2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section{Tidying up Core}
7 module TidyPgm( mkBootModDetailsDs, mkBootModDetailsTc,
8 tidyProgram, globaliseAndTidyId ) where
10 #include "HsVersions.h"
22 import CoreArity ( exprArity )
23 import Class ( classSelIds )
45 import FastBool hiding ( fastOr )
48 import Data.List ( sortBy )
49 import Data.IORef ( IORef, readIORef, writeIORef )
53 Constructing the TypeEnv, Instances, Rules from which the ModIface is
54 constructed, and which goes on to subsequent modules in --make mode.
56 Most of the interface file is obtained simply by serialising the
57 TypeEnv. One important consequence is that if the *interface file*
58 has pragma info if and only if the final TypeEnv does. This is not so
59 important for *this* module, but it's essential for ghc --make:
60 subsequent compilations must not see (e.g.) the arity if the interface
61 file does not contain arity If they do, they'll exploit the arity;
62 then the arity might change, but the iface file doesn't change =>
63 recompilation does not happen => disaster.
65 For data types, the final TypeEnv will have a TyThing for the TyCon,
66 plus one for each DataCon; the interface file will contain just one
67 data type declaration, but it is de-serialised back into a collection
70 %************************************************************************
74 %************************************************************************
77 Plan A: mkBootModDetails: omit pragmas, make interfaces small
78 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
81 * Drop all WiredIn things from the TypeEnv
82 (we never want them in interface files)
84 * Retain all TyCons and Classes in the TypeEnv, to avoid
85 having to find which ones are mentioned in the
88 * Trim off the constructors of non-exported TyCons, both
89 from the TyCon and from the TypeEnv
91 * Drop non-exported Ids from the TypeEnv
93 * Tidy the types of the DFunIds of Instances,
94 make them into GlobalIds, (they already have External Names)
95 and add them to the TypeEnv
97 * Tidy the types of the (exported) Ids in the TypeEnv,
98 make them into GlobalIds (they already have External Names)
100 * Drop rules altogether
102 * Tidy the bindings, to ensure that the Caf and Arity
103 information is correct for each top-level binder; the
104 code generator needs it. And to ensure that local names have
105 distinct OccNames in case of object-file splitting
108 -- This is Plan A: make a small type env when typechecking only,
109 -- or when compiling a hs-boot file, or simply when not using -O
111 -- We don't look at the bindings at all -- there aren't any
114 mkBootModDetailsTc :: HscEnv -> TcGblEnv -> IO ModDetails
115 mkBootModDetailsTc hsc_env
116 TcGblEnv{ tcg_exports = exports,
117 tcg_type_env = type_env,
119 tcg_fam_insts = fam_insts
121 = mkBootModDetails hsc_env exports type_env insts fam_insts
123 mkBootModDetailsDs :: HscEnv -> ModGuts -> IO ModDetails
124 mkBootModDetailsDs hsc_env
125 ModGuts{ mg_exports = exports,
128 mg_fam_insts = fam_insts
130 = mkBootModDetails hsc_env exports type_env insts fam_insts
132 mkBootModDetails :: HscEnv -> [AvailInfo] -> NameEnv TyThing
133 -> [Instance] -> [FamInstEnv.FamInst] -> IO ModDetails
134 mkBootModDetails hsc_env exports type_env insts fam_insts
135 = do { let dflags = hsc_dflags hsc_env
136 ; showPass dflags "Tidy [hoot] type env"
138 ; let { insts' = tidyInstances globaliseAndTidyId insts
139 ; dfun_ids = map instanceDFunId insts'
140 ; type_env1 = tidyBootTypeEnv (availsToNameSet exports) type_env
141 ; type_env' = extendTypeEnvWithIds type_env1 dfun_ids
143 ; return (ModDetails { md_types = type_env'
145 , md_fam_insts = fam_insts
148 , md_exports = exports
149 , md_vect_info = noVectInfo
154 tidyBootTypeEnv :: NameSet -> TypeEnv -> TypeEnv
155 tidyBootTypeEnv exports type_env
156 = tidyTypeEnv True False exports type_env final_ids
158 -- Find the LocalIds in the type env that are exported
159 -- Make them into GlobalIds, and tidy their types
161 -- It's very important to remove the non-exported ones
162 -- because we don't tidy the OccNames, and if we don't remove
163 -- the non-exported ones we'll get many things with the
164 -- same name in the interface file, giving chaos.
165 final_ids = [ globaliseAndTidyId id
166 | id <- typeEnvIds type_env
170 -- default methods have their export flag set, but everything
171 -- else doesn't (yet), because this is pre-desugaring, so we
173 keep_it id = isExportedId id || idName id `elemNameSet` exports
177 globaliseAndTidyId :: Id -> Id
178 -- Takes an LocalId with an External Name,
179 -- makes it into a GlobalId
180 -- * unchanged Name (might be Internal or External)
181 -- * unchanged details
182 -- * VanillaIdInfo (makes a conservative assumption about Caf-hood)
183 globaliseAndTidyId id
184 = Id.setIdType (globaliseId id) tidy_type
186 tidy_type = tidyTopType (idType id)
190 %************************************************************************
192 Plan B: tidy bindings, make TypeEnv full of IdInfo
194 %************************************************************************
196 Plan B: include pragmas, make interfaces
197 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
198 * Figure out which Ids are externally visible
200 * Tidy the bindings, externalising appropriate Ids
202 * Drop all Ids from the TypeEnv, and add all the External Ids from
203 the bindings. (This adds their IdInfo to the TypeEnv; and adds
204 floated-out Ids that weren't even in the TypeEnv before.)
206 Step 1: Figure out external Ids
207 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
208 Note [choosing external names]
210 See also the section "Interface stability" in the
211 RecompilationAvoidance commentary:
212 http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/RecompilationAvoidance
214 First we figure out which Ids are "external" Ids. An
215 "external" Id is one that is visible from outside the compilation
217 a) the user exported ones
218 b) ones mentioned in the unfoldings, workers,
219 or rules of externally-visible ones
221 While figuring out which Ids are external, we pick a "tidy" OccName
222 for each one. That is, we make its OccName distinct from the other
223 external OccNames in this module, so that in interface files and
224 object code we can refer to it unambiguously by its OccName. The
225 OccName for each binder is prefixed by the name of the exported Id
226 that references it; e.g. if "f" references "x" in its unfolding, then
227 "x" is renamed to "f_x". This helps distinguish the different "x"s
228 from each other, and means that if "f" is later removed, things that
229 depend on the other "x"s will not need to be recompiled. Of course,
230 if there are multiple "f_x"s, then we have to disambiguate somehow; we
231 use "f_x0", "f_x1" etc.
233 As far as possible we should assign names in a deterministic fashion.
234 Each time this module is compiled with the same options, we should end
235 up with the same set of external names with the same types. That is,
236 the ABI hash in the interface should not change. This turns out to be
237 quite tricky, since the order of the bindings going into the tidy
238 phase is already non-deterministic, as it is based on the ordering of
239 Uniques, which are assigned unpredictably.
241 To name things in a stable way, we do a depth-first-search of the
242 bindings, starting from the exports sorted by name. This way, as long
243 as the bindings themselves are deterministic (they sometimes aren't!),
244 the order in which they are presented to the tidying phase does not
245 affect the names we assign.
247 Step 2: Tidy the program
248 ~~~~~~~~~~~~~~~~~~~~~~~~
249 Next we traverse the bindings top to bottom. For each *top-level*
252 1. Make it into a GlobalId; its IdDetails becomes VanillaGlobal,
253 reflecting the fact that from now on we regard it as a global,
256 2. Give it a system-wide Unique.
257 [Even non-exported things need system-wide Uniques because the
258 byte-code generator builds a single Name->BCO symbol table.]
260 We use the NameCache kept in the HscEnv as the
261 source of such system-wide uniques.
263 For external Ids, use the original-name cache in the NameCache
264 to ensure that the unique assigned is the same as the Id had
265 in any previous compilation run.
267 3. Rename top-level Ids according to the names we chose in step 1.
268 If it's an external Id, make it have a External Name, otherwise
269 make it have an Internal Name. This is used by the code generator
270 to decide whether to make the label externally visible
272 4. Give it its UTTERLY FINAL IdInfo; in ptic,
273 * its unfolding, if it should have one
275 * its arity, computed from the number of visible lambdas
277 * its CAF info, computed from what is free in its RHS
280 Finally, substitute these new top-level binders consistently
281 throughout, including in unfoldings. We also tidy binders in
282 RHSs, so that they print nicely in interfaces.
285 tidyProgram :: HscEnv -> ModGuts -> IO (CgGuts, ModDetails)
286 tidyProgram hsc_env (ModGuts { mg_module = mod, mg_exports = exports,
288 mg_insts = insts, mg_fam_insts = fam_insts,
290 mg_rules = imp_rules,
291 mg_vect_info = vect_info,
292 mg_dir_imps = dir_imps,
295 mg_foreign = foreign_stubs,
296 mg_hpc_info = hpc_info,
297 mg_modBreaks = modBreaks })
299 = do { let dflags = hsc_dflags hsc_env
300 ; showPass dflags "Tidy Core"
302 ; let { omit_prags = dopt Opt_OmitInterfacePragmas dflags
303 ; th = dopt Opt_TemplateHaskell dflags
306 ; let { implicit_binds = getImplicitBinds type_env }
308 ; (unfold_env, tidy_occ_env)
309 <- chooseExternalIds hsc_env mod omit_prags binds implicit_binds
313 | otherwise = findExternalRules binds imp_rules unfold_env
314 -- findExternalRules filters imp_rules to avoid binders that
315 -- aren't externally visible; but the externally-visible binders
316 -- are computed (by findExternalIds) assuming that all orphan
317 -- rules are exported (they get their Exported flag set in the desugarer)
318 -- So in fact we may export more than we need.
319 -- (It's a sort of mutual recursion.)
322 ; let { (tidy_env, tidy_binds)
323 = tidyTopBinds hsc_env unfold_env tidy_occ_env binds }
325 ; let { export_set = availsToNameSet exports
326 ; final_ids = [ id | id <- bindersOfBinds tidy_binds,
327 isExternalName (idName id)]
328 ; tidy_type_env = tidyTypeEnv omit_prags th export_set
330 ; tidy_insts = tidyInstances (lookup_dfun tidy_type_env) insts
331 -- A DFunId will have a binding in tidy_binds, and so
332 -- will now be in final_env, replete with IdInfo
333 -- Its name will be unchanged since it was born, but
334 -- we want Global, IdInfo-rich (or not) DFunId in the
337 ; tidy_rules = tidyRules tidy_env ext_rules
338 -- You might worry that the tidy_env contains IdInfo-rich stuff
339 -- and indeed it does, but if omit_prags is on, ext_rules is
342 -- See Note [Injecting implicit bindings]
343 ; all_tidy_binds = implicit_binds ++ tidy_binds
345 ; alg_tycons = filter isAlgTyCon (typeEnvTyCons type_env)
348 ; endPass dflags "Tidy Core" Opt_D_dump_simpl all_tidy_binds
349 ; dumpIfSet_core dflags Opt_D_dump_simpl
351 (pprRules tidy_rules)
353 ; let dir_imp_mods = moduleEnvKeys dir_imps
355 ; return (CgGuts { cg_module = mod,
356 cg_tycons = alg_tycons,
357 cg_binds = all_tidy_binds,
358 cg_dir_imps = dir_imp_mods,
359 cg_foreign = foreign_stubs,
360 cg_dep_pkgs = dep_pkgs deps,
361 cg_hpc_info = hpc_info,
362 cg_modBreaks = modBreaks },
364 ModDetails { md_types = tidy_type_env,
365 md_rules = tidy_rules,
366 md_insts = tidy_insts,
367 md_fam_insts = fam_insts,
368 md_exports = exports,
369 md_anns = anns, -- are already tidy
370 md_vect_info = vect_info --
374 lookup_dfun :: TypeEnv -> Var -> Id
375 lookup_dfun type_env dfun_id
376 = case lookupTypeEnv type_env (idName dfun_id) of
377 Just (AnId dfun_id') -> dfun_id'
378 _other -> pprPanic "lookup_dfun" (ppr dfun_id)
380 --------------------------
381 tidyTypeEnv :: Bool -- Compiling without -O, so omit prags
382 -> Bool -- Template Haskell is on
383 -> NameSet -> TypeEnv -> [Id] -> TypeEnv
385 -- The competed type environment is gotten from
386 -- Dropping any wired-in things, and then
387 -- a) keeping the types and classes
388 -- b) removing all Ids,
389 -- c) adding Ids with correct IdInfo, including unfoldings,
390 -- gotten from the bindings
391 -- From (c) we keep only those Ids with External names;
392 -- the CoreTidy pass makes sure these are all and only
393 -- the externally-accessible ones
394 -- This truncates the type environment to include only the
395 -- exported Ids and things needed from them, which saves space
397 tidyTypeEnv omit_prags th exports type_env final_ids
398 = let type_env1 = filterNameEnv keep_it type_env
399 type_env2 = extendTypeEnvWithIds type_env1 final_ids
400 type_env3 | omit_prags = mapNameEnv (trimThing th exports) type_env2
401 | otherwise = type_env2
405 -- We keep GlobalIds, because they won't appear
406 -- in the bindings from which final_ids are derived!
407 -- (The bindings bind LocalIds.)
408 keep_it thing | isWiredInThing thing = False
409 keep_it (AnId id) = isGlobalId id -- Keep GlobalIds (e.g. class ops)
410 keep_it _other = True -- Keep all TyCons, DataCons, and Classes
412 --------------------------
413 isWiredInThing :: TyThing -> Bool
414 isWiredInThing thing = isWiredInName (getName thing)
416 --------------------------
417 trimThing :: Bool -> NameSet -> TyThing -> TyThing
418 -- Trim off inessentials, for boot files and no -O
419 trimThing th exports (ATyCon tc)
420 | not th && not (mustExposeTyCon exports tc)
421 = ATyCon (makeTyConAbstract tc) -- Note [Trimming and Template Haskell]
423 trimThing _th _exports (AnId id)
424 | not (isImplicitId id)
425 = AnId (id `setIdInfo` vanillaIdInfo)
427 trimThing _th _exports other_thing
431 {- Note [Trimming and Template Haskell]
432 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
433 Consider (Trac #2386) this
434 module M(T, makeOne) where
436 makeOne = [| Yay "Yep" |]
437 Notice that T is exported abstractly, but makeOne effectively exports it too!
438 A module that splices in $(makeOne) will then look for a declartion of Yay,
439 so it'd better be there. Hence, brutally but simply, we switch off type
440 constructor trimming if TH is enabled in this module. -}
443 mustExposeTyCon :: NameSet -- Exports
444 -> TyCon -- The tycon
445 -> Bool -- Can its rep be hidden?
446 -- We are compiling without -O, and thus trying to write as little as
447 -- possible into the interface file. But we must expose the details of
448 -- any data types whose constructors or fields are exported
449 mustExposeTyCon exports tc
450 | not (isAlgTyCon tc) -- Synonyms
452 | isEnumerationTyCon tc -- For an enumeration, exposing the constructors
453 = True -- won't lead to the need for further exposure
454 -- (This includes data types with no constructors.)
455 | isOpenTyCon tc -- Open type family
458 | otherwise -- Newtype, datatype
459 = any exported_con (tyConDataCons tc)
460 -- Expose rep if any datacon or field is exported
462 || (isNewTyCon tc && isFFITy (snd (newTyConRhs tc)))
463 -- Expose the rep for newtypes if the rep is an FFI type.
464 -- For a very annoying reason. 'Foreign import' is meant to
465 -- be able to look through newtypes transparently, but it
466 -- can only do that if it can "see" the newtype representation
468 exported_con con = any (`elemNameSet` exports)
469 (dataConName con : dataConFieldLabels con)
471 tidyInstances :: (DFunId -> DFunId) -> [Instance] -> [Instance]
472 tidyInstances tidy_dfun ispecs
475 tidy ispec = setInstanceDFunId ispec $
476 tidy_dfun (instanceDFunId ispec)
480 %************************************************************************
484 %************************************************************************
486 Note [Injecting implicit bindings]
487 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
488 We inject the implict bindings right at the end, in CoreTidy.
489 Some of these bindings, notably record selectors, are not
490 constructed in an optimised form. E.g. record selector for
491 data T = MkT { x :: {-# UNPACK #-} !Int }
492 Then the unfolding looks like
493 x = \t. case t of MkT x1 -> let x = I# x1 in x
494 This generates bad code unless it's first simplified a bit. That is
495 why CoreUnfold.mkImplicitUnfolding uses simleExprOpt to do a bit of
496 optimisation first. (Only matters when the selector is used curried;
497 eg map x ys.) See Trac #2070.
499 At one time I tried injecting the implicit bindings *early*, at the
500 beginning of SimplCore. But that gave rise to real difficulty,
501 becuase GlobalIds are supposed to have *fixed* IdInfo, but the
502 simplifier and other core-to-core passes mess with IdInfo all the
503 time. The straw that broke the camels back was when a class selector
504 got the wrong arity -- ie the simplifier gave it arity 2, whereas
505 importing modules were expecting it to have arity 1 (Trac #2844).
506 It's much safer just to inject them right at the end, after tidying.
508 Oh: two other reasons for injecting them late:
509 - If implicit Ids are already in the bindings when we start TidyPgm,
510 we'd have to be careful not to treat them as external Ids (in
511 the sense of findExternalIds); else the Ids mentioned in *their*
512 RHSs will be treated as external and you get an interface file
514 but nothing refererring to a18 (because the implicit Id is the
517 - More seriously, the tidied type-envt will include the implicit
518 Id replete with a18 in its unfolding; but we won't take account
519 of a18 when computing a fingerprint for the class; result chaos.
523 getImplicitBinds :: TypeEnv -> [CoreBind]
524 getImplicitBinds type_env
525 = map get_defn (concatMap implicit_ids (typeEnvElts type_env))
527 implicit_ids (ATyCon tc) = mapCatMaybes dataConWrapId_maybe (tyConDataCons tc)
528 implicit_ids (AClass cls) = classSelIds cls
531 get_defn :: Id -> CoreBind
532 get_defn id = NonRec id (unfoldingTemplate (idUnfolding id))
536 %************************************************************************
538 \subsection{Step 1: finding externals}
540 %************************************************************************
542 Sete Note [choosing external names].
545 type UnfoldEnv = IdEnv (Name{-new name-}, Bool {-show unfolding-})
546 -- maps each top-level Id to its new Name (the Id is tidied in step 2)
547 -- The Unique is unchanged. If the new Id is external, it will be
548 -- visible in the interface file.
550 -- Bool => expose unfolding or not.
552 chooseExternalIds :: HscEnv
557 -> IO (UnfoldEnv, TidyOccEnv)
558 -- Step 1 from the notes above
560 chooseExternalIds hsc_env mod omit_prags binds implicit_binds
562 (unfold_env1,occ_env1)
563 <- search (zip sorted_exports sorted_exports) emptyVarEnv init_occ_env
564 let internal_ids = filter (not . (`elemVarEnv` unfold_env1)) binders
565 tidy_internal internal_ids unfold_env1 occ_env1
567 nc_var = hsc_NC hsc_env
569 -- the exports, sorted by OccName. This is a deterministic list of
570 -- Ids (i.e. it's the same list every time this module is compiled),
571 -- in contrast to the bindings, which are ordered
572 -- non-deterministically.
574 -- This list will serve as a starting point for finding a
575 -- deterministic, tidy, renaming for all external Ids in this
577 sorted_exports = sortBy (compare `on` getOccName) $
578 filter isExportedId binders
580 binders = bindersOfBinds binds
581 implicit_binders = bindersOfBinds implicit_binds
583 bind_env :: IdEnv (Id,CoreExpr)
584 bind_env = mkVarEnv (zip (map fst bs) bs) where bs = flattenBinds binds
586 avoids = [getOccName name | bndr <- binders ++ implicit_binders,
587 let name = idName bndr,
588 isExternalName name ]
589 -- In computing our "avoids" list, we must include
591 -- all things with global names (assigned once and for
592 -- all by the renamer)
593 -- since their names are "taken".
594 -- The type environment is a convenient source of such things.
595 -- In particular, the set of binders doesn't include
596 -- implicit Ids at this stage.
598 -- We also make sure to avoid any exported binders. Consider
599 -- f{-u1-} = 1 -- Local decl
601 -- f{-u2-} = 2 -- Exported decl
603 -- The second exported decl must 'get' the name 'f', so we
604 -- have to put 'f' in the avoids list before we get to the first
605 -- decl. tidyTopId then does a no-op on exported binders.
606 init_occ_env = initTidyOccEnv avoids
609 search :: [(Id,Id)] -- (external id, referrring id)
610 -> UnfoldEnv -- id -> (new Name, show_unfold)
611 -> TidyOccEnv -- occ env for choosing new Names
612 -> IO (UnfoldEnv, TidyOccEnv)
614 search [] unfold_env occ_env = return (unfold_env, occ_env)
616 search ((idocc,referrer) : rest) unfold_env occ_env
617 | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env
619 (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc
621 (id, rhs) = expectJust (showSDoc (text "chooseExternalIds: " <>
623 lookupVarEnv bind_env idocc
624 -- NB. idocc might be an *occurrence* of an Id, whereas we want
625 -- the Id from the binding site, because only the latter is
626 -- guaranteed to have the unfolding attached. This is why we
627 -- keep binding site Ids in the bind_env.
628 (new_ids, show_unfold)
629 | omit_prags = ([], False)
630 | otherwise = addExternal id rhs
631 unfold_env' = extendVarEnv unfold_env id (name',show_unfold)
632 referrer' | isExportedId id = id
633 | otherwise = referrer
635 search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env'
637 tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv
638 -> IO (UnfoldEnv, TidyOccEnv)
639 tidy_internal [] unfold_env occ_env = return (unfold_env,occ_env)
640 tidy_internal (id:ids) unfold_env occ_env = do
641 (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id
642 let unfold_env' = extendVarEnv unfold_env id (name',False)
643 tidy_internal ids unfold_env' occ_env'
645 addExternal :: Id -> CoreExpr -> ([Id],Bool)
646 addExternal id rhs = (new_needed_ids, show_unfold)
648 new_needed_ids = unfold_ids ++
649 filter (\id -> isLocalId id &&
650 not (id `elemVarSet` unfold_set))
651 (varSetElems worker_ids ++
652 varSetElems spec_ids) -- XXX non-det ordering
655 dont_inline = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo))
656 loop_breaker = isNonRuleLoopBreaker (occInfo idinfo)
657 bottoming_fn = isBottomingSig (newStrictnessInfo idinfo `orElse` topSig)
658 spec_ids = specInfoFreeVars (specInfo idinfo)
659 worker_info = workerInfo idinfo
661 -- Stuff to do with the Id's unfolding
662 -- The simplifier has put an up-to-date unfolding
663 -- in the IdInfo, but the RHS will do just as well
664 unfolding = unfoldingInfo idinfo
665 rhs_is_small = not (neverUnfold unfolding)
667 -- We leave the unfolding there even if there is a worker
668 -- In GHCI the unfolding is used by importers
669 -- When writing an interface file, we omit the unfolding
670 -- if there is a worker
671 show_unfold = not bottoming_fn && -- Not necessary
674 rhs_is_small -- Small enough
676 (unfold_set, unfold_ids)
677 | show_unfold = freeVarsInDepthFirstOrder rhs
678 | otherwise = (emptyVarSet, [])
680 worker_ids = case worker_info of
681 HasWorker work_id _ -> unitVarSet work_id
682 _otherwise -> emptyVarSet
685 -- We want a deterministic free-variable list. exprFreeVars gives us
686 -- a VarSet, which is in a non-deterministic order when converted to a
687 -- list. Hence, here we define a free-variable finder that returns
688 -- the free variables in the order that they are encountered.
690 -- Note [choosing external names]
692 freeVarsInDepthFirstOrder :: CoreExpr -> (VarSet, [Id])
693 freeVarsInDepthFirstOrder e =
695 DFFV m -> case m emptyVarSet [] of
696 (set,ids,_) -> (set,ids)
698 newtype DFFV a = DFFV (VarSet -> [Var] -> (VarSet,[Var],a))
700 instance Monad DFFV where
701 return a = DFFV $ \set ids -> (set, ids, a)
702 (DFFV m) >>= k = DFFV $ \set ids ->
704 (set',ids',a) -> case k a of
705 DFFV f -> f set' ids'
707 insert :: Var -> DFFV ()
708 insert v = DFFV $ \ set ids -> case () of
709 _ | v `elemVarSet` set -> (set,ids,())
710 | otherwise -> (extendVarSet set v, v:ids, ())
712 dffvExpr :: CoreExpr -> DFFV ()
713 dffvExpr e = go emptyVarSet e
715 go scope e = case e of
716 Var v | isLocalId v && not (v `elemVarSet` scope) -> insert v
717 App e1 e2 -> do go scope e1; go scope e2
718 Lam v e -> go (extendVarSet scope v) e
719 Note _ e -> go scope e
720 Cast e _ -> go scope e
721 Let (NonRec x r) e -> do go scope r; go (extendVarSet scope x) e
722 Let (Rec prs) e -> do let scope' = extendVarSetList scope (map fst prs)
723 mapM_ (go scope') (map snd prs)
725 Case e b _ as -> do go scope e
726 mapM_ (go_alt (extendVarSet scope b)) as
729 go_alt scope (_,xs,r) = go (extendVarSetList scope xs) r
733 --------------------------------------------------------------------
735 -- This is where we set names to local/global based on whether they really are
736 -- externally visible (see comment at the top of this module). If the name
737 -- was previously local, we have to give it a unique occurrence name if
738 -- we intend to externalise it.
741 tidyTopName :: Module -> IORef NameCache -> Maybe Id -> TidyOccEnv
742 -> Id -> IO (TidyOccEnv, Name)
743 tidyTopName mod nc_var maybe_ref occ_env id
744 | global && internal = return (occ_env, localiseName name)
746 | global && external = return (occ_env, name)
747 -- Global names are assumed to have been allocated by the renamer,
748 -- so they already have the "right" unique
749 -- And it's a system-wide unique too
751 -- Now we get to the real reason that all this is in the IO Monad:
752 -- we have to update the name cache in a nice atomic fashion
754 | local && internal = do { nc <- readIORef nc_var
755 ; let (nc', new_local_name) = mk_new_local nc
756 ; writeIORef nc_var nc'
757 ; return (occ_env', new_local_name) }
758 -- Even local, internal names must get a unique occurrence, because
759 -- if we do -split-objs we externalise the name later, in the code generator
761 -- Similarly, we must make sure it has a system-wide Unique, because
762 -- the byte-code generator builds a system-wide Name->BCO symbol table
764 | local && external = do { nc <- readIORef nc_var
765 ; let (nc', new_external_name) = mk_new_external nc
766 ; writeIORef nc_var nc'
767 ; return (occ_env', new_external_name) }
769 | otherwise = panic "tidyTopName"
772 external = isJust maybe_ref
773 global = isExternalName name
775 internal = not external
776 loc = nameSrcSpan name
778 old_occ = nameOccName name
780 | Just ref <- maybe_ref, ref /= id =
781 mkOccName (occNameSpace old_occ) $
783 ref_str = occNameString (getOccName ref)
784 occ_str = occNameString old_occ
788 -- workers: the worker for a function already
789 -- includes the occname for its parent, so there's
790 -- no need to prepend the referrer.
791 _other | isSystemName name -> ref_str
792 | otherwise -> ref_str ++ '_' : occ_str
793 -- If this name was system-generated, then don't bother
794 -- to retain its OccName, just use the referrer. These
795 -- system-generated names will become "f1", "f2", etc. for
797 | otherwise = old_occ
799 (occ_env', occ') = tidyOccName occ_env new_occ
801 mk_new_local nc = (nc { nsUniqs = us2 }, mkInternalName uniq occ' loc)
803 (us1, us2) = splitUniqSupply (nsUniqs nc)
804 uniq = uniqFromSupply us1
806 mk_new_external nc = allocateGlobalBinder nc mod occ' loc
807 -- If we want to externalise a currently-local name, check
808 -- whether we have already assigned a unique for it.
809 -- If so, use it; if not, extend the table.
810 -- All this is done by allcoateGlobalBinder.
811 -- This is needed when *re*-compiling a module in GHCi; we must
812 -- use the same name for externally-visible things as we did before.
816 findExternalRules :: [CoreBind]
817 -> [CoreRule] -- Non-local rules (i.e. ones for imported fns)
818 -> UnfoldEnv -- Ids that are exported, so we need their rules
820 -- The complete rules are gotten by combining
821 -- a) the non-local rules
822 -- b) rules embedded in the top-level Ids
823 findExternalRules binds non_local_rules unfold_env
824 = filter (not . internal_rule) (non_local_rules ++ local_rules)
827 | id <- bindersOfBinds binds,
829 rule <- idCoreRules id
833 = any (not . external_id) (varSetElems (ruleLhsFreeIds rule))
834 -- Don't export a rule whose LHS mentions a locally-defined
835 -- Id that is completely internal (i.e. not visible to an
839 | Just (name,_) <- lookupVarEnv unfold_env id = isExternalName name
845 %************************************************************************
847 \subsection{Step 2: top-level tidying}
849 %************************************************************************
853 -- TopTidyEnv: when tidying we need to know
854 -- * nc_var: The NameCache, containing a unique supply and any pre-ordained Names.
855 -- These may have arisen because the
856 -- renamer read in an interface file mentioning M.$wf, say,
857 -- and assigned it unique r77. If, on this compilation, we've
858 -- invented an Id whose name is $wf (but with a different unique)
859 -- we want to rename it to have unique r77, so that we can do easy
860 -- comparisons with stuff from the interface file
862 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
865 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
867 tidyTopBinds :: HscEnv
871 -> (TidyEnv, [CoreBind])
873 tidyTopBinds hsc_env unfold_env init_occ_env binds
874 = tidy init_env binds
876 init_env = (init_occ_env, emptyVarEnv)
878 this_pkg = thisPackage (hsc_dflags hsc_env)
880 tidy env [] = (env, [])
881 tidy env (b:bs) = let (env1, b') = tidyTopBind this_pkg unfold_env env b
882 (env2, bs') = tidy env1 bs
886 ------------------------
887 tidyTopBind :: PackageId
891 -> (TidyEnv, CoreBind)
893 tidyTopBind this_pkg unfold_env (occ_env,subst1) (NonRec bndr rhs)
894 = (tidy_env2, NonRec bndr' rhs')
896 Just (name',show_unfold) = lookupVarEnv unfold_env bndr
897 caf_info = hasCafRefs this_pkg subst1 (idArity bndr) rhs
898 (bndr', rhs') = tidyTopPair show_unfold tidy_env2 caf_info name' (bndr, rhs)
899 subst2 = extendVarEnv subst1 bndr bndr'
900 tidy_env2 = (occ_env, subst2)
902 tidyTopBind this_pkg unfold_env (occ_env,subst1) (Rec prs)
903 = (tidy_env2, Rec prs')
905 prs' = [ tidyTopPair show_unfold tidy_env2 caf_info name' (id,rhs)
907 let (name',show_unfold) =
908 expectJust "tidyTopBind" $ lookupVarEnv unfold_env id
911 subst2 = extendVarEnvList subst1 (bndrs `zip` map fst prs')
912 tidy_env2 = (occ_env, subst2)
916 -- the CafInfo for a recursive group says whether *any* rhs in
917 -- the group may refer indirectly to a CAF (because then, they all do).
919 | or [ mayHaveCafRefs (hasCafRefs this_pkg subst1 (idArity bndr) rhs)
920 | (bndr,rhs) <- prs ] = MayHaveCafRefs
921 | otherwise = NoCafRefs
923 -----------------------------------------------------------
924 tidyTopPair :: Bool -- show unfolding
925 -> TidyEnv -- The TidyEnv is used to tidy the IdInfo
926 -- It is knot-tied: don't look at it!
929 -> (Id, CoreExpr) -- Binder and RHS before tidying
931 -- This function is the heart of Step 2
932 -- The rec_tidy_env is the one to use for the IdInfo
933 -- It's necessary because when we are dealing with a recursive
934 -- group, a variable late in the group might be mentioned
935 -- in the IdInfo of one early in the group
937 tidyTopPair show_unfold rhs_tidy_env caf_info name' (bndr, rhs)
940 bndr' = mkGlobalId details name' ty' idinfo'
941 details = idDetails bndr -- Preserve the IdDetails
942 ty' = tidyTopType (idType bndr)
943 rhs' = tidyExpr rhs_tidy_env rhs
945 idinfo' = tidyTopIdInfo (isExternalName name')
946 idinfo unfold_info worker_info
949 unfold_info | show_unfold = mkTopUnfolding rhs'
950 | otherwise = noUnfolding
951 worker_info = tidyWorker rhs_tidy_env show_unfold (workerInfo idinfo)
953 -- Usually the Id will have an accurate arity on it, because
954 -- the simplifier has just run, but not always.
955 -- One case I found was when the last thing the simplifier
956 -- did was to let-bind a non-atomic argument and then float
957 -- it to the top level. So it seems more robust just to
959 arity = exprArity rhs
962 -- tidyTopIdInfo creates the final IdInfo for top-level
963 -- binders. There are two delicate pieces:
965 -- * Arity. After CoreTidy, this arity must not change any more.
966 -- Indeed, CorePrep must eta expand where necessary to make
967 -- the manifest arity equal to the claimed arity.
969 -- * CAF info. This must also remain valid through to code generation.
970 -- We add the info here so that it propagates to all
971 -- occurrences of the binders in RHSs, and hence to occurrences in
972 -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES.
973 -- CoreToStg makes use of this when constructing SRTs.
974 tidyTopIdInfo :: Bool -> IdInfo -> Unfolding
975 -> WorkerInfo -> ArityInfo -> CafInfo
977 tidyTopIdInfo is_external idinfo unfold_info worker_info arity caf_info
978 | not is_external -- For internal Ids (not externally visible)
979 = vanillaIdInfo -- we only need enough info for code generation
980 -- Arity and strictness info are enough;
981 -- c.f. CoreTidy.tidyLetBndr
982 `setCafInfo` caf_info
984 `setAllStrictnessInfo` newStrictnessInfo idinfo
986 | otherwise -- Externally-visible Ids get the whole lot
988 `setCafInfo` caf_info
990 `setAllStrictnessInfo` newStrictnessInfo idinfo
991 `setInlinePragInfo` inlinePragInfo idinfo
992 `setUnfoldingInfo` unfold_info
993 `setWorkerInfo` worker_info
994 -- NB: we throw away the Rules
995 -- They have already been extracted by findExternalRules
999 ------------ Worker --------------
1000 tidyWorker :: TidyEnv -> Bool -> WorkerInfo -> WorkerInfo
1001 tidyWorker _tidy_env _show_unfold NoWorker
1003 tidyWorker tidy_env show_unfold (HasWorker work_id wrap_arity)
1004 | show_unfold = HasWorker (tidyVarOcc tidy_env work_id) wrap_arity
1005 | otherwise = NoWorker
1006 -- NB: do *not* expose the worker if show_unfold is off,
1007 -- because that means this thing is a loop breaker or
1008 -- marked NOINLINE or something like that
1009 -- This is important: if you expose the worker for a loop-breaker
1010 -- then you can make the simplifier go into an infinite loop, because
1011 -- in effect the unfolding is exposed. See Trac #1709
1013 -- You might think that if show_unfold is False, then the thing should
1014 -- not be w/w'd in the first place. But a legitimate reason is this:
1015 -- the function returns bottom
1016 -- In this case, show_unfold will be false (we don't expose unfoldings
1017 -- for bottoming functions), but we might still have a worker/wrapper
1018 -- split (see Note [Worker-wrapper for bottoming functions] in WorkWrap.lhs
1021 %************************************************************************
1023 \subsection{Figuring out CafInfo for an expression}
1025 %************************************************************************
1027 hasCafRefs decides whether a top-level closure can point into the dynamic heap.
1028 We mark such things as `MayHaveCafRefs' because this information is
1029 used to decide whether a particular closure needs to be referenced
1032 There are two reasons for setting MayHaveCafRefs:
1033 a) The RHS is a CAF: a top-level updatable thunk.
1034 b) The RHS refers to something that MayHaveCafRefs
1036 Possible improvement: In an effort to keep the number of CAFs (and
1037 hence the size of the SRTs) down, we could also look at the expression and
1038 decide whether it requires a small bounded amount of heap, so we can ignore
1039 it as a CAF. In these cases however, we would need to use an additional
1040 CAF list to keep track of non-collectable CAFs.
1043 hasCafRefs :: PackageId -> VarEnv Var -> Arity -> CoreExpr -> CafInfo
1044 hasCafRefs this_pkg p arity expr
1045 | is_caf || mentions_cafs
1047 | otherwise = NoCafRefs
1049 mentions_cafs = isFastTrue (cafRefs p expr)
1050 is_caf = not (arity > 0 || rhsIsStatic this_pkg expr)
1052 -- NB. we pass in the arity of the expression, which is expected
1053 -- to be calculated by exprArity. This is because exprArity
1054 -- knows how much eta expansion is going to be done by
1055 -- CorePrep later on, and we don't want to duplicate that
1056 -- knowledge in rhsIsStatic below.
1058 cafRefs :: VarEnv Id -> Expr a -> FastBool
1060 -- imported Ids first:
1061 | not (isLocalId id) = fastBool (mayHaveCafRefs (idCafInfo id))
1062 -- now Ids local to this module:
1064 case lookupVarEnv p id of
1065 Just id' -> fastBool (mayHaveCafRefs (idCafInfo id'))
1066 Nothing -> fastBool False
1068 cafRefs _ (Lit _) = fastBool False
1069 cafRefs p (App f a) = fastOr (cafRefs p f) (cafRefs p) a
1070 cafRefs p (Lam _ e) = cafRefs p e
1071 cafRefs p (Let b e) = fastOr (cafRefss p (rhssOfBind b)) (cafRefs p) e
1072 cafRefs p (Case e _bndr _ alts) = fastOr (cafRefs p e) (cafRefss p) (rhssOfAlts alts)
1073 cafRefs p (Note _n e) = cafRefs p e
1074 cafRefs p (Cast e _co) = cafRefs p e
1075 cafRefs _ (Type _) = fastBool False
1077 cafRefss :: VarEnv Id -> [Expr a] -> FastBool
1078 cafRefss _ [] = fastBool False
1079 cafRefss p (e:es) = fastOr (cafRefs p e) (cafRefss p) es
1081 fastOr :: FastBool -> (a -> FastBool) -> a -> FastBool
1082 -- hack for lazy-or over FastBool.
1083 fastOr a f x = fastBool (isFastTrue a || isFastTrue (f x))