafRefsE expr is_dynamic_name = isDllName dflags this_mod is_caf = not (arity > 0 || rhsIsStatic (targetPlatform dflags) is_dynamic_name cvt_literal expr) -- NB. we pass in the arity of the expression, which is expected -- to be calculated by exprArity. This is because exprArity -- knows how much eta expansion is going to be done by -- CorePrep later on, and we don't want to duplicate that -- knowledge in rhsIsStatic below. cafRefsE :: Expr a -> Bool cafRefsE (Var id) = cafRefsV id cafRefsE (Lit lit) = cafRefsL lit cafRefsE (App f a) = cafRefsE f || cafRefsE a cafRefsE (Lam _ e) = cafRefsE e cafRefsE (Let b e) = cafRefsEs (rhssOfBind b) || cafRefsE e cafRefsE (Case e _ _ alts) = cafRefsE e || cafRefsEs (rhssOfAlts alts) cafRefsE (Tick _n e) = cafRefsE e cafRefsE (Cast e _co) = cafRefsE e cafRefsE (Type _) = False cafRefsE (Coercion _) = False cafRefsEs :: [Expr a] -> Bool cafRefsEs [] = False cafRefsEs (e:es) = cafRefsE e || cafRefsEs es cafRefsL :: Literal -> Bool -- Don't forget that mk_integer id might have Caf refs! -- We first need to convert the Integer into its final form, to -- see whether mkInteger is used. Same for LitNatural. cafRefsL (LitNumber nt i _) = case cvt_literal nt i of Just e -> cafRefsE e Nothing -> False cafRefsL _ = False cafRefsV :: Id -> Bool cafRefsV id | not (isLocalId id) = mayHaveCafRefs (idCafInfo id) | Just id' <- lookupVarEnv subst id = mayHaveCafRefs (idCafInfo id') | otherwise = False {- ************************************************************************ * * Old, dead, type-trimming code * * ************************************************************************ We used to try to "trim off" the constructors of data types that are not exported, to reduce the size of interface files, at least without -O. But that is not always possible: see the old Note [When we can't trim types] below for exceptions. Then (Trac #7445) I realised that the TH problem arises for any data type that we have deriving( Data ), because we can invoke Language.Haskell.TH.Quote.dataToExpQ to get a TH Exp representation of a value built from that data type. You don't even need {-# LANGUAGE TemplateHaskell #-}. At this point I give up. The pain of trimming constructors just doesn't seem worth the gain. So I've dumped all the code, and am just leaving it here at the end of the module in case something like this is ever resurrected. Note [When we can't trim types] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The basic idea of type trimming is to export algebraic data types abstractly (without their data constructors) when compiling without -O, unless of course they are explicitly exported by the user. We always export synonyms, because they can be mentioned in the type of an exported Id. We could do a full dependency analysis starting from the explicit exports, but that's quite painful, and not done for now. But there are some times we can't do that, indicated by the 'no_trim_types' flag. First, Template Haskell. Consider (Trac #2386) this module M(T, makeOne) where data T = Yay String makeOne = [| Yay "Yep" |] Notice that T is exported abstractly, but makeOne effectively exports it too! A module that splices in $(makeOne) will then look for a declaration of Yay, so it'd better be there. Hence, brutally but simply, we switch off type constructor trimming if TH is enabled in this module. Second, data kinds. Consider (Trac #5912) {-# LANGUAGE DataKinds #-} module M() where data UnaryTypeC a = UnaryDataC a type Bug = 'UnaryDataC We always export synonyms, so Bug is exposed, and that means that UnaryTypeC must be too, even though it's not explicitly exported. In effect, DataKinds means that we'd need to do a full dependency analysis to see what data constructors are mentioned. But we don't do that yet. In these two cases we just switch off type trimming altogether. mustExposeTyCon :: Bool -- Type-trimming flag -> NameSet -- Exports -> TyCon -- The tycon -> Bool -- Can its rep be hidden? -- We are compiling without -O, and thus trying to write as little as -- possible into the interface file. But we must expose the details of -- any data types whose constructors or fields are exported mustExposeTyCon no_trim_types exports tc | no_trim_types -- See Note [When we can't trim types] = True | not (isAlgTyCon tc) -- Always expose synonyms (otherwise we'd have to -- figure out whether it was mentioned in the type -- of any other exported thing) = True | isEnumerationTyCon tc -- For an enumeration, exposing the constructors = True -- won't lead to the need for further exposure | isFamilyTyCon tc -- Open type family = True -- Below here we just have data/newtype decls or family instances | null data_cons -- Ditto if there are no data constructors = True -- (NB: empty data types do not count as enumerations -- see Note [Enumeration types] in TyCon | any exported_con data_cons -- Expose rep if any datacon or field is exported = True | isNewTyCon tc && isFFITy (snd (newTyConRhs tc)) = True -- Expose the rep for newtypes if the rep is an FFI type. -- For a very annoying reason. 'Foreign import' is meant to -- be able to look through newtypes transparently, but it -- can only do that if it can "see" the newtype representation | otherwise = False where data_cons = tyConDataCons tc exported_con con = any (`elemNameSet` exports) (dataConName con : dataConFieldLabels con) -}