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"
21 import CoreArity ( exprArity, exprBotStrictness_maybe )
22 import Class ( classSelIds )
31 import Name hiding (varName)
44 import FastBool hiding ( fastOr )
47 import Data.List ( sortBy )
48 import Data.IORef ( IORef, readIORef, writeIORef )
52 Constructing the TypeEnv, Instances, Rules, VectInfo from which the
53 ModIface is constructed, and which goes on to subsequent modules in
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 ; omit_prags = dopt Opt_OmitInterfacePragmas dflags
301 ; expose_all = dopt Opt_ExposeAllUnfoldings dflags
302 ; th = dopt Opt_TemplateHaskell dflags
304 ; showPass dflags "Tidy Core"
306 ; let { implicit_binds = getImplicitBinds type_env }
308 ; (unfold_env, tidy_occ_env)
309 <- chooseExternalIds hsc_env mod omit_prags expose_all
310 binds implicit_binds imp_rules
312 ; let { ext_rules = findExternalRules omit_prags binds imp_rules unfold_env }
313 -- See Note [Which rules to expose]
315 ; let { (tidy_env, tidy_binds)
316 = tidyTopBinds hsc_env unfold_env tidy_occ_env binds }
318 ; let { export_set = availsToNameSet exports
319 ; final_ids = [ id | id <- bindersOfBinds tidy_binds,
320 isExternalName (idName id)]
321 ; tidy_type_env = tidyTypeEnv omit_prags th export_set
323 ; tidy_insts = tidyInstances (lookup_dfun tidy_type_env) insts
324 -- A DFunId will have a binding in tidy_binds, and so
325 -- will now be in final_env, replete with IdInfo
326 -- Its name will be unchanged since it was born, but
327 -- we want Global, IdInfo-rich (or not) DFunId in the
330 ; tidy_rules = tidyRules tidy_env ext_rules
331 -- You might worry that the tidy_env contains IdInfo-rich stuff
332 -- and indeed it does, but if omit_prags is on, ext_rules is
335 ; tidy_vect_info = tidyVectInfo tidy_env vect_info
337 -- See Note [Injecting implicit bindings]
338 ; all_tidy_binds = implicit_binds ++ tidy_binds
340 ; alg_tycons = filter isAlgTyCon (typeEnvTyCons type_env)
343 ; endPass dflags "Tidy Core" Opt_D_dump_simpl all_tidy_binds tidy_rules
344 ; let dir_imp_mods = moduleEnvKeys dir_imps
346 ; return (CgGuts { cg_module = mod,
347 cg_tycons = alg_tycons,
348 cg_binds = all_tidy_binds,
349 cg_dir_imps = dir_imp_mods,
350 cg_foreign = foreign_stubs,
351 cg_dep_pkgs = dep_pkgs deps,
352 cg_hpc_info = hpc_info,
353 cg_modBreaks = modBreaks },
355 ModDetails { md_types = tidy_type_env,
356 md_rules = tidy_rules,
357 md_insts = tidy_insts,
358 md_vect_info = tidy_vect_info,
359 md_fam_insts = fam_insts,
360 md_exports = exports,
361 md_anns = anns -- are already tidy
365 lookup_dfun :: TypeEnv -> Var -> Id
366 lookup_dfun type_env dfun_id
367 = case lookupTypeEnv type_env (idName dfun_id) of
368 Just (AnId dfun_id') -> dfun_id'
369 _other -> pprPanic "lookup_dfun" (ppr dfun_id)
371 --------------------------
372 tidyTypeEnv :: Bool -- Compiling without -O, so omit prags
373 -> Bool -- Template Haskell is on
374 -> NameSet -> TypeEnv -> [Id] -> TypeEnv
376 -- The competed type environment is gotten from
377 -- Dropping any wired-in things, and then
378 -- a) keeping the types and classes
379 -- b) removing all Ids,
380 -- c) adding Ids with correct IdInfo, including unfoldings,
381 -- gotten from the bindings
382 -- From (c) we keep only those Ids with External names;
383 -- the CoreTidy pass makes sure these are all and only
384 -- the externally-accessible ones
385 -- This truncates the type environment to include only the
386 -- exported Ids and things needed from them, which saves space
388 tidyTypeEnv omit_prags th exports type_env final_ids
389 = let type_env1 = filterNameEnv keep_it type_env
390 type_env2 = extendTypeEnvWithIds type_env1 final_ids
391 type_env3 | omit_prags = mapNameEnv (trimThing th exports) type_env2
392 | otherwise = type_env2
396 -- We keep GlobalIds, because they won't appear
397 -- in the bindings from which final_ids are derived!
398 -- (The bindings bind LocalIds.)
399 keep_it thing | isWiredInThing thing = False
400 keep_it (AnId id) = isGlobalId id -- Keep GlobalIds (e.g. class ops)
401 keep_it _other = True -- Keep all TyCons, DataCons, and Classes
403 --------------------------
404 isWiredInThing :: TyThing -> Bool
405 isWiredInThing thing = isWiredInName (getName thing)
407 --------------------------
408 trimThing :: Bool -> NameSet -> TyThing -> TyThing
409 -- Trim off inessentials, for boot files and no -O
410 trimThing th exports (ATyCon tc)
411 | not th && not (mustExposeTyCon exports tc)
412 = ATyCon (makeTyConAbstract tc) -- Note [Trimming and Template Haskell]
414 trimThing _th _exports (AnId id)
415 | not (isImplicitId id)
416 = AnId (id `setIdInfo` vanillaIdInfo)
418 trimThing _th _exports other_thing
422 {- Note [Trimming and Template Haskell]
423 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
424 Consider (Trac #2386) this
425 module M(T, makeOne) where
427 makeOne = [| Yay "Yep" |]
428 Notice that T is exported abstractly, but makeOne effectively exports it too!
429 A module that splices in $(makeOne) will then look for a declartion of Yay,
430 so it'd better be there. Hence, brutally but simply, we switch off type
431 constructor trimming if TH is enabled in this module. -}
434 mustExposeTyCon :: NameSet -- Exports
435 -> TyCon -- The tycon
436 -> Bool -- Can its rep be hidden?
437 -- We are compiling without -O, and thus trying to write as little as
438 -- possible into the interface file. But we must expose the details of
439 -- any data types whose constructors or fields are exported
440 mustExposeTyCon exports tc
441 | not (isAlgTyCon tc) -- Synonyms
443 | isEnumerationTyCon tc -- For an enumeration, exposing the constructors
444 = True -- won't lead to the need for further exposure
445 -- (This includes data types with no constructors.)
446 | isOpenTyCon tc -- Open type family
449 | otherwise -- Newtype, datatype
450 = any exported_con (tyConDataCons tc)
451 -- Expose rep if any datacon or field is exported
453 || (isNewTyCon tc && isFFITy (snd (newTyConRhs tc)))
454 -- Expose the rep for newtypes if the rep is an FFI type.
455 -- For a very annoying reason. 'Foreign import' is meant to
456 -- be able to look through newtypes transparently, but it
457 -- can only do that if it can "see" the newtype representation
459 exported_con con = any (`elemNameSet` exports)
460 (dataConName con : dataConFieldLabels con)
462 tidyInstances :: (DFunId -> DFunId) -> [Instance] -> [Instance]
463 tidyInstances tidy_dfun ispecs
466 tidy ispec = setInstanceDFunId ispec $
467 tidy_dfun (instanceDFunId ispec)
471 tidyVectInfo :: TidyEnv -> VectInfo -> VectInfo
472 tidyVectInfo (_, var_env) info@(VectInfo { vectInfoVar = vars
473 , vectInfoPADFun = pas
474 , vectInfoIso = isos })
475 = info { vectInfoVar = tidy_vars
476 , vectInfoPADFun = tidy_pas
477 , vectInfoIso = tidy_isos }
480 $ map tidy_var_mapping
483 tidy_pas = mapNameEnv tidy_snd_var pas
484 tidy_isos = mapNameEnv tidy_snd_var isos
486 tidy_var_mapping (from, to) = (from', (from', lookup_var to))
487 where from' = lookup_var from
488 tidy_snd_var (x, var) = (x, lookup_var var)
490 lookup_var var = lookupWithDefaultVarEnv var_env var var
494 %************************************************************************
498 %************************************************************************
500 Note [Injecting implicit bindings]
501 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
502 We inject the implict bindings right at the end, in CoreTidy.
503 Some of these bindings, notably record selectors, are not
504 constructed in an optimised form. E.g. record selector for
505 data T = MkT { x :: {-# UNPACK #-} !Int }
506 Then the unfolding looks like
507 x = \t. case t of MkT x1 -> let x = I# x1 in x
508 This generates bad code unless it's first simplified a bit. That is
509 why CoreUnfold.mkImplicitUnfolding uses simleExprOpt to do a bit of
510 optimisation first. (Only matters when the selector is used curried;
511 eg map x ys.) See Trac #2070.
513 [Oct 09: in fact, record selectors are no longer implicit Ids at all,
514 because we really do want to optimise them properly. They are treated
515 much like any other Id. But doing "light" optimisation on an implicit
516 Id still makes sense.]
518 At one time I tried injecting the implicit bindings *early*, at the
519 beginning of SimplCore. But that gave rise to real difficulty,
520 becuase GlobalIds are supposed to have *fixed* IdInfo, but the
521 simplifier and other core-to-core passes mess with IdInfo all the
522 time. The straw that broke the camels back was when a class selector
523 got the wrong arity -- ie the simplifier gave it arity 2, whereas
524 importing modules were expecting it to have arity 1 (Trac #2844).
525 It's much safer just to inject them right at the end, after tidying.
527 Oh: two other reasons for injecting them late:
529 - If implicit Ids are already in the bindings when we start TidyPgm,
530 we'd have to be careful not to treat them as external Ids (in
531 the sense of findExternalIds); else the Ids mentioned in *their*
532 RHSs will be treated as external and you get an interface file
534 but nothing refererring to a18 (because the implicit Id is the
535 one that does, and implicit Ids don't appear in interface files).
537 - More seriously, the tidied type-envt will include the implicit
538 Id replete with a18 in its unfolding; but we won't take account
539 of a18 when computing a fingerprint for the class; result chaos.
541 There is one sort of implicit binding that is injected still later,
542 namely those for data constructor workers. Reason (I think): it's
543 really just a code generation trick.... binding itself makes no sense.
544 See CorePrep Note [Data constructor workers].
547 getImplicitBinds :: TypeEnv -> [CoreBind]
548 getImplicitBinds type_env
549 = map get_defn (concatMap implicit_ids (typeEnvElts type_env))
551 implicit_ids (ATyCon tc) = mapCatMaybes dataConWrapId_maybe (tyConDataCons tc)
552 implicit_ids (AClass cls) = classSelIds cls
555 get_defn :: Id -> CoreBind
556 get_defn id = NonRec id (unfoldingTemplate (realIdUnfolding id))
560 %************************************************************************
562 \subsection{Step 1: finding externals}
564 %************************************************************************
566 Sete Note [choosing external names].
569 type UnfoldEnv = IdEnv (Name{-new name-}, Bool {-show unfolding-})
570 -- Maps each top-level Id to its new Name (the Id is tidied in step 2)
571 -- The Unique is unchanged. If the new Name is external, it will be
572 -- visible in the interface file.
574 -- Bool => expose unfolding or not.
576 chooseExternalIds :: HscEnv
582 -> IO (UnfoldEnv, TidyOccEnv)
583 -- Step 1 from the notes above
585 chooseExternalIds hsc_env mod omit_prags expose_all binds implicit_binds imp_id_rules
586 = do { (unfold_env1,occ_env1) <- search init_work_list emptyVarEnv init_occ_env
587 ; let internal_ids = filter (not . (`elemVarEnv` unfold_env1)) binders
588 ; tidy_internal internal_ids unfold_env1 occ_env1 }
590 nc_var = hsc_NC hsc_env
592 -- init_ext_ids is the intial list of Ids that should be
593 -- externalised. It serves as the starting point for finding a
594 -- deterministic, tidy, renaming for all external Ids in this
597 -- It is sorted, so that it has adeterministic order (i.e. it's the
598 -- same list every time this module is compiled), in contrast to the
599 -- bindings, which are ordered non-deterministically.
600 init_work_list = zip init_ext_ids init_ext_ids
601 init_ext_ids = sortBy (compare `on` getOccName) $
602 filter is_external binders
604 -- An Id should be external if either (a) it is exported or
605 -- (b) it appears in the RHS of a local rule for an imported Id.
606 -- See Note [Which rules to expose]
607 is_external id = isExportedId id || id `elemVarSet` rule_rhs_vars
608 rule_rhs_vars = foldr (unionVarSet . ruleRhsFreeVars) emptyVarSet imp_id_rules
610 binders = bindersOfBinds binds
611 implicit_binders = bindersOfBinds implicit_binds
612 binder_set = mkVarSet binders
614 avoids = [getOccName name | bndr <- binders ++ implicit_binders,
615 let name = idName bndr,
616 isExternalName name ]
617 -- In computing our "avoids" list, we must include
619 -- all things with global names (assigned once and for
620 -- all by the renamer)
621 -- since their names are "taken".
622 -- The type environment is a convenient source of such things.
623 -- In particular, the set of binders doesn't include
624 -- implicit Ids at this stage.
626 -- We also make sure to avoid any exported binders. Consider
627 -- f{-u1-} = 1 -- Local decl
629 -- f{-u2-} = 2 -- Exported decl
631 -- The second exported decl must 'get' the name 'f', so we
632 -- have to put 'f' in the avoids list before we get to the first
633 -- decl. tidyTopId then does a no-op on exported binders.
634 init_occ_env = initTidyOccEnv avoids
637 search :: [(Id,Id)] -- The work-list: (external id, referrring id)
638 -- Make a tidy, external Name for the external id,
639 -- add it to the UnfoldEnv, and do the same for the
640 -- transitive closure of Ids it refers to
641 -- The referring id is used to generate a tidy
642 --- name for the external id
643 -> UnfoldEnv -- id -> (new Name, show_unfold)
644 -> TidyOccEnv -- occ env for choosing new Names
645 -> IO (UnfoldEnv, TidyOccEnv)
647 search [] unfold_env occ_env = return (unfold_env, occ_env)
649 search ((idocc,referrer) : rest) unfold_env occ_env
650 | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env
652 (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc
654 (new_ids, show_unfold)
655 | omit_prags = ([], False)
656 | otherwise = addExternal expose_all refined_id
658 -- 'idocc' is an *occurrence*, but we need to see the
659 -- unfolding in the *definition*; so look up in binder_set
660 refined_id = case lookupVarSet binder_set idocc of
662 Nothing -> WARN( True, ppr idocc ) idocc
664 unfold_env' = extendVarEnv unfold_env idocc (name',show_unfold)
665 referrer' | isExportedId refined_id = refined_id
666 | otherwise = referrer
668 search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env'
670 tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv
671 -> IO (UnfoldEnv, TidyOccEnv)
672 tidy_internal [] unfold_env occ_env = return (unfold_env,occ_env)
673 tidy_internal (id:ids) unfold_env occ_env = do
674 (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id
675 let unfold_env' = extendVarEnv unfold_env id (name',False)
676 tidy_internal ids unfold_env' occ_env'
678 addExternal :: Bool -> Id -> ([Id],Bool)
679 addExternal expose_all id = (new_needed_ids, show_unfold)
681 new_needed_ids = unfold_ids ++
682 filter (\id -> isLocalId id &&
683 not (id `elemVarSet` unfold_set))
684 (varSetElems spec_ids) -- XXX non-det ordering
687 dont_inline = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo))
688 loop_breaker = isNonRuleLoopBreaker (occInfo idinfo)
689 bottoming_fn = isBottomingSig (strictnessInfo idinfo `orElse` topSig)
690 spec_ids = specInfoFreeVars (specInfo idinfo)
692 -- Stuff to do with the Id's unfolding
693 -- We leave the unfolding there even if there is a worker
694 -- In GHCI the unfolding is used by importers
695 show_unfold = isJust mb_unfold_ids
696 (unfold_set, unfold_ids) = mb_unfold_ids `orElse` (emptyVarSet, [])
698 mb_unfold_ids :: Maybe (IdSet, [Id]) -- Nothing => don't unfold
699 mb_unfold_ids = case unfoldingInfo idinfo of
700 CoreUnfolding { uf_tmpl = unf_rhs, uf_guidance = guide }
701 | expose_all || -- expose_all says to expose all
702 -- unfoldings willy-nilly
703 not (bottoming_fn -- No need to inline bottom functions
704 || dont_inline -- Or ones that say not to
705 || loop_breaker -- Or that are loop breakers
706 || neverUnfoldGuidance guide)
707 -> Just (exprFvsInOrder unf_rhs)
708 DFunUnfolding _ ops -> Just (exprsFvsInOrder ops)
711 -- We want a deterministic free-variable list. exprFreeVars gives us
712 -- a VarSet, which is in a non-deterministic order when converted to a
713 -- list. Hence, here we define a free-variable finder that returns
714 -- the free variables in the order that they are encountered.
716 -- Note [choosing external names]
718 exprFvsInOrder :: CoreExpr -> (VarSet, [Id])
719 exprFvsInOrder e = run (dffvExpr e)
721 exprsFvsInOrder :: [CoreExpr] -> (VarSet, [Id])
722 exprsFvsInOrder es = run (mapM_ dffvExpr es)
724 run :: DFFV () -> (VarSet, [Id])
725 run (DFFV m) = case m emptyVarSet [] of
726 (set,ids,_) -> (set,ids)
728 newtype DFFV a = DFFV (VarSet -> [Var] -> (VarSet,[Var],a))
730 instance Monad DFFV where
731 return a = DFFV $ \set ids -> (set, ids, a)
732 (DFFV m) >>= k = DFFV $ \set ids ->
734 (set',ids',a) -> case k a of
735 DFFV f -> f set' ids'
737 insert :: Var -> DFFV ()
738 insert v = DFFV $ \ set ids -> case () of
739 _ | v `elemVarSet` set -> (set,ids,())
740 | otherwise -> (extendVarSet set v, v:ids, ())
742 dffvExpr :: CoreExpr -> DFFV ()
743 dffvExpr e = go emptyVarSet e
745 go scope e = case e of
746 Var v | isLocalId v && not (v `elemVarSet` scope) -> insert v
747 App e1 e2 -> do go scope e1; go scope e2
748 Lam v e -> go (extendVarSet scope v) e
749 Note _ e -> go scope e
750 Cast e _ -> go scope e
751 Let (NonRec x r) e -> do go scope r; go (extendVarSet scope x) e
752 Let (Rec prs) e -> do let scope' = extendVarSetList scope (map fst prs)
753 mapM_ (go scope') (map snd prs)
755 Case e b _ as -> do go scope e
756 mapM_ (go_alt (extendVarSet scope b)) as
759 go_alt scope (_,xs,r) = go (extendVarSetList scope xs) r
763 --------------------------------------------------------------------
765 -- This is where we set names to local/global based on whether they really are
766 -- externally visible (see comment at the top of this module). If the name
767 -- was previously local, we have to give it a unique occurrence name if
768 -- we intend to externalise it.
771 tidyTopName :: Module -> IORef NameCache -> Maybe Id -> TidyOccEnv
772 -> Id -> IO (TidyOccEnv, Name)
773 tidyTopName mod nc_var maybe_ref occ_env id
774 | global && internal = return (occ_env, localiseName name)
776 | global && external = return (occ_env, name)
777 -- Global names are assumed to have been allocated by the renamer,
778 -- so they already have the "right" unique
779 -- And it's a system-wide unique too
781 -- Now we get to the real reason that all this is in the IO Monad:
782 -- we have to update the name cache in a nice atomic fashion
784 | local && internal = do { nc <- readIORef nc_var
785 ; let (nc', new_local_name) = mk_new_local nc
786 ; writeIORef nc_var nc'
787 ; return (occ_env', new_local_name) }
788 -- Even local, internal names must get a unique occurrence, because
789 -- if we do -split-objs we externalise the name later, in the code generator
791 -- Similarly, we must make sure it has a system-wide Unique, because
792 -- the byte-code generator builds a system-wide Name->BCO symbol table
794 | local && external = do { nc <- readIORef nc_var
795 ; let (nc', new_external_name) = mk_new_external nc
796 ; writeIORef nc_var nc'
797 ; return (occ_env', new_external_name) }
799 | otherwise = panic "tidyTopName"
802 external = isJust maybe_ref
803 global = isExternalName name
805 internal = not external
806 loc = nameSrcSpan name
808 old_occ = nameOccName name
810 | Just ref <- maybe_ref, ref /= id =
811 mkOccName (occNameSpace old_occ) $
813 ref_str = occNameString (getOccName ref)
814 occ_str = occNameString old_occ
818 -- workers: the worker for a function already
819 -- includes the occname for its parent, so there's
820 -- no need to prepend the referrer.
821 _other | isSystemName name -> ref_str
822 | otherwise -> ref_str ++ '_' : occ_str
823 -- If this name was system-generated, then don't bother
824 -- to retain its OccName, just use the referrer. These
825 -- system-generated names will become "f1", "f2", etc. for
827 | otherwise = old_occ
829 (occ_env', occ') = tidyOccName occ_env new_occ
831 mk_new_local nc = (nc { nsUniqs = us2 }, mkInternalName uniq occ' loc)
833 (us1, us2) = splitUniqSupply (nsUniqs nc)
834 uniq = uniqFromSupply us1
836 mk_new_external nc = allocateGlobalBinder nc mod occ' loc
837 -- If we want to externalise a currently-local name, check
838 -- whether we have already assigned a unique for it.
839 -- If so, use it; if not, extend the table.
840 -- All this is done by allcoateGlobalBinder.
841 -- This is needed when *re*-compiling a module in GHCi; we must
842 -- use the same name for externally-visible things as we did before.
846 findExternalRules :: Bool -- Omit pragmas
848 -> [CoreRule] -- Local rules for imported fns
849 -> UnfoldEnv -- Ids that are exported, so we need their rules
851 -- The complete rules are gotten by combining
852 -- a) local rules for imported Ids
853 -- b) rules embedded in the top-level Ids
854 findExternalRules omit_prags binds imp_id_rules unfold_env
856 | otherwise = filterOut internal_rule (imp_id_rules ++ local_rules)
859 | id <- bindersOfBinds binds,
861 rule <- idCoreRules id
865 = any (not . external_id) (varSetElems (ruleLhsFreeIds rule))
866 -- Don't export a rule whose LHS mentions a locally-defined
867 -- Id that is completely internal (i.e. not visible to an
871 | Just (name,_) <- lookupVarEnv unfold_env id = isExternalName name
875 Note [Which rules to expose]
876 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
877 findExternalRules filters imp_rules to avoid binders that
878 aren't externally visible; but the externally-visible binders
879 are computed (by findExternalIds) assuming that all orphan
880 rules are externalised (see init_ext_ids in function
881 'search'). So in fact we may export more than we need.
882 (It's a sort of mutual recursion.)
884 %************************************************************************
886 \subsection{Step 2: top-level tidying}
888 %************************************************************************
892 -- TopTidyEnv: when tidying we need to know
893 -- * nc_var: The NameCache, containing a unique supply and any pre-ordained Names.
894 -- These may have arisen because the
895 -- renamer read in an interface file mentioning M.$wf, say,
896 -- and assigned it unique r77. If, on this compilation, we've
897 -- invented an Id whose name is $wf (but with a different unique)
898 -- we want to rename it to have unique r77, so that we can do easy
899 -- comparisons with stuff from the interface file
901 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
904 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
906 tidyTopBinds :: HscEnv
910 -> (TidyEnv, [CoreBind])
912 tidyTopBinds hsc_env unfold_env init_occ_env binds
913 = tidy init_env binds
915 init_env = (init_occ_env, emptyVarEnv)
917 this_pkg = thisPackage (hsc_dflags hsc_env)
919 tidy env [] = (env, [])
920 tidy env (b:bs) = let (env1, b') = tidyTopBind this_pkg unfold_env env b
921 (env2, bs') = tidy env1 bs
925 ------------------------
926 tidyTopBind :: PackageId
930 -> (TidyEnv, CoreBind)
932 tidyTopBind this_pkg unfold_env (occ_env,subst1) (NonRec bndr rhs)
933 = (tidy_env2, NonRec bndr' rhs')
935 Just (name',show_unfold) = lookupVarEnv unfold_env bndr
936 caf_info = hasCafRefs this_pkg subst1 (idArity bndr) rhs
937 (bndr', rhs') = tidyTopPair show_unfold tidy_env2 caf_info name' (bndr, rhs)
938 subst2 = extendVarEnv subst1 bndr bndr'
939 tidy_env2 = (occ_env, subst2)
941 tidyTopBind this_pkg unfold_env (occ_env,subst1) (Rec prs)
942 = (tidy_env2, Rec prs')
944 prs' = [ tidyTopPair show_unfold tidy_env2 caf_info name' (id,rhs)
946 let (name',show_unfold) =
947 expectJust "tidyTopBind" $ lookupVarEnv unfold_env id
950 subst2 = extendVarEnvList subst1 (bndrs `zip` map fst prs')
951 tidy_env2 = (occ_env, subst2)
955 -- the CafInfo for a recursive group says whether *any* rhs in
956 -- the group may refer indirectly to a CAF (because then, they all do).
958 | or [ mayHaveCafRefs (hasCafRefs this_pkg subst1 (idArity bndr) rhs)
959 | (bndr,rhs) <- prs ] = MayHaveCafRefs
960 | otherwise = NoCafRefs
962 -----------------------------------------------------------
963 tidyTopPair :: Bool -- show unfolding
964 -> TidyEnv -- The TidyEnv is used to tidy the IdInfo
965 -- It is knot-tied: don't look at it!
968 -> (Id, CoreExpr) -- Binder and RHS before tidying
970 -- This function is the heart of Step 2
971 -- The rec_tidy_env is the one to use for the IdInfo
972 -- It's necessary because when we are dealing with a recursive
973 -- group, a variable late in the group might be mentioned
974 -- in the IdInfo of one early in the group
976 tidyTopPair show_unfold rhs_tidy_env caf_info name' (bndr, rhs)
977 = WARN( not _bottom_exposed, ppr bndr1 )
980 -- If the cheap-and-cheerful bottom analyser can see that
981 -- the RHS is bottom, it should jolly well be exposed
982 _bottom_exposed = case exprBotStrictness_maybe rhs of
984 Just (arity, _) -> appIsBottom str arity
986 str = strictnessInfo idinfo `orElse` topSig
988 bndr1 = mkGlobalId details name' ty' idinfo'
989 details = idDetails bndr -- Preserve the IdDetails
990 ty' = tidyTopType (idType bndr)
991 rhs1 = tidyExpr rhs_tidy_env rhs
993 idinfo' = tidyTopIdInfo (isExternalName name')
998 unfold_info | show_unfold = tidyUnfolding rhs_tidy_env rhs1 (unfoldingInfo idinfo)
999 | otherwise = noUnfolding
1000 -- NB: do *not* expose the worker if show_unfold is off,
1001 -- because that means this thing is a loop breaker or
1002 -- marked NOINLINE or something like that
1003 -- This is important: if you expose the worker for a loop-breaker
1004 -- then you can make the simplifier go into an infinite loop, because
1005 -- in effect the unfolding is exposed. See Trac #1709
1007 -- You might think that if show_unfold is False, then the thing should
1008 -- not be w/w'd in the first place. But a legitimate reason is this:
1009 -- the function returns bottom
1010 -- In this case, show_unfold will be false (we don't expose unfoldings
1011 -- for bottoming functions), but we might still have a worker/wrapper
1012 -- split (see Note [Worker-wrapper for bottoming functions] in WorkWrap.lhs
1014 -- Usually the Id will have an accurate arity on it, because
1015 -- the simplifier has just run, but not always.
1016 -- One case I found was when the last thing the simplifier
1017 -- did was to let-bind a non-atomic argument and then float
1018 -- it to the top level. So it seems more robust just to
1020 arity = exprArity rhs
1023 -- tidyTopIdInfo creates the final IdInfo for top-level
1024 -- binders. There are two delicate pieces:
1026 -- * Arity. After CoreTidy, this arity must not change any more.
1027 -- Indeed, CorePrep must eta expand where necessary to make
1028 -- the manifest arity equal to the claimed arity.
1030 -- * CAF info. This must also remain valid through to code generation.
1031 -- We add the info here so that it propagates to all
1032 -- occurrences of the binders in RHSs, and hence to occurrences in
1033 -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES.
1034 -- CoreToStg makes use of this when constructing SRTs.
1035 tidyTopIdInfo :: Bool -> IdInfo -> Unfolding
1036 -> ArityInfo -> CafInfo -> OccInfo
1038 tidyTopIdInfo is_external idinfo unfold_info arity caf_info occ_info
1039 | not is_external -- For internal Ids (not externally visible)
1040 = vanillaIdInfo -- we only need enough info for code generation
1041 -- Arity and strictness info are enough;
1042 -- c.f. CoreTidy.tidyLetBndr
1043 `setOccInfo` robust_occ_info
1044 `setCafInfo` caf_info
1045 `setArityInfo` arity
1046 `setStrictnessInfo` strictnessInfo idinfo
1048 | otherwise -- Externally-visible Ids get the whole lot
1050 `setOccInfo` robust_occ_info
1051 `setCafInfo` caf_info
1052 `setArityInfo` arity
1053 `setStrictnessInfo` strictnessInfo idinfo
1054 `setInlinePragInfo` inlinePragInfo idinfo
1055 `setUnfoldingInfo` unfold_info
1056 -- NB: we throw away the Rules
1057 -- They have already been extracted by findExternalRules
1059 robust_occ_info = zapFragileOcc occ_info
1060 -- It's important to keep loop-breaker information
1061 -- when we are doing -fexpose-all-unfoldings
1065 ------------ Unfolding --------------
1066 tidyUnfolding :: TidyEnv -> CoreExpr -> Unfolding -> Unfolding
1067 tidyUnfolding tidy_env _ unf@(CoreUnfolding { uf_tmpl = rhs
1068 , uf_guidance = guide@(InlineRule {}) })
1069 = unf { uf_tmpl = tidyExpr tidy_env rhs, -- Preserves OccInfo
1070 uf_guidance = guide { ir_info = tidyInl tidy_env (ir_info guide) } }
1071 tidyUnfolding tidy_env _ (DFunUnfolding con ids)
1072 = DFunUnfolding con (map (tidyExpr tidy_env) ids)
1073 tidyUnfolding _ tidy_rhs (CoreUnfolding {})
1074 = mkTopUnfolding tidy_rhs
1075 tidyUnfolding _ _ unf = unf
1077 tidyInl :: TidyEnv -> InlineRuleInfo -> InlineRuleInfo
1078 tidyInl tidy_env (InlWrapper w) = InlWrapper (tidyVarOcc tidy_env w)
1079 tidyInl _ inl_info = inl_info
1082 %************************************************************************
1084 \subsection{Figuring out CafInfo for an expression}
1086 %************************************************************************
1088 hasCafRefs decides whether a top-level closure can point into the dynamic heap.
1089 We mark such things as `MayHaveCafRefs' because this information is
1090 used to decide whether a particular closure needs to be referenced
1093 There are two reasons for setting MayHaveCafRefs:
1094 a) The RHS is a CAF: a top-level updatable thunk.
1095 b) The RHS refers to something that MayHaveCafRefs
1097 Possible improvement: In an effort to keep the number of CAFs (and
1098 hence the size of the SRTs) down, we could also look at the expression and
1099 decide whether it requires a small bounded amount of heap, so we can ignore
1100 it as a CAF. In these cases however, we would need to use an additional
1101 CAF list to keep track of non-collectable CAFs.
1104 hasCafRefs :: PackageId -> VarEnv Var -> Arity -> CoreExpr -> CafInfo
1105 hasCafRefs this_pkg p arity expr
1106 | is_caf || mentions_cafs
1108 | otherwise = NoCafRefs
1110 mentions_cafs = isFastTrue (cafRefs p expr)
1111 is_caf = not (arity > 0 || rhsIsStatic this_pkg expr)
1113 -- NB. we pass in the arity of the expression, which is expected
1114 -- to be calculated by exprArity. This is because exprArity
1115 -- knows how much eta expansion is going to be done by
1116 -- CorePrep later on, and we don't want to duplicate that
1117 -- knowledge in rhsIsStatic below.
1119 cafRefs :: VarEnv Id -> Expr a -> FastBool
1121 -- imported Ids first:
1122 | not (isLocalId id) = fastBool (mayHaveCafRefs (idCafInfo id))
1123 -- now Ids local to this module:
1125 case lookupVarEnv p id of
1126 Just id' -> fastBool (mayHaveCafRefs (idCafInfo id'))
1127 Nothing -> fastBool False
1129 cafRefs _ (Lit _) = fastBool False
1130 cafRefs p (App f a) = fastOr (cafRefs p f) (cafRefs p) a
1131 cafRefs p (Lam _ e) = cafRefs p e
1132 cafRefs p (Let b e) = fastOr (cafRefss p (rhssOfBind b)) (cafRefs p) e
1133 cafRefs p (Case e _bndr _ alts) = fastOr (cafRefs p e) (cafRefss p) (rhssOfAlts alts)
1134 cafRefs p (Note _n e) = cafRefs p e
1135 cafRefs p (Cast e _co) = cafRefs p e
1136 cafRefs _ (Type _) = fastBool False
1138 cafRefss :: VarEnv Id -> [Expr a] -> FastBool
1139 cafRefss _ [] = fastBool False
1140 cafRefss p (e:es) = fastOr (cafRefs p e) (cafRefss p) es
1142 fastOr :: FastBool -> (a -> FastBool) -> a -> FastBool
1143 -- hack for lazy-or over FastBool.
1144 fastOr a f x = fastBool (isFastTrue a || isFastTrue (f x))