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 )
32 import Name hiding (varName)
45 import FastBool hiding ( fastOr )
48 import Data.List ( sortBy )
49 import Data.IORef ( IORef, readIORef, writeIORef )
53 Constructing the TypeEnv, Instances, Rules, VectInfo from which the
54 ModIface is constructed, and which goes on to subsequent modules in
57 Most of the interface file is obtained simply by serialising the
58 TypeEnv. One important consequence is that if the *interface file*
59 has pragma info if and only if the final TypeEnv does. This is not so
60 important for *this* module, but it's essential for ghc --make:
61 subsequent compilations must not see (e.g.) the arity if the interface
62 file does not contain arity If they do, they'll exploit the arity;
63 then the arity might change, but the iface file doesn't change =>
64 recompilation does not happen => disaster.
66 For data types, the final TypeEnv will have a TyThing for the TyCon,
67 plus one for each DataCon; the interface file will contain just one
68 data type declaration, but it is de-serialised back into a collection
71 %************************************************************************
75 %************************************************************************
78 Plan A: mkBootModDetails: omit pragmas, make interfaces small
79 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
82 * Drop all WiredIn things from the TypeEnv
83 (we never want them in interface files)
85 * Retain all TyCons and Classes in the TypeEnv, to avoid
86 having to find which ones are mentioned in the
89 * Trim off the constructors of non-exported TyCons, both
90 from the TyCon and from the TypeEnv
92 * Drop non-exported Ids from the TypeEnv
94 * Tidy the types of the DFunIds of Instances,
95 make them into GlobalIds, (they already have External Names)
96 and add them to the TypeEnv
98 * Tidy the types of the (exported) Ids in the TypeEnv,
99 make them into GlobalIds (they already have External Names)
101 * Drop rules altogether
103 * Tidy the bindings, to ensure that the Caf and Arity
104 information is correct for each top-level binder; the
105 code generator needs it. And to ensure that local names have
106 distinct OccNames in case of object-file splitting
109 -- This is Plan A: make a small type env when typechecking only,
110 -- or when compiling a hs-boot file, or simply when not using -O
112 -- We don't look at the bindings at all -- there aren't any
115 mkBootModDetailsTc :: HscEnv -> TcGblEnv -> IO ModDetails
116 mkBootModDetailsTc hsc_env
117 TcGblEnv{ tcg_exports = exports,
118 tcg_type_env = type_env,
120 tcg_fam_insts = fam_insts
122 = mkBootModDetails hsc_env exports type_env insts fam_insts
124 mkBootModDetailsDs :: HscEnv -> ModGuts -> IO ModDetails
125 mkBootModDetailsDs hsc_env
126 ModGuts{ mg_exports = exports,
129 mg_fam_insts = fam_insts
131 = mkBootModDetails hsc_env exports type_env insts fam_insts
133 mkBootModDetails :: HscEnv -> [AvailInfo] -> NameEnv TyThing
134 -> [Instance] -> [FamInstEnv.FamInst] -> IO ModDetails
135 mkBootModDetails hsc_env exports type_env insts fam_insts
136 = do { let dflags = hsc_dflags hsc_env
137 ; showPass dflags "Tidy [hoot] type env"
139 ; let { insts' = tidyInstances globaliseAndTidyId insts
140 ; dfun_ids = map instanceDFunId insts'
141 ; type_env1 = tidyBootTypeEnv (availsToNameSet exports) type_env
142 ; type_env' = extendTypeEnvWithIds type_env1 dfun_ids
144 ; return (ModDetails { md_types = type_env'
146 , md_fam_insts = fam_insts
149 , md_exports = exports
150 , md_vect_info = noVectInfo
155 tidyBootTypeEnv :: NameSet -> TypeEnv -> TypeEnv
156 tidyBootTypeEnv exports type_env
157 = tidyTypeEnv True False exports type_env final_ids
159 -- Find the LocalIds in the type env that are exported
160 -- Make them into GlobalIds, and tidy their types
162 -- It's very important to remove the non-exported ones
163 -- because we don't tidy the OccNames, and if we don't remove
164 -- the non-exported ones we'll get many things with the
165 -- same name in the interface file, giving chaos.
166 final_ids = [ globaliseAndTidyId id
167 | id <- typeEnvIds type_env
171 -- default methods have their export flag set, but everything
172 -- else doesn't (yet), because this is pre-desugaring, so we
174 keep_it id = isExportedId id || idName id `elemNameSet` exports
178 globaliseAndTidyId :: Id -> Id
179 -- Takes an LocalId with an External Name,
180 -- makes it into a GlobalId
181 -- * unchanged Name (might be Internal or External)
182 -- * unchanged details
183 -- * VanillaIdInfo (makes a conservative assumption about Caf-hood)
184 globaliseAndTidyId id
185 = Id.setIdType (globaliseId id) tidy_type
187 tidy_type = tidyTopType (idType id)
191 %************************************************************************
193 Plan B: tidy bindings, make TypeEnv full of IdInfo
195 %************************************************************************
197 Plan B: include pragmas, make interfaces
198 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
199 * Figure out which Ids are externally visible
201 * Tidy the bindings, externalising appropriate Ids
203 * Drop all Ids from the TypeEnv, and add all the External Ids from
204 the bindings. (This adds their IdInfo to the TypeEnv; and adds
205 floated-out Ids that weren't even in the TypeEnv before.)
207 Step 1: Figure out external Ids
208 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
209 Note [choosing external names]
211 See also the section "Interface stability" in the
212 RecompilationAvoidance commentary:
213 http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/RecompilationAvoidance
215 First we figure out which Ids are "external" Ids. An
216 "external" Id is one that is visible from outside the compilation
218 a) the user exported ones
219 b) ones mentioned in the unfoldings, workers,
220 or rules of externally-visible ones
222 While figuring out which Ids are external, we pick a "tidy" OccName
223 for each one. That is, we make its OccName distinct from the other
224 external OccNames in this module, so that in interface files and
225 object code we can refer to it unambiguously by its OccName. The
226 OccName for each binder is prefixed by the name of the exported Id
227 that references it; e.g. if "f" references "x" in its unfolding, then
228 "x" is renamed to "f_x". This helps distinguish the different "x"s
229 from each other, and means that if "f" is later removed, things that
230 depend on the other "x"s will not need to be recompiled. Of course,
231 if there are multiple "f_x"s, then we have to disambiguate somehow; we
232 use "f_x0", "f_x1" etc.
234 As far as possible we should assign names in a deterministic fashion.
235 Each time this module is compiled with the same options, we should end
236 up with the same set of external names with the same types. That is,
237 the ABI hash in the interface should not change. This turns out to be
238 quite tricky, since the order of the bindings going into the tidy
239 phase is already non-deterministic, as it is based on the ordering of
240 Uniques, which are assigned unpredictably.
242 To name things in a stable way, we do a depth-first-search of the
243 bindings, starting from the exports sorted by name. This way, as long
244 as the bindings themselves are deterministic (they sometimes aren't!),
245 the order in which they are presented to the tidying phase does not
246 affect the names we assign.
248 Step 2: Tidy the program
249 ~~~~~~~~~~~~~~~~~~~~~~~~
250 Next we traverse the bindings top to bottom. For each *top-level*
253 1. Make it into a GlobalId; its IdDetails becomes VanillaGlobal,
254 reflecting the fact that from now on we regard it as a global,
257 2. Give it a system-wide Unique.
258 [Even non-exported things need system-wide Uniques because the
259 byte-code generator builds a single Name->BCO symbol table.]
261 We use the NameCache kept in the HscEnv as the
262 source of such system-wide uniques.
264 For external Ids, use the original-name cache in the NameCache
265 to ensure that the unique assigned is the same as the Id had
266 in any previous compilation run.
268 3. Rename top-level Ids according to the names we chose in step 1.
269 If it's an external Id, make it have a External Name, otherwise
270 make it have an Internal Name. This is used by the code generator
271 to decide whether to make the label externally visible
273 4. Give it its UTTERLY FINAL IdInfo; in ptic,
274 * its unfolding, if it should have one
276 * its arity, computed from the number of visible lambdas
278 * its CAF info, computed from what is free in its RHS
281 Finally, substitute these new top-level binders consistently
282 throughout, including in unfoldings. We also tidy binders in
283 RHSs, so that they print nicely in interfaces.
286 tidyProgram :: HscEnv -> ModGuts -> IO (CgGuts, ModDetails)
287 tidyProgram hsc_env (ModGuts { mg_module = mod, mg_exports = exports,
289 mg_insts = insts, mg_fam_insts = fam_insts,
291 mg_rules = imp_rules,
292 mg_vect_info = vect_info,
293 mg_dir_imps = dir_imps,
296 mg_foreign = foreign_stubs,
297 mg_hpc_info = hpc_info,
298 mg_modBreaks = modBreaks })
300 = do { let dflags = hsc_dflags hsc_env
301 ; showPass dflags "Tidy Core"
303 ; let { omit_prags = dopt Opt_OmitInterfacePragmas dflags
304 ; th = dopt Opt_TemplateHaskell dflags
307 ; let { implicit_binds = getImplicitBinds type_env }
309 ; (unfold_env, tidy_occ_env)
310 <- chooseExternalIds hsc_env mod omit_prags binds implicit_binds
314 | otherwise = findExternalRules binds imp_rules unfold_env
315 -- findExternalRules filters imp_rules to avoid binders that
316 -- aren't externally visible; but the externally-visible binders
317 -- are computed (by findExternalIds) assuming that all orphan
318 -- rules are exported (they get their Exported flag set in the desugarer)
319 -- So in fact we may export more than we need.
320 -- (It's a sort of mutual recursion.)
323 ; let { (tidy_env, tidy_binds)
324 = tidyTopBinds hsc_env unfold_env tidy_occ_env binds }
326 ; let { export_set = availsToNameSet exports
327 ; final_ids = [ id | id <- bindersOfBinds tidy_binds,
328 isExternalName (idName id)]
329 ; tidy_type_env = tidyTypeEnv omit_prags th export_set
331 ; tidy_insts = tidyInstances (lookup_dfun tidy_type_env) insts
332 -- A DFunId will have a binding in tidy_binds, and so
333 -- will now be in final_env, replete with IdInfo
334 -- Its name will be unchanged since it was born, but
335 -- we want Global, IdInfo-rich (or not) DFunId in the
338 ; tidy_rules = tidyRules tidy_env ext_rules
339 -- You might worry that the tidy_env contains IdInfo-rich stuff
340 -- and indeed it does, but if omit_prags is on, ext_rules is
343 ; tidy_vect_info = tidyVectInfo tidy_env vect_info
345 -- See Note [Injecting implicit bindings]
346 ; all_tidy_binds = implicit_binds ++ tidy_binds
348 ; alg_tycons = filter isAlgTyCon (typeEnvTyCons type_env)
351 ; endPass dflags "Tidy Core" Opt_D_dump_simpl all_tidy_binds
352 ; dumpIfSet_core dflags Opt_D_dump_simpl
354 (pprRules tidy_rules)
356 ; let dir_imp_mods = moduleEnvKeys dir_imps
358 ; return (CgGuts { cg_module = mod,
359 cg_tycons = alg_tycons,
360 cg_binds = all_tidy_binds,
361 cg_dir_imps = dir_imp_mods,
362 cg_foreign = foreign_stubs,
363 cg_dep_pkgs = dep_pkgs deps,
364 cg_hpc_info = hpc_info,
365 cg_modBreaks = modBreaks },
367 ModDetails { md_types = tidy_type_env,
368 md_rules = tidy_rules,
369 md_insts = tidy_insts,
370 md_vect_info = tidy_vect_info, md_fam_insts = fam_insts,
371 md_exports = exports,
372 md_anns = anns -- are already tidy
376 lookup_dfun :: TypeEnv -> Var -> Id
377 lookup_dfun type_env dfun_id
378 = case lookupTypeEnv type_env (idName dfun_id) of
379 Just (AnId dfun_id') -> dfun_id'
380 _other -> pprPanic "lookup_dfun" (ppr dfun_id)
382 --------------------------
383 tidyTypeEnv :: Bool -- Compiling without -O, so omit prags
384 -> Bool -- Template Haskell is on
385 -> NameSet -> TypeEnv -> [Id] -> TypeEnv
387 -- The competed type environment is gotten from
388 -- Dropping any wired-in things, and then
389 -- a) keeping the types and classes
390 -- b) removing all Ids,
391 -- c) adding Ids with correct IdInfo, including unfoldings,
392 -- gotten from the bindings
393 -- From (c) we keep only those Ids with External names;
394 -- the CoreTidy pass makes sure these are all and only
395 -- the externally-accessible ones
396 -- This truncates the type environment to include only the
397 -- exported Ids and things needed from them, which saves space
399 tidyTypeEnv omit_prags th exports type_env final_ids
400 = let type_env1 = filterNameEnv keep_it type_env
401 type_env2 = extendTypeEnvWithIds type_env1 final_ids
402 type_env3 | omit_prags = mapNameEnv (trimThing th exports) type_env2
403 | otherwise = type_env2
407 -- We keep GlobalIds, because they won't appear
408 -- in the bindings from which final_ids are derived!
409 -- (The bindings bind LocalIds.)
410 keep_it thing | isWiredInThing thing = False
411 keep_it (AnId id) = isGlobalId id -- Keep GlobalIds (e.g. class ops)
412 keep_it _other = True -- Keep all TyCons, DataCons, and Classes
414 --------------------------
415 isWiredInThing :: TyThing -> Bool
416 isWiredInThing thing = isWiredInName (getName thing)
418 --------------------------
419 trimThing :: Bool -> NameSet -> TyThing -> TyThing
420 -- Trim off inessentials, for boot files and no -O
421 trimThing th exports (ATyCon tc)
422 | not th && not (mustExposeTyCon exports tc)
423 = ATyCon (makeTyConAbstract tc) -- Note [Trimming and Template Haskell]
425 trimThing _th _exports (AnId id)
426 | not (isImplicitId id)
427 = AnId (id `setIdInfo` vanillaIdInfo)
429 trimThing _th _exports other_thing
433 {- Note [Trimming and Template Haskell]
434 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
435 Consider (Trac #2386) this
436 module M(T, makeOne) where
438 makeOne = [| Yay "Yep" |]
439 Notice that T is exported abstractly, but makeOne effectively exports it too!
440 A module that splices in $(makeOne) will then look for a declartion of Yay,
441 so it'd better be there. Hence, brutally but simply, we switch off type
442 constructor trimming if TH is enabled in this module. -}
445 mustExposeTyCon :: NameSet -- Exports
446 -> TyCon -- The tycon
447 -> Bool -- Can its rep be hidden?
448 -- We are compiling without -O, and thus trying to write as little as
449 -- possible into the interface file. But we must expose the details of
450 -- any data types whose constructors or fields are exported
451 mustExposeTyCon exports tc
452 | not (isAlgTyCon tc) -- Synonyms
454 | isEnumerationTyCon tc -- For an enumeration, exposing the constructors
455 = True -- won't lead to the need for further exposure
456 -- (This includes data types with no constructors.)
457 | isOpenTyCon tc -- Open type family
460 | otherwise -- Newtype, datatype
461 = any exported_con (tyConDataCons tc)
462 -- Expose rep if any datacon or field is exported
464 || (isNewTyCon tc && isFFITy (snd (newTyConRhs tc)))
465 -- Expose the rep for newtypes if the rep is an FFI type.
466 -- For a very annoying reason. 'Foreign import' is meant to
467 -- be able to look through newtypes transparently, but it
468 -- can only do that if it can "see" the newtype representation
470 exported_con con = any (`elemNameSet` exports)
471 (dataConName con : dataConFieldLabels con)
473 tidyInstances :: (DFunId -> DFunId) -> [Instance] -> [Instance]
474 tidyInstances tidy_dfun ispecs
477 tidy ispec = setInstanceDFunId ispec $
478 tidy_dfun (instanceDFunId ispec)
482 tidyVectInfo :: TidyEnv -> VectInfo -> VectInfo
483 tidyVectInfo (_, var_env) info@(VectInfo { vectInfoVar = vars
484 , vectInfoPADFun = pas
485 , vectInfoIso = isos })
486 = info { vectInfoVar = tidy_vars
487 , vectInfoPADFun = tidy_pas
488 , vectInfoIso = tidy_isos }
491 $ map tidy_var_mapping
494 tidy_pas = mapNameEnv tidy_snd_var pas
495 tidy_isos = mapNameEnv tidy_snd_var isos
497 tidy_var_mapping (from, to) = (from', (from', lookup_var to))
498 where from' = lookup_var from
499 tidy_snd_var (x, var) = (x, lookup_var var)
501 lookup_var var = lookupWithDefaultVarEnv var_env var var
505 %************************************************************************
509 %************************************************************************
511 Note [Injecting implicit bindings]
512 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
513 We inject the implict bindings right at the end, in CoreTidy.
514 Some of these bindings, notably record selectors, are not
515 constructed in an optimised form. E.g. record selector for
516 data T = MkT { x :: {-# UNPACK #-} !Int }
517 Then the unfolding looks like
518 x = \t. case t of MkT x1 -> let x = I# x1 in x
519 This generates bad code unless it's first simplified a bit. That is
520 why CoreUnfold.mkImplicitUnfolding uses simleExprOpt to do a bit of
521 optimisation first. (Only matters when the selector is used curried;
522 eg map x ys.) See Trac #2070.
524 [Oct 09: in fact, record selectors are no longer implicit Ids at all,
525 because we really do want to optimise them properly. They are treated
526 much like any other Id. But doing "light" optimisation on an implicit
527 Id still makes sense.]
529 At one time I tried injecting the implicit bindings *early*, at the
530 beginning of SimplCore. But that gave rise to real difficulty,
531 becuase GlobalIds are supposed to have *fixed* IdInfo, but the
532 simplifier and other core-to-core passes mess with IdInfo all the
533 time. The straw that broke the camels back was when a class selector
534 got the wrong arity -- ie the simplifier gave it arity 2, whereas
535 importing modules were expecting it to have arity 1 (Trac #2844).
536 It's much safer just to inject them right at the end, after tidying.
538 Oh: two other reasons for injecting them late:
540 - If implicit Ids are already in the bindings when we start TidyPgm,
541 we'd have to be careful not to treat them as external Ids (in
542 the sense of findExternalIds); else the Ids mentioned in *their*
543 RHSs will be treated as external and you get an interface file
545 but nothing refererring to a18 (because the implicit Id is the
546 one that does, and implicit Ids don't appear in interface files).
548 - More seriously, the tidied type-envt will include the implicit
549 Id replete with a18 in its unfolding; but we won't take account
550 of a18 when computing a fingerprint for the class; result chaos.
552 There is one sort of implicit binding that is injected still later,
553 namely those for data constructor workers. Reason (I think): it's
554 really just a code generation trick.... binding itself makes no sense.
555 See CorePrep Note [Data constructor workers].
558 getImplicitBinds :: TypeEnv -> [CoreBind]
559 getImplicitBinds type_env
560 = map get_defn (concatMap implicit_ids (typeEnvElts type_env))
562 implicit_ids (ATyCon tc) = mapCatMaybes dataConWrapId_maybe (tyConDataCons tc)
563 implicit_ids (AClass cls) = classSelIds cls
566 get_defn :: Id -> CoreBind
567 get_defn id = NonRec id (unfoldingTemplate (idUnfolding id))
571 %************************************************************************
573 \subsection{Step 1: finding externals}
575 %************************************************************************
577 Sete Note [choosing external names].
580 type UnfoldEnv = IdEnv (Name{-new name-}, Bool {-show unfolding-})
581 -- maps each top-level Id to its new Name (the Id is tidied in step 2)
582 -- The Unique is unchanged. If the new Id is external, it will be
583 -- visible in the interface file.
585 -- Bool => expose unfolding or not.
587 chooseExternalIds :: HscEnv
592 -> IO (UnfoldEnv, TidyOccEnv)
593 -- Step 1 from the notes above
595 chooseExternalIds hsc_env mod omit_prags binds implicit_binds
597 (unfold_env1,occ_env1)
598 <- search (zip sorted_exports sorted_exports) emptyVarEnv init_occ_env
599 let internal_ids = filter (not . (`elemVarEnv` unfold_env1)) binders
600 tidy_internal internal_ids unfold_env1 occ_env1
602 nc_var = hsc_NC hsc_env
604 -- the exports, sorted by OccName. This is a deterministic list of
605 -- Ids (i.e. it's the same list every time this module is compiled),
606 -- in contrast to the bindings, which are ordered
607 -- non-deterministically.
609 -- This list will serve as a starting point for finding a
610 -- deterministic, tidy, renaming for all external Ids in this
612 sorted_exports = sortBy (compare `on` getOccName) $
613 filter isExportedId binders
615 binders = bindersOfBinds binds
616 implicit_binders = bindersOfBinds implicit_binds
618 bind_env :: IdEnv (Id,CoreExpr)
619 bind_env = mkVarEnv (zip (map fst bs) bs) where bs = flattenBinds binds
621 avoids = [getOccName name | bndr <- binders ++ implicit_binders,
622 let name = idName bndr,
623 isExternalName name ]
624 -- In computing our "avoids" list, we must include
626 -- all things with global names (assigned once and for
627 -- all by the renamer)
628 -- since their names are "taken".
629 -- The type environment is a convenient source of such things.
630 -- In particular, the set of binders doesn't include
631 -- implicit Ids at this stage.
633 -- We also make sure to avoid any exported binders. Consider
634 -- f{-u1-} = 1 -- Local decl
636 -- f{-u2-} = 2 -- Exported decl
638 -- The second exported decl must 'get' the name 'f', so we
639 -- have to put 'f' in the avoids list before we get to the first
640 -- decl. tidyTopId then does a no-op on exported binders.
641 init_occ_env = initTidyOccEnv avoids
644 search :: [(Id,Id)] -- (external id, referrring id)
645 -> UnfoldEnv -- id -> (new Name, show_unfold)
646 -> TidyOccEnv -- occ env for choosing new Names
647 -> IO (UnfoldEnv, TidyOccEnv)
649 search [] unfold_env occ_env = return (unfold_env, occ_env)
651 search ((idocc,referrer) : rest) unfold_env occ_env
652 | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env
654 (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc
656 (id, rhs) = expectJust (showSDoc (text "chooseExternalIds: " <>
658 lookupVarEnv bind_env idocc
659 -- NB. idocc might be an *occurrence* of an Id, whereas we want
660 -- the Id from the binding site, because only the latter is
661 -- guaranteed to have the unfolding attached. This is why we
662 -- keep binding site Ids in the bind_env.
663 (new_ids, show_unfold)
664 | omit_prags = ([], False)
665 | otherwise = addExternal id rhs
666 unfold_env' = extendVarEnv unfold_env id (name',show_unfold)
667 referrer' | isExportedId id = id
668 | otherwise = referrer
670 search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env'
672 tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv
673 -> IO (UnfoldEnv, TidyOccEnv)
674 tidy_internal [] unfold_env occ_env = return (unfold_env,occ_env)
675 tidy_internal (id:ids) unfold_env occ_env = do
676 (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id
677 let unfold_env' = extendVarEnv unfold_env id (name',False)
678 tidy_internal ids unfold_env' occ_env'
680 addExternal :: Id -> CoreExpr -> ([Id],Bool)
681 addExternal id rhs = (new_needed_ids, show_unfold)
683 new_needed_ids = unfold_ids ++
684 filter (\id -> isLocalId id &&
685 not (id `elemVarSet` unfold_set))
686 (varSetElems worker_ids ++
687 varSetElems spec_ids) -- XXX non-det ordering
690 dont_inline = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo))
691 loop_breaker = isNonRuleLoopBreaker (occInfo idinfo)
692 bottoming_fn = isBottomingSig (newStrictnessInfo idinfo `orElse` topSig)
693 spec_ids = specInfoFreeVars (specInfo idinfo)
694 worker_info = workerInfo idinfo
696 -- Stuff to do with the Id's unfolding
697 -- The simplifier has put an up-to-date unfolding
698 -- in the IdInfo, but the RHS will do just as well
699 unfolding = unfoldingInfo idinfo
700 rhs_is_small = not (neverUnfold unfolding)
702 -- We leave the unfolding there even if there is a worker
703 -- In GHCI the unfolding is used by importers
704 -- When writing an interface file, we omit the unfolding
705 -- if there is a worker
706 show_unfold = not bottoming_fn && -- Not necessary
709 rhs_is_small -- Small enough
711 (unfold_set, unfold_ids)
712 | show_unfold = freeVarsInDepthFirstOrder rhs
713 | otherwise = (emptyVarSet, [])
715 worker_ids = case worker_info of
716 HasWorker work_id _ -> unitVarSet work_id
717 _otherwise -> emptyVarSet
720 -- We want a deterministic free-variable list. exprFreeVars gives us
721 -- a VarSet, which is in a non-deterministic order when converted to a
722 -- list. Hence, here we define a free-variable finder that returns
723 -- the free variables in the order that they are encountered.
725 -- Note [choosing external names]
727 freeVarsInDepthFirstOrder :: CoreExpr -> (VarSet, [Id])
728 freeVarsInDepthFirstOrder e =
730 DFFV m -> case m emptyVarSet [] of
731 (set,ids,_) -> (set,ids)
733 newtype DFFV a = DFFV (VarSet -> [Var] -> (VarSet,[Var],a))
735 instance Monad DFFV where
736 return a = DFFV $ \set ids -> (set, ids, a)
737 (DFFV m) >>= k = DFFV $ \set ids ->
739 (set',ids',a) -> case k a of
740 DFFV f -> f set' ids'
742 insert :: Var -> DFFV ()
743 insert v = DFFV $ \ set ids -> case () of
744 _ | v `elemVarSet` set -> (set,ids,())
745 | otherwise -> (extendVarSet set v, v:ids, ())
747 dffvExpr :: CoreExpr -> DFFV ()
748 dffvExpr e = go emptyVarSet e
750 go scope e = case e of
751 Var v | isLocalId v && not (v `elemVarSet` scope) -> insert v
752 App e1 e2 -> do go scope e1; go scope e2
753 Lam v e -> go (extendVarSet scope v) e
754 Note _ e -> go scope e
755 Cast e _ -> go scope e
756 Let (NonRec x r) e -> do go scope r; go (extendVarSet scope x) e
757 Let (Rec prs) e -> do let scope' = extendVarSetList scope (map fst prs)
758 mapM_ (go scope') (map snd prs)
760 Case e b _ as -> do go scope e
761 mapM_ (go_alt (extendVarSet scope b)) as
764 go_alt scope (_,xs,r) = go (extendVarSetList scope xs) r
768 --------------------------------------------------------------------
770 -- This is where we set names to local/global based on whether they really are
771 -- externally visible (see comment at the top of this module). If the name
772 -- was previously local, we have to give it a unique occurrence name if
773 -- we intend to externalise it.
776 tidyTopName :: Module -> IORef NameCache -> Maybe Id -> TidyOccEnv
777 -> Id -> IO (TidyOccEnv, Name)
778 tidyTopName mod nc_var maybe_ref occ_env id
779 | global && internal = return (occ_env, localiseName name)
781 | global && external = return (occ_env, name)
782 -- Global names are assumed to have been allocated by the renamer,
783 -- so they already have the "right" unique
784 -- And it's a system-wide unique too
786 -- Now we get to the real reason that all this is in the IO Monad:
787 -- we have to update the name cache in a nice atomic fashion
789 | local && internal = do { nc <- readIORef nc_var
790 ; let (nc', new_local_name) = mk_new_local nc
791 ; writeIORef nc_var nc'
792 ; return (occ_env', new_local_name) }
793 -- Even local, internal names must get a unique occurrence, because
794 -- if we do -split-objs we externalise the name later, in the code generator
796 -- Similarly, we must make sure it has a system-wide Unique, because
797 -- the byte-code generator builds a system-wide Name->BCO symbol table
799 | local && external = do { nc <- readIORef nc_var
800 ; let (nc', new_external_name) = mk_new_external nc
801 ; writeIORef nc_var nc'
802 ; return (occ_env', new_external_name) }
804 | otherwise = panic "tidyTopName"
807 external = isJust maybe_ref
808 global = isExternalName name
810 internal = not external
811 loc = nameSrcSpan name
813 old_occ = nameOccName name
815 | Just ref <- maybe_ref, ref /= id =
816 mkOccName (occNameSpace old_occ) $
818 ref_str = occNameString (getOccName ref)
819 occ_str = occNameString old_occ
823 -- workers: the worker for a function already
824 -- includes the occname for its parent, so there's
825 -- no need to prepend the referrer.
826 _other | isSystemName name -> ref_str
827 | otherwise -> ref_str ++ '_' : occ_str
828 -- If this name was system-generated, then don't bother
829 -- to retain its OccName, just use the referrer. These
830 -- system-generated names will become "f1", "f2", etc. for
832 | otherwise = old_occ
834 (occ_env', occ') = tidyOccName occ_env new_occ
836 mk_new_local nc = (nc { nsUniqs = us2 }, mkInternalName uniq occ' loc)
838 (us1, us2) = splitUniqSupply (nsUniqs nc)
839 uniq = uniqFromSupply us1
841 mk_new_external nc = allocateGlobalBinder nc mod occ' loc
842 -- If we want to externalise a currently-local name, check
843 -- whether we have already assigned a unique for it.
844 -- If so, use it; if not, extend the table.
845 -- All this is done by allcoateGlobalBinder.
846 -- This is needed when *re*-compiling a module in GHCi; we must
847 -- use the same name for externally-visible things as we did before.
851 findExternalRules :: [CoreBind]
852 -> [CoreRule] -- Non-local rules (i.e. ones for imported fns)
853 -> UnfoldEnv -- Ids that are exported, so we need their rules
855 -- The complete rules are gotten by combining
856 -- a) the non-local rules
857 -- b) rules embedded in the top-level Ids
858 findExternalRules binds non_local_rules unfold_env
859 = filter (not . internal_rule) (non_local_rules ++ local_rules)
862 | id <- bindersOfBinds binds,
864 rule <- idCoreRules id
868 = any (not . external_id) (varSetElems (ruleLhsFreeIds rule))
869 -- Don't export a rule whose LHS mentions a locally-defined
870 -- Id that is completely internal (i.e. not visible to an
874 | Just (name,_) <- lookupVarEnv unfold_env id = isExternalName name
880 %************************************************************************
882 \subsection{Step 2: top-level tidying}
884 %************************************************************************
888 -- TopTidyEnv: when tidying we need to know
889 -- * nc_var: The NameCache, containing a unique supply and any pre-ordained Names.
890 -- These may have arisen because the
891 -- renamer read in an interface file mentioning M.$wf, say,
892 -- and assigned it unique r77. If, on this compilation, we've
893 -- invented an Id whose name is $wf (but with a different unique)
894 -- we want to rename it to have unique r77, so that we can do easy
895 -- comparisons with stuff from the interface file
897 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
900 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
902 tidyTopBinds :: HscEnv
906 -> (TidyEnv, [CoreBind])
908 tidyTopBinds hsc_env unfold_env init_occ_env binds
909 = tidy init_env binds
911 init_env = (init_occ_env, emptyVarEnv)
913 this_pkg = thisPackage (hsc_dflags hsc_env)
915 tidy env [] = (env, [])
916 tidy env (b:bs) = let (env1, b') = tidyTopBind this_pkg unfold_env env b
917 (env2, bs') = tidy env1 bs
921 ------------------------
922 tidyTopBind :: PackageId
926 -> (TidyEnv, CoreBind)
928 tidyTopBind this_pkg unfold_env (occ_env,subst1) (NonRec bndr rhs)
929 = (tidy_env2, NonRec bndr' rhs')
931 Just (name',show_unfold) = lookupVarEnv unfold_env bndr
932 caf_info = hasCafRefs this_pkg subst1 (idArity bndr) rhs
933 (bndr', rhs') = tidyTopPair show_unfold tidy_env2 caf_info name' (bndr, rhs)
934 subst2 = extendVarEnv subst1 bndr bndr'
935 tidy_env2 = (occ_env, subst2)
937 tidyTopBind this_pkg unfold_env (occ_env,subst1) (Rec prs)
938 = (tidy_env2, Rec prs')
940 prs' = [ tidyTopPair show_unfold tidy_env2 caf_info name' (id,rhs)
942 let (name',show_unfold) =
943 expectJust "tidyTopBind" $ lookupVarEnv unfold_env id
946 subst2 = extendVarEnvList subst1 (bndrs `zip` map fst prs')
947 tidy_env2 = (occ_env, subst2)
951 -- the CafInfo for a recursive group says whether *any* rhs in
952 -- the group may refer indirectly to a CAF (because then, they all do).
954 | or [ mayHaveCafRefs (hasCafRefs this_pkg subst1 (idArity bndr) rhs)
955 | (bndr,rhs) <- prs ] = MayHaveCafRefs
956 | otherwise = NoCafRefs
958 -----------------------------------------------------------
959 tidyTopPair :: Bool -- show unfolding
960 -> TidyEnv -- The TidyEnv is used to tidy the IdInfo
961 -- It is knot-tied: don't look at it!
964 -> (Id, CoreExpr) -- Binder and RHS before tidying
966 -- This function is the heart of Step 2
967 -- The rec_tidy_env is the one to use for the IdInfo
968 -- It's necessary because when we are dealing with a recursive
969 -- group, a variable late in the group might be mentioned
970 -- in the IdInfo of one early in the group
972 tidyTopPair show_unfold rhs_tidy_env caf_info name' (bndr, rhs)
975 bndr' = mkGlobalId details name' ty' idinfo'
976 details = idDetails bndr -- Preserve the IdDetails
977 ty' = tidyTopType (idType bndr)
978 rhs' = tidyExpr rhs_tidy_env rhs
980 idinfo' = tidyTopIdInfo (isExternalName name')
981 idinfo unfold_info worker_info
984 unfold_info | show_unfold = mkTopUnfolding rhs'
985 | otherwise = noUnfolding
986 worker_info = tidyWorker rhs_tidy_env show_unfold (workerInfo idinfo)
988 -- Usually the Id will have an accurate arity on it, because
989 -- the simplifier has just run, but not always.
990 -- One case I found was when the last thing the simplifier
991 -- did was to let-bind a non-atomic argument and then float
992 -- it to the top level. So it seems more robust just to
994 arity = exprArity rhs
997 -- tidyTopIdInfo creates the final IdInfo for top-level
998 -- binders. There are two delicate pieces:
1000 -- * Arity. After CoreTidy, this arity must not change any more.
1001 -- Indeed, CorePrep must eta expand where necessary to make
1002 -- the manifest arity equal to the claimed arity.
1004 -- * CAF info. This must also remain valid through to code generation.
1005 -- We add the info here so that it propagates to all
1006 -- occurrences of the binders in RHSs, and hence to occurrences in
1007 -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES.
1008 -- CoreToStg makes use of this when constructing SRTs.
1009 tidyTopIdInfo :: Bool -> IdInfo -> Unfolding
1010 -> WorkerInfo -> ArityInfo -> CafInfo
1012 tidyTopIdInfo is_external idinfo unfold_info worker_info arity caf_info
1013 | not is_external -- For internal Ids (not externally visible)
1014 = vanillaIdInfo -- we only need enough info for code generation
1015 -- Arity and strictness info are enough;
1016 -- c.f. CoreTidy.tidyLetBndr
1017 `setCafInfo` caf_info
1018 `setArityInfo` arity
1019 `setAllStrictnessInfo` newStrictnessInfo idinfo
1021 | otherwise -- Externally-visible Ids get the whole lot
1023 `setCafInfo` caf_info
1024 `setArityInfo` arity
1025 `setAllStrictnessInfo` newStrictnessInfo idinfo
1026 `setInlinePragInfo` inlinePragInfo idinfo
1027 `setUnfoldingInfo` unfold_info
1028 `setWorkerInfo` worker_info
1029 -- NB: we throw away the Rules
1030 -- They have already been extracted by findExternalRules
1034 ------------ Worker --------------
1035 tidyWorker :: TidyEnv -> Bool -> WorkerInfo -> WorkerInfo
1036 tidyWorker _tidy_env _show_unfold NoWorker
1038 tidyWorker tidy_env show_unfold (HasWorker work_id wrap_arity)
1039 | show_unfold = HasWorker (tidyVarOcc tidy_env work_id) wrap_arity
1040 | otherwise = NoWorker
1041 -- NB: do *not* expose the worker if show_unfold is off,
1042 -- because that means this thing is a loop breaker or
1043 -- marked NOINLINE or something like that
1044 -- This is important: if you expose the worker for a loop-breaker
1045 -- then you can make the simplifier go into an infinite loop, because
1046 -- in effect the unfolding is exposed. See Trac #1709
1048 -- You might think that if show_unfold is False, then the thing should
1049 -- not be w/w'd in the first place. But a legitimate reason is this:
1050 -- the function returns bottom
1051 -- In this case, show_unfold will be false (we don't expose unfoldings
1052 -- for bottoming functions), but we might still have a worker/wrapper
1053 -- split (see Note [Worker-wrapper for bottoming functions] in WorkWrap.lhs
1056 %************************************************************************
1058 \subsection{Figuring out CafInfo for an expression}
1060 %************************************************************************
1062 hasCafRefs decides whether a top-level closure can point into the dynamic heap.
1063 We mark such things as `MayHaveCafRefs' because this information is
1064 used to decide whether a particular closure needs to be referenced
1067 There are two reasons for setting MayHaveCafRefs:
1068 a) The RHS is a CAF: a top-level updatable thunk.
1069 b) The RHS refers to something that MayHaveCafRefs
1071 Possible improvement: In an effort to keep the number of CAFs (and
1072 hence the size of the SRTs) down, we could also look at the expression and
1073 decide whether it requires a small bounded amount of heap, so we can ignore
1074 it as a CAF. In these cases however, we would need to use an additional
1075 CAF list to keep track of non-collectable CAFs.
1078 hasCafRefs :: PackageId -> VarEnv Var -> Arity -> CoreExpr -> CafInfo
1079 hasCafRefs this_pkg p arity expr
1080 | is_caf || mentions_cafs
1082 | otherwise = NoCafRefs
1084 mentions_cafs = isFastTrue (cafRefs p expr)
1085 is_caf = not (arity > 0 || rhsIsStatic this_pkg expr)
1087 -- NB. we pass in the arity of the expression, which is expected
1088 -- to be calculated by exprArity. This is because exprArity
1089 -- knows how much eta expansion is going to be done by
1090 -- CorePrep later on, and we don't want to duplicate that
1091 -- knowledge in rhsIsStatic below.
1093 cafRefs :: VarEnv Id -> Expr a -> FastBool
1095 -- imported Ids first:
1096 | not (isLocalId id) = fastBool (mayHaveCafRefs (idCafInfo id))
1097 -- now Ids local to this module:
1099 case lookupVarEnv p id of
1100 Just id' -> fastBool (mayHaveCafRefs (idCafInfo id'))
1101 Nothing -> fastBool False
1103 cafRefs _ (Lit _) = fastBool False
1104 cafRefs p (App f a) = fastOr (cafRefs p f) (cafRefs p) a
1105 cafRefs p (Lam _ e) = cafRefs p e
1106 cafRefs p (Let b e) = fastOr (cafRefss p (rhssOfBind b)) (cafRefs p) e
1107 cafRefs p (Case e _bndr _ alts) = fastOr (cafRefs p e) (cafRefss p) (rhssOfAlts alts)
1108 cafRefs p (Note _n e) = cafRefs p e
1109 cafRefs p (Cast e _co) = cafRefs p e
1110 cafRefs _ (Type _) = fastBool False
1112 cafRefss :: VarEnv Id -> [Expr a] -> FastBool
1113 cafRefss _ [] = fastBool False
1114 cafRefss p (e:es) = fastOr (cafRefs p e) (cafRefss p) es
1116 fastOr :: FastBool -> (a -> FastBool) -> a -> FastBool
1117 -- hack for lazy-or over FastBool.
1118 fastOr a f x = fastBool (isFastTrue a || isFastTrue (f x))