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, exprBotStrictness_maybe )
23 import Class ( classAllSelIds )
32 import Name hiding (varName)
40 import Packages( isDllName )
45 import FastBool hiding ( fastOr )
49 import Data.List ( sortBy )
50 import Data.IORef ( IORef, readIORef, writeIORef )
54 Constructing the TypeEnv, Instances, Rules, VectInfo from which the
55 ModIface is constructed, and which goes on to subsequent modules in
58 Most of the interface file is obtained simply by serialising the
59 TypeEnv. One important consequence is that if the *interface file*
60 has pragma info if and only if the final TypeEnv does. This is not so
61 important for *this* module, but it's essential for ghc --make:
62 subsequent compilations must not see (e.g.) the arity if the interface
63 file does not contain arity If they do, they'll exploit the arity;
64 then the arity might change, but the iface file doesn't change =>
65 recompilation does not happen => disaster.
67 For data types, the final TypeEnv will have a TyThing for the TyCon,
68 plus one for each DataCon; the interface file will contain just one
69 data type declaration, but it is de-serialised back into a collection
72 %************************************************************************
76 %************************************************************************
79 Plan A: mkBootModDetails: omit pragmas, make interfaces small
80 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
83 * Drop all WiredIn things from the TypeEnv
84 (we never want them in interface files)
86 * Retain all TyCons and Classes in the TypeEnv, to avoid
87 having to find which ones are mentioned in the
90 * Trim off the constructors of non-exported TyCons, both
91 from the TyCon and from the TypeEnv
93 * Drop non-exported Ids from the TypeEnv
95 * Tidy the types of the DFunIds of Instances,
96 make them into GlobalIds, (they already have External Names)
97 and add them to the TypeEnv
99 * Tidy the types of the (exported) Ids in the TypeEnv,
100 make them into GlobalIds (they already have External Names)
102 * Drop rules altogether
104 * Tidy the bindings, to ensure that the Caf and Arity
105 information is correct for each top-level binder; the
106 code generator needs it. And to ensure that local names have
107 distinct OccNames in case of object-file splitting
110 -- This is Plan A: make a small type env when typechecking only,
111 -- or when compiling a hs-boot file, or simply when not using -O
113 -- We don't look at the bindings at all -- there aren't any
116 mkBootModDetailsTc :: HscEnv -> TcGblEnv -> IO ModDetails
117 mkBootModDetailsTc hsc_env
118 TcGblEnv{ tcg_exports = exports,
119 tcg_type_env = type_env,
121 tcg_fam_insts = fam_insts
123 = mkBootModDetails hsc_env exports type_env insts fam_insts
125 mkBootModDetailsDs :: HscEnv -> ModGuts -> IO ModDetails
126 mkBootModDetailsDs hsc_env
127 ModGuts{ mg_exports = exports,
130 mg_fam_insts = fam_insts
132 = mkBootModDetails hsc_env exports type_env insts fam_insts
134 mkBootModDetails :: HscEnv -> [AvailInfo] -> NameEnv TyThing
135 -> [Instance] -> [FamInstEnv.FamInst] -> IO ModDetails
136 mkBootModDetails hsc_env exports type_env insts fam_insts
137 = do { let dflags = hsc_dflags hsc_env
138 ; showPass dflags CoreTidy
140 ; let { insts' = tidyInstances globaliseAndTidyId insts
141 ; dfun_ids = map instanceDFunId insts'
142 ; type_env1 = tidyBootTypeEnv (availsToNameSet exports) type_env
143 ; type_env' = extendTypeEnvWithIds type_env1 dfun_ids
145 ; return (ModDetails { md_types = type_env'
147 , md_fam_insts = fam_insts
150 , md_exports = exports
151 , md_vect_info = noVectInfo
156 tidyBootTypeEnv :: NameSet -> TypeEnv -> TypeEnv
157 tidyBootTypeEnv exports type_env
158 = tidyTypeEnv True False exports type_env final_ids
160 -- Find the LocalIds in the type env that are exported
161 -- Make them into GlobalIds, and tidy their types
163 -- It's very important to remove the non-exported ones
164 -- because we don't tidy the OccNames, and if we don't remove
165 -- the non-exported ones we'll get many things with the
166 -- same name in the interface file, giving chaos.
167 final_ids = [ globaliseAndTidyId id
168 | id <- typeEnvIds type_env
172 -- default methods have their export flag set, but everything
173 -- else doesn't (yet), because this is pre-desugaring, so we
175 keep_it id = isExportedId id || idName id `elemNameSet` exports
179 globaliseAndTidyId :: Id -> Id
180 -- Takes an LocalId with an External Name,
181 -- makes it into a GlobalId
182 -- * unchanged Name (might be Internal or External)
183 -- * unchanged details
184 -- * VanillaIdInfo (makes a conservative assumption about Caf-hood)
185 globaliseAndTidyId id
186 = Id.setIdType (globaliseId id) tidy_type
188 tidy_type = tidyTopType (idType id)
192 %************************************************************************
194 Plan B: tidy bindings, make TypeEnv full of IdInfo
196 %************************************************************************
198 Plan B: include pragmas, make interfaces
199 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
200 * Figure out which Ids are externally visible
202 * Tidy the bindings, externalising appropriate Ids
204 * Drop all Ids from the TypeEnv, and add all the External Ids from
205 the bindings. (This adds their IdInfo to the TypeEnv; and adds
206 floated-out Ids that weren't even in the TypeEnv before.)
208 Step 1: Figure out external Ids
209 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
210 Note [choosing external names]
212 See also the section "Interface stability" in the
213 RecompilationAvoidance commentary:
214 http://hackage.haskell.org/trac/ghc/wiki/Commentary/Compiler/RecompilationAvoidance
216 First we figure out which Ids are "external" Ids. An
217 "external" Id is one that is visible from outside the compilation
219 a) the user exported ones
220 b) ones mentioned in the unfoldings, workers,
221 or rules of externally-visible ones
223 While figuring out which Ids are external, we pick a "tidy" OccName
224 for each one. That is, we make its OccName distinct from the other
225 external OccNames in this module, so that in interface files and
226 object code we can refer to it unambiguously by its OccName. The
227 OccName for each binder is prefixed by the name of the exported Id
228 that references it; e.g. if "f" references "x" in its unfolding, then
229 "x" is renamed to "f_x". This helps distinguish the different "x"s
230 from each other, and means that if "f" is later removed, things that
231 depend on the other "x"s will not need to be recompiled. Of course,
232 if there are multiple "f_x"s, then we have to disambiguate somehow; we
233 use "f_x0", "f_x1" etc.
235 As far as possible we should assign names in a deterministic fashion.
236 Each time this module is compiled with the same options, we should end
237 up with the same set of external names with the same types. That is,
238 the ABI hash in the interface should not change. This turns out to be
239 quite tricky, since the order of the bindings going into the tidy
240 phase is already non-deterministic, as it is based on the ordering of
241 Uniques, which are assigned unpredictably.
243 To name things in a stable way, we do a depth-first-search of the
244 bindings, starting from the exports sorted by name. This way, as long
245 as the bindings themselves are deterministic (they sometimes aren't!),
246 the order in which they are presented to the tidying phase does not
247 affect the names we assign.
249 Step 2: Tidy the program
250 ~~~~~~~~~~~~~~~~~~~~~~~~
251 Next we traverse the bindings top to bottom. For each *top-level*
254 1. Make it into a GlobalId; its IdDetails becomes VanillaGlobal,
255 reflecting the fact that from now on we regard it as a global,
258 2. Give it a system-wide Unique.
259 [Even non-exported things need system-wide Uniques because the
260 byte-code generator builds a single Name->BCO symbol table.]
262 We use the NameCache kept in the HscEnv as the
263 source of such system-wide uniques.
265 For external Ids, use the original-name cache in the NameCache
266 to ensure that the unique assigned is the same as the Id had
267 in any previous compilation run.
269 3. Rename top-level Ids according to the names we chose in step 1.
270 If it's an external Id, make it have a External Name, otherwise
271 make it have an Internal Name. This is used by the code generator
272 to decide whether to make the label externally visible
274 4. Give it its UTTERLY FINAL IdInfo; in ptic,
275 * its unfolding, if it should have one
277 * its arity, computed from the number of visible lambdas
279 * its CAF info, computed from what is free in its RHS
282 Finally, substitute these new top-level binders consistently
283 throughout, including in unfoldings. We also tidy binders in
284 RHSs, so that they print nicely in interfaces.
287 tidyProgram :: HscEnv -> ModGuts -> IO (CgGuts, ModDetails)
288 tidyProgram hsc_env (ModGuts { mg_module = mod, mg_exports = exports,
290 mg_insts = insts, mg_fam_insts = fam_insts,
292 mg_rules = imp_rules,
293 mg_vect_info = vect_info,
294 mg_dir_imps = dir_imps,
297 mg_foreign = foreign_stubs,
298 mg_hpc_info = hpc_info,
299 mg_modBreaks = modBreaks })
301 = do { let { dflags = hsc_dflags hsc_env
302 ; omit_prags = dopt Opt_OmitInterfacePragmas dflags
303 ; expose_all = dopt Opt_ExposeAllUnfoldings dflags
304 ; th = xopt Opt_TemplateHaskell dflags
306 ; showPass dflags CoreTidy
308 ; let { implicit_binds = getImplicitBinds type_env }
310 ; (unfold_env, tidy_occ_env)
311 <- chooseExternalIds hsc_env mod omit_prags expose_all
312 binds implicit_binds imp_rules
314 ; let { ext_rules = findExternalRules omit_prags binds imp_rules unfold_env }
315 -- Glom together imp_rules and rules currently attached to binders
316 -- Then pick just the ones we need to expose
317 -- See Note [Which rules to expose]
319 ; let { (tidy_env, tidy_binds)
320 = tidyTopBinds hsc_env unfold_env tidy_occ_env binds }
322 ; let { export_set = availsToNameSet exports
323 ; final_ids = [ id | id <- bindersOfBinds tidy_binds,
324 isExternalName (idName id)]
325 ; tidy_type_env = tidyTypeEnv omit_prags th export_set
327 ; tidy_insts = tidyInstances (lookup_dfun tidy_type_env) insts
328 -- A DFunId will have a binding in tidy_binds, and so
329 -- will now be in final_env, replete with IdInfo
330 -- Its name will be unchanged since it was born, but
331 -- we want Global, IdInfo-rich (or not) DFunId in the
334 ; tidy_rules = tidyRules tidy_env ext_rules
335 -- You might worry that the tidy_env contains IdInfo-rich stuff
336 -- and indeed it does, but if omit_prags is on, ext_rules is
339 ; tidy_vect_info = tidyVectInfo tidy_env vect_info
341 -- See Note [Injecting implicit bindings]
342 ; all_tidy_binds = implicit_binds ++ tidy_binds
344 ; alg_tycons = filter isAlgTyCon (typeEnvTyCons type_env)
347 ; endPass dflags CoreTidy all_tidy_binds tidy_rules
349 -- If the endPass didn't print the rules, but ddump-rules is on, print now
350 ; dumpIfSet (dopt Opt_D_dump_rules dflags
351 && (not (dopt Opt_D_dump_simpl dflags)))
353 (ptext (sLit "rules"))
354 (pprRulesForUser 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,
371 md_fam_insts = fam_insts,
372 md_exports = exports,
373 md_anns = anns -- are already tidy
377 lookup_dfun :: TypeEnv -> Var -> Id
378 lookup_dfun type_env dfun_id
379 = case lookupTypeEnv type_env (idName dfun_id) of
380 Just (AnId dfun_id') -> dfun_id'
381 _other -> pprPanic "lookup_dfun" (ppr dfun_id)
383 --------------------------
384 tidyTypeEnv :: Bool -- Compiling without -O, so omit prags
385 -> Bool -- Template Haskell is on
386 -> NameSet -> TypeEnv -> [Id] -> TypeEnv
388 -- The competed type environment is gotten from
389 -- Dropping any wired-in things, and then
390 -- a) keeping the types and classes
391 -- b) removing all Ids,
392 -- c) adding Ids with correct IdInfo, including unfoldings,
393 -- gotten from the bindings
394 -- From (c) we keep only those Ids with External names;
395 -- the CoreTidy pass makes sure these are all and only
396 -- the externally-accessible ones
397 -- This truncates the type environment to include only the
398 -- exported Ids and things needed from them, which saves space
400 tidyTypeEnv omit_prags th exports type_env final_ids
401 = let type_env1 = filterNameEnv keep_it type_env
402 type_env2 = extendTypeEnvWithIds type_env1 final_ids
403 type_env3 | omit_prags = mapNameEnv (trimThing th exports) type_env2
404 | otherwise = type_env2
408 -- We keep GlobalIds, because they won't appear
409 -- in the bindings from which final_ids are derived!
410 -- (The bindings bind LocalIds.)
411 keep_it thing | isWiredInThing thing = False
412 keep_it (AnId id) = isGlobalId id -- Keep GlobalIds (e.g. class ops)
413 keep_it _other = True -- Keep all TyCons, DataCons, and Classes
415 --------------------------
416 isWiredInThing :: TyThing -> Bool
417 isWiredInThing thing = isWiredInName (getName thing)
419 --------------------------
420 trimThing :: Bool -> NameSet -> TyThing -> TyThing
421 -- Trim off inessentials, for boot files and no -O
422 trimThing th exports (ATyCon tc)
423 | not th && not (mustExposeTyCon exports tc)
424 = ATyCon (makeTyConAbstract tc) -- Note [Trimming and Template Haskell]
426 trimThing _th _exports (AnId id)
427 | not (isImplicitId id)
428 = AnId (id `setIdInfo` vanillaIdInfo)
430 trimThing _th _exports other_thing
434 {- Note [Trimming and Template Haskell]
435 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
436 Consider (Trac #2386) this
437 module M(T, makeOne) where
439 makeOne = [| Yay "Yep" |]
440 Notice that T is exported abstractly, but makeOne effectively exports it too!
441 A module that splices in $(makeOne) will then look for a declartion of Yay,
442 so it'd better be there. Hence, brutally but simply, we switch off type
443 constructor trimming if TH is enabled in this module. -}
446 mustExposeTyCon :: NameSet -- Exports
447 -> TyCon -- The tycon
448 -> Bool -- Can its rep be hidden?
449 -- We are compiling without -O, and thus trying to write as little as
450 -- possible into the interface file. But we must expose the details of
451 -- any data types whose constructors or fields are exported
452 mustExposeTyCon exports tc
453 | not (isAlgTyCon tc) -- Synonyms
455 | isEnumerationTyCon tc -- For an enumeration, exposing the constructors
456 = True -- won't lead to the need for further exposure
457 -- (This includes data types with no constructors.)
458 | isFamilyTyCon tc -- Open type family
461 | otherwise -- Newtype, datatype
462 = any exported_con (tyConDataCons tc)
463 -- Expose rep if any datacon or field is exported
465 || (isNewTyCon tc && isFFITy (snd (newTyConRhs tc)))
466 -- Expose the rep for newtypes if the rep is an FFI type.
467 -- For a very annoying reason. 'Foreign import' is meant to
468 -- be able to look through newtypes transparently, but it
469 -- can only do that if it can "see" the newtype representation
471 exported_con con = any (`elemNameSet` exports)
472 (dataConName con : dataConFieldLabels con)
474 tidyInstances :: (DFunId -> DFunId) -> [Instance] -> [Instance]
475 tidyInstances tidy_dfun ispecs
478 tidy ispec = setInstanceDFunId ispec $
479 tidy_dfun (instanceDFunId ispec)
483 tidyVectInfo :: TidyEnv -> VectInfo -> VectInfo
484 tidyVectInfo (_, var_env) info@(VectInfo { vectInfoVar = vars
485 , vectInfoPADFun = pas
486 , vectInfoIso = isos })
487 = info { vectInfoVar = tidy_vars
488 , vectInfoPADFun = tidy_pas
489 , vectInfoIso = tidy_isos }
492 $ map tidy_var_mapping
495 tidy_pas = mapNameEnv tidy_snd_var pas
496 tidy_isos = mapNameEnv tidy_snd_var isos
498 tidy_var_mapping (from, to) = (from', (from', lookup_var to))
499 where from' = lookup_var from
500 tidy_snd_var (x, var) = (x, lookup_var var)
502 lookup_var var = lookupWithDefaultVarEnv var_env var var
506 %************************************************************************
510 %************************************************************************
512 Note [Injecting implicit bindings]
513 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
514 We inject the implict bindings right at the end, in CoreTidy.
515 Some of these bindings, notably record selectors, are not
516 constructed in an optimised form. E.g. record selector for
517 data T = MkT { x :: {-# UNPACK #-} !Int }
518 Then the unfolding looks like
519 x = \t. case t of MkT x1 -> let x = I# x1 in x
520 This generates bad code unless it's first simplified a bit. That is
521 why CoreUnfold.mkImplicitUnfolding uses simleExprOpt to do a bit of
522 optimisation first. (Only matters when the selector is used curried;
523 eg map x ys.) See Trac #2070.
525 [Oct 09: in fact, record selectors are no longer implicit Ids at all,
526 because we really do want to optimise them properly. They are treated
527 much like any other Id. But doing "light" optimisation on an implicit
528 Id still makes sense.]
530 At one time I tried injecting the implicit bindings *early*, at the
531 beginning of SimplCore. But that gave rise to real difficulty,
532 becuase GlobalIds are supposed to have *fixed* IdInfo, but the
533 simplifier and other core-to-core passes mess with IdInfo all the
534 time. The straw that broke the camels back was when a class selector
535 got the wrong arity -- ie the simplifier gave it arity 2, whereas
536 importing modules were expecting it to have arity 1 (Trac #2844).
537 It's much safer just to inject them right at the end, after tidying.
539 Oh: two other reasons for injecting them late:
541 - If implicit Ids are already in the bindings when we start TidyPgm,
542 we'd have to be careful not to treat them as external Ids (in
543 the sense of findExternalIds); else the Ids mentioned in *their*
544 RHSs will be treated as external and you get an interface file
546 but nothing refererring to a18 (because the implicit Id is the
547 one that does, and implicit Ids don't appear in interface files).
549 - More seriously, the tidied type-envt will include the implicit
550 Id replete with a18 in its unfolding; but we won't take account
551 of a18 when computing a fingerprint for the class; result chaos.
553 There is one sort of implicit binding that is injected still later,
554 namely those for data constructor workers. Reason (I think): it's
555 really just a code generation trick.... binding itself makes no sense.
556 See CorePrep Note [Data constructor workers].
559 getImplicitBinds :: TypeEnv -> [CoreBind]
560 getImplicitBinds type_env
561 = map get_defn (concatMap implicit_ids (typeEnvElts type_env))
563 implicit_ids (ATyCon tc) = mapCatMaybes dataConWrapId_maybe (tyConDataCons tc)
564 implicit_ids (AClass cls) = classAllSelIds cls
567 get_defn :: Id -> CoreBind
568 get_defn id = NonRec id (unfoldingTemplate (realIdUnfolding id))
572 %************************************************************************
574 \subsection{Step 1: finding externals}
576 %************************************************************************
578 Sete Note [choosing external names].
581 type UnfoldEnv = IdEnv (Name{-new name-}, Bool {-show unfolding-})
582 -- Maps each top-level Id to its new Name (the Id is tidied in step 2)
583 -- The Unique is unchanged. If the new Name is external, it will be
584 -- visible in the interface file.
586 -- Bool => expose unfolding or not.
588 chooseExternalIds :: HscEnv
594 -> IO (UnfoldEnv, TidyOccEnv)
595 -- Step 1 from the notes above
597 chooseExternalIds hsc_env mod omit_prags expose_all binds implicit_binds imp_id_rules
598 = do { (unfold_env1,occ_env1) <- search init_work_list 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 -- init_ext_ids is the intial list of Ids that should be
605 -- externalised. It serves as the starting point for finding a
606 -- deterministic, tidy, renaming for all external Ids in this
609 -- It is sorted, so that it has adeterministic order (i.e. it's the
610 -- same list every time this module is compiled), in contrast to the
611 -- bindings, which are ordered non-deterministically.
612 init_work_list = zip init_ext_ids init_ext_ids
613 init_ext_ids = sortBy (compare `on` getOccName) $
614 filter is_external binders
616 -- An Id should be external if either (a) it is exported or
617 -- (b) it appears in the RHS of a local rule for an imported Id.
618 -- See Note [Which rules to expose]
619 is_external id = isExportedId id || id `elemVarSet` rule_rhs_vars
620 rule_rhs_vars = foldr (unionVarSet . ruleRhsFreeVars) emptyVarSet imp_id_rules
622 binders = bindersOfBinds binds
623 implicit_binders = bindersOfBinds implicit_binds
624 binder_set = mkVarSet binders
626 avoids = [getOccName name | bndr <- binders ++ implicit_binders,
627 let name = idName bndr,
628 isExternalName name ]
629 -- In computing our "avoids" list, we must include
631 -- all things with global names (assigned once and for
632 -- all by the renamer)
633 -- since their names are "taken".
634 -- The type environment is a convenient source of such things.
635 -- In particular, the set of binders doesn't include
636 -- implicit Ids at this stage.
638 -- We also make sure to avoid any exported binders. Consider
639 -- f{-u1-} = 1 -- Local decl
641 -- f{-u2-} = 2 -- Exported decl
643 -- The second exported decl must 'get' the name 'f', so we
644 -- have to put 'f' in the avoids list before we get to the first
645 -- decl. tidyTopId then does a no-op on exported binders.
646 init_occ_env = initTidyOccEnv avoids
649 search :: [(Id,Id)] -- The work-list: (external id, referrring id)
650 -- Make a tidy, external Name for the external id,
651 -- add it to the UnfoldEnv, and do the same for the
652 -- transitive closure of Ids it refers to
653 -- The referring id is used to generate a tidy
654 --- name for the external id
655 -> UnfoldEnv -- id -> (new Name, show_unfold)
656 -> TidyOccEnv -- occ env for choosing new Names
657 -> IO (UnfoldEnv, TidyOccEnv)
659 search [] unfold_env occ_env = return (unfold_env, occ_env)
661 search ((idocc,referrer) : rest) unfold_env occ_env
662 | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env
664 (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc
666 (new_ids, show_unfold)
667 | omit_prags = ([], False)
668 | otherwise = addExternal expose_all refined_id
670 -- 'idocc' is an *occurrence*, but we need to see the
671 -- unfolding in the *definition*; so look up in binder_set
672 refined_id = case lookupVarSet binder_set idocc of
674 Nothing -> WARN( True, ppr idocc ) idocc
676 unfold_env' = extendVarEnv unfold_env idocc (name',show_unfold)
677 referrer' | isExportedId refined_id = refined_id
678 | otherwise = referrer
680 search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env'
682 tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv
683 -> IO (UnfoldEnv, TidyOccEnv)
684 tidy_internal [] unfold_env occ_env = return (unfold_env,occ_env)
685 tidy_internal (id:ids) unfold_env occ_env = do
686 (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id
687 let unfold_env' = extendVarEnv unfold_env id (name',False)
688 tidy_internal ids unfold_env' occ_env'
690 addExternal :: Bool -> Id -> ([Id],Bool)
691 addExternal expose_all id = (new_needed_ids, show_unfold)
693 new_needed_ids = unfold_ids ++
694 filter (\id -> isLocalId id &&
695 not (id `elemVarSet` unfold_set))
696 (varSetElems spec_ids) -- XXX non-det ordering
699 never_active = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo))
700 loop_breaker = isNonRuleLoopBreaker (occInfo idinfo)
701 bottoming_fn = isBottomingSig (strictnessInfo idinfo `orElse` topSig)
702 spec_ids = specInfoFreeVars (specInfo idinfo)
704 -- Stuff to do with the Id's unfolding
705 -- We leave the unfolding there even if there is a worker
706 -- In GHCI the unfolding is used by importers
707 show_unfold = isJust mb_unfold_ids
708 (unfold_set, unfold_ids) = mb_unfold_ids `orElse` (emptyVarSet, [])
710 mb_unfold_ids :: Maybe (IdSet, [Id]) -- Nothing => don't unfold
711 mb_unfold_ids = case unfoldingInfo idinfo of
712 CoreUnfolding { uf_tmpl = unf_rhs, uf_src = src, uf_guidance = guide }
713 | show_unfolding src guide
714 -> Just (unf_ext_ids src unf_rhs)
715 DFunUnfolding _ _ ops -> Just (exprsFvsInOrder (dfunArgExprs ops))
718 unf_ext_ids (InlineWrapper v) _ = (unitVarSet v, [v])
719 unf_ext_ids _ unf_rhs = exprFvsInOrder unf_rhs
720 -- For a wrapper, externalise the wrapper id rather than the
721 -- fvs of the rhs. The two usually come down to the same thing
722 -- but I've seen cases where we had a wrapper id $w but a
723 -- rhs where $w had been inlined; see Trac #3922
725 show_unfolding unf_source unf_guidance
726 = expose_all -- 'expose_all' says to expose all
727 -- unfoldings willy-nilly
729 || isStableSource unf_source -- Always expose things whose
730 -- source is an inline rule
732 || not (bottoming_fn -- No need to inline bottom functions
733 || never_active -- Or ones that say not to
734 || loop_breaker -- Or that are loop breakers
735 || neverUnfoldGuidance unf_guidance)
737 -- We want a deterministic free-variable list. exprFreeVars gives us
738 -- a VarSet, which is in a non-deterministic order when converted to a
739 -- list. Hence, here we define a free-variable finder that returns
740 -- the free variables in the order that they are encountered.
742 -- Note [choosing external names]
744 exprFvsInOrder :: CoreExpr -> (VarSet, [Id])
745 exprFvsInOrder e = run (dffvExpr e)
747 exprsFvsInOrder :: [CoreExpr] -> (VarSet, [Id])
748 exprsFvsInOrder es = run (mapM_ dffvExpr es)
750 run :: DFFV () -> (VarSet, [Id])
751 run (DFFV m) = case m emptyVarSet [] of
752 (set,ids,_) -> (set,ids)
754 newtype DFFV a = DFFV (VarSet -> [Var] -> (VarSet,[Var],a))
756 instance Monad DFFV where
757 return a = DFFV $ \set ids -> (set, ids, a)
758 (DFFV m) >>= k = DFFV $ \set ids ->
760 (set',ids',a) -> case k a of
761 DFFV f -> f set' ids'
763 insert :: Var -> DFFV ()
764 insert v = DFFV $ \ set ids -> case () of
765 _ | v `elemVarSet` set -> (set,ids,())
766 | otherwise -> (extendVarSet set v, v:ids, ())
768 dffvExpr :: CoreExpr -> DFFV ()
769 dffvExpr e = go emptyVarSet e
771 go scope e = case e of
772 Var v | isLocalId v && not (v `elemVarSet` scope) -> insert v
773 App e1 e2 -> do go scope e1; go scope e2
774 Lam v e -> go (extendVarSet scope v) e
775 Note _ e -> go scope e
776 Cast e _ -> go scope e
777 Let (NonRec x r) e -> do go scope r; go (extendVarSet scope x) e
778 Let (Rec prs) e -> do let scope' = extendVarSetList scope (map fst prs)
779 mapM_ (go scope') (map snd prs)
781 Case e b _ as -> do go scope e
782 mapM_ (go_alt (extendVarSet scope b)) as
785 go_alt scope (_,xs,r) = go (extendVarSetList scope xs) r
789 --------------------------------------------------------------------
791 -- This is where we set names to local/global based on whether they really are
792 -- externally visible (see comment at the top of this module). If the name
793 -- was previously local, we have to give it a unique occurrence name if
794 -- we intend to externalise it.
797 tidyTopName :: Module -> IORef NameCache -> Maybe Id -> TidyOccEnv
798 -> Id -> IO (TidyOccEnv, Name)
799 tidyTopName mod nc_var maybe_ref occ_env id
800 | global && internal = return (occ_env, localiseName name)
802 | global && external = return (occ_env, name)
803 -- Global names are assumed to have been allocated by the renamer,
804 -- so they already have the "right" unique
805 -- And it's a system-wide unique too
807 -- Now we get to the real reason that all this is in the IO Monad:
808 -- we have to update the name cache in a nice atomic fashion
810 | local && internal = do { nc <- readIORef nc_var
811 ; let (nc', new_local_name) = mk_new_local nc
812 ; writeIORef nc_var nc'
813 ; return (occ_env', new_local_name) }
814 -- Even local, internal names must get a unique occurrence, because
815 -- if we do -split-objs we externalise the name later, in the code generator
817 -- Similarly, we must make sure it has a system-wide Unique, because
818 -- the byte-code generator builds a system-wide Name->BCO symbol table
820 | local && external = do { nc <- readIORef nc_var
821 ; let (nc', new_external_name) = mk_new_external nc
822 ; writeIORef nc_var nc'
823 ; return (occ_env', new_external_name) }
825 | otherwise = panic "tidyTopName"
828 external = isJust maybe_ref
829 global = isExternalName name
831 internal = not external
832 loc = nameSrcSpan name
834 old_occ = nameOccName name
836 | Just ref <- maybe_ref, ref /= id =
837 mkOccName (occNameSpace old_occ) $
839 ref_str = occNameString (getOccName ref)
840 occ_str = occNameString old_occ
844 -- workers: the worker for a function already
845 -- includes the occname for its parent, so there's
846 -- no need to prepend the referrer.
847 _other | isSystemName name -> ref_str
848 | otherwise -> ref_str ++ '_' : occ_str
849 -- If this name was system-generated, then don't bother
850 -- to retain its OccName, just use the referrer. These
851 -- system-generated names will become "f1", "f2", etc. for
853 | otherwise = old_occ
855 (occ_env', occ') = tidyOccName occ_env new_occ
857 mk_new_local nc = (nc { nsUniqs = us }, mkInternalName uniq occ' loc)
859 (uniq, us) = takeUniqFromSupply (nsUniqs nc)
861 mk_new_external nc = allocateGlobalBinder nc mod occ' loc
862 -- If we want to externalise a currently-local name, check
863 -- whether we have already assigned a unique for it.
864 -- If so, use it; if not, extend the table.
865 -- All this is done by allcoateGlobalBinder.
866 -- This is needed when *re*-compiling a module in GHCi; we must
867 -- use the same name for externally-visible things as we did before.
871 findExternalRules :: Bool -- Omit pragmas
873 -> [CoreRule] -- Local rules for imported fns
874 -> UnfoldEnv -- Ids that are exported, so we need their rules
876 -- The complete rules are gotten by combining
877 -- a) local rules for imported Ids
878 -- b) rules embedded in the top-level Ids
879 findExternalRules omit_prags binds imp_id_rules unfold_env
881 | otherwise = filterOut internal_rule (imp_id_rules ++ local_rules)
884 | id <- bindersOfBinds binds,
886 rule <- idCoreRules id
890 = any (not . external_id) (varSetElems (ruleLhsFreeIds rule))
891 -- Don't export a rule whose LHS mentions a locally-defined
892 -- Id that is completely internal (i.e. not visible to an
896 | Just (name,_) <- lookupVarEnv unfold_env id = isExternalName name
900 Note [Which rules to expose]
901 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
902 findExternalRules filters imp_rules to avoid binders that
903 aren't externally visible; but the externally-visible binders
904 are computed (by findExternalIds) assuming that all orphan
905 rules are externalised (see init_ext_ids in function
906 'search'). So in fact we may export more than we need.
907 (It's a sort of mutual recursion.)
909 %************************************************************************
911 \subsection{Step 2: top-level tidying}
913 %************************************************************************
917 -- TopTidyEnv: when tidying we need to know
918 -- * nc_var: The NameCache, containing a unique supply and any pre-ordained Names.
919 -- These may have arisen because the
920 -- renamer read in an interface file mentioning M.$wf, say,
921 -- and assigned it unique r77. If, on this compilation, we've
922 -- invented an Id whose name is $wf (but with a different unique)
923 -- we want to rename it to have unique r77, so that we can do easy
924 -- comparisons with stuff from the interface file
926 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
929 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
931 tidyTopBinds :: HscEnv
935 -> (TidyEnv, [CoreBind])
937 tidyTopBinds hsc_env unfold_env init_occ_env binds
938 = tidy init_env binds
940 init_env = (init_occ_env, emptyVarEnv)
942 this_pkg = thisPackage (hsc_dflags hsc_env)
944 tidy env [] = (env, [])
945 tidy env (b:bs) = let (env1, b') = tidyTopBind this_pkg unfold_env env b
946 (env2, bs') = tidy env1 bs
950 ------------------------
951 tidyTopBind :: PackageId
955 -> (TidyEnv, CoreBind)
957 tidyTopBind this_pkg unfold_env (occ_env,subst1) (NonRec bndr rhs)
958 = (tidy_env2, NonRec bndr' rhs')
960 Just (name',show_unfold) = lookupVarEnv unfold_env bndr
961 caf_info = hasCafRefs this_pkg subst1 (idArity bndr) rhs
962 (bndr', rhs') = tidyTopPair show_unfold tidy_env2 caf_info name' (bndr, rhs)
963 subst2 = extendVarEnv subst1 bndr bndr'
964 tidy_env2 = (occ_env, subst2)
966 tidyTopBind this_pkg unfold_env (occ_env,subst1) (Rec prs)
967 = (tidy_env2, Rec prs')
969 prs' = [ tidyTopPair show_unfold tidy_env2 caf_info name' (id,rhs)
971 let (name',show_unfold) =
972 expectJust "tidyTopBind" $ lookupVarEnv unfold_env id
975 subst2 = extendVarEnvList subst1 (bndrs `zip` map fst prs')
976 tidy_env2 = (occ_env, subst2)
980 -- the CafInfo for a recursive group says whether *any* rhs in
981 -- the group may refer indirectly to a CAF (because then, they all do).
983 | or [ mayHaveCafRefs (hasCafRefs this_pkg subst1 (idArity bndr) rhs)
984 | (bndr,rhs) <- prs ] = MayHaveCafRefs
985 | otherwise = NoCafRefs
987 -----------------------------------------------------------
988 tidyTopPair :: Bool -- show unfolding
989 -> TidyEnv -- The TidyEnv is used to tidy the IdInfo
990 -- It is knot-tied: don't look at it!
993 -> (Id, CoreExpr) -- Binder and RHS before tidying
995 -- This function is the heart of Step 2
996 -- The rec_tidy_env is the one to use for the IdInfo
997 -- It's necessary because when we are dealing with a recursive
998 -- group, a variable late in the group might be mentioned
999 -- in the IdInfo of one early in the group
1001 tidyTopPair show_unfold rhs_tidy_env caf_info name' (bndr, rhs)
1004 bndr1 = mkGlobalId details name' ty' idinfo'
1005 details = idDetails bndr -- Preserve the IdDetails
1006 ty' = tidyTopType (idType bndr)
1007 rhs1 = tidyExpr rhs_tidy_env rhs
1008 idinfo' = tidyTopIdInfo rhs_tidy_env name' rhs rhs1 (idInfo bndr)
1009 show_unfold caf_info
1011 -- tidyTopIdInfo creates the final IdInfo for top-level
1012 -- binders. There are two delicate pieces:
1014 -- * Arity. After CoreTidy, this arity must not change any more.
1015 -- Indeed, CorePrep must eta expand where necessary to make
1016 -- the manifest arity equal to the claimed arity.
1018 -- * CAF info. This must also remain valid through to code generation.
1019 -- We add the info here so that it propagates to all
1020 -- occurrences of the binders in RHSs, and hence to occurrences in
1021 -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES.
1022 -- CoreToStg makes use of this when constructing SRTs.
1023 tidyTopIdInfo :: TidyEnv -> Name -> CoreExpr -> CoreExpr
1024 -> IdInfo -> Bool -> CafInfo -> IdInfo
1025 tidyTopIdInfo rhs_tidy_env name orig_rhs tidy_rhs idinfo show_unfold caf_info
1026 | not is_external -- For internal Ids (not externally visible)
1027 = vanillaIdInfo -- we only need enough info for code generation
1028 -- Arity and strictness info are enough;
1029 -- c.f. CoreTidy.tidyLetBndr
1030 `setCafInfo` caf_info
1031 `setArityInfo` arity
1032 `setStrictnessInfo` final_sig
1034 | otherwise -- Externally-visible Ids get the whole lot
1036 `setCafInfo` caf_info
1037 `setArityInfo` arity
1038 `setStrictnessInfo` final_sig
1039 `setOccInfo` robust_occ_info
1040 `setInlinePragInfo` (inlinePragInfo idinfo)
1041 `setUnfoldingInfo` unfold_info
1042 -- NB: we throw away the Rules
1043 -- They have already been extracted by findExternalRules
1045 is_external = isExternalName name
1047 --------- OccInfo ------------
1048 robust_occ_info = zapFragileOcc (occInfo idinfo)
1049 -- It's important to keep loop-breaker information
1050 -- when we are doing -fexpose-all-unfoldings
1052 --------- Strictness ------------
1053 final_sig | Just sig <- strictnessInfo idinfo
1054 = WARN( _bottom_hidden sig, ppr name ) Just sig
1055 | Just (_, sig) <- mb_bot_str = Just sig
1056 | otherwise = Nothing
1058 -- If the cheap-and-cheerful bottom analyser can see that
1059 -- the RHS is bottom, it should jolly well be exposed
1060 _bottom_hidden id_sig = case mb_bot_str of
1062 Just (arity, _) -> not (appIsBottom id_sig arity)
1064 mb_bot_str = exprBotStrictness_maybe orig_rhs
1066 --------- Unfolding ------------
1067 unf_info = unfoldingInfo idinfo
1068 unfold_info | show_unfold = tidyUnfolding rhs_tidy_env unf_info unf_from_rhs
1069 | otherwise = noUnfolding
1070 unf_from_rhs = mkTopUnfolding is_bot tidy_rhs
1071 is_bot = case final_sig of
1072 Just sig -> isBottomingSig sig
1074 -- NB: do *not* expose the worker if show_unfold is off,
1075 -- because that means this thing is a loop breaker or
1076 -- marked NOINLINE or something like that
1077 -- This is important: if you expose the worker for a loop-breaker
1078 -- then you can make the simplifier go into an infinite loop, because
1079 -- in effect the unfolding is exposed. See Trac #1709
1081 -- You might think that if show_unfold is False, then the thing should
1082 -- not be w/w'd in the first place. But a legitimate reason is this:
1083 -- the function returns bottom
1084 -- In this case, show_unfold will be false (we don't expose unfoldings
1085 -- for bottoming functions), but we might still have a worker/wrapper
1086 -- split (see Note [Worker-wrapper for bottoming functions] in WorkWrap.lhs
1088 --------- Arity ------------
1089 -- Usually the Id will have an accurate arity on it, because
1090 -- the simplifier has just run, but not always.
1091 -- One case I found was when the last thing the simplifier
1092 -- did was to let-bind a non-atomic argument and then float
1093 -- it to the top level. So it seems more robust just to
1095 arity = exprArity orig_rhs
1098 %************************************************************************
1100 \subsection{Figuring out CafInfo for an expression}
1102 %************************************************************************
1104 hasCafRefs decides whether a top-level closure can point into the dynamic heap.
1105 We mark such things as `MayHaveCafRefs' because this information is
1106 used to decide whether a particular closure needs to be referenced
1109 There are two reasons for setting MayHaveCafRefs:
1110 a) The RHS is a CAF: a top-level updatable thunk.
1111 b) The RHS refers to something that MayHaveCafRefs
1113 Possible improvement: In an effort to keep the number of CAFs (and
1114 hence the size of the SRTs) down, we could also look at the expression and
1115 decide whether it requires a small bounded amount of heap, so we can ignore
1116 it as a CAF. In these cases however, we would need to use an additional
1117 CAF list to keep track of non-collectable CAFs.
1120 hasCafRefs :: PackageId -> VarEnv Var -> Arity -> CoreExpr -> CafInfo
1121 hasCafRefs this_pkg p arity expr
1122 | is_caf || mentions_cafs = MayHaveCafRefs
1123 | otherwise = NoCafRefs
1125 mentions_cafs = isFastTrue (cafRefs p expr)
1126 is_dynamic_name = isDllName this_pkg
1127 is_caf = not (arity > 0 || rhsIsStatic is_dynamic_name expr)
1129 -- NB. we pass in the arity of the expression, which is expected
1130 -- to be calculated by exprArity. This is because exprArity
1131 -- knows how much eta expansion is going to be done by
1132 -- CorePrep later on, and we don't want to duplicate that
1133 -- knowledge in rhsIsStatic below.
1135 cafRefs :: VarEnv Id -> Expr a -> FastBool
1137 -- imported Ids first:
1138 | not (isLocalId id) = fastBool (mayHaveCafRefs (idCafInfo id))
1139 -- now Ids local to this module:
1141 case lookupVarEnv p id of
1142 Just id' -> fastBool (mayHaveCafRefs (idCafInfo id'))
1143 Nothing -> fastBool False
1145 cafRefs _ (Lit _) = fastBool False
1146 cafRefs p (App f a) = fastOr (cafRefs p f) (cafRefs p) a
1147 cafRefs p (Lam _ e) = cafRefs p e
1148 cafRefs p (Let b e) = fastOr (cafRefss p (rhssOfBind b)) (cafRefs p) e
1149 cafRefs p (Case e _bndr _ alts) = fastOr (cafRefs p e) (cafRefss p) (rhssOfAlts alts)
1150 cafRefs p (Note _n e) = cafRefs p e
1151 cafRefs p (Cast e _co) = cafRefs p e
1152 cafRefs _ (Type _) = fastBool False
1154 cafRefss :: VarEnv Id -> [Expr a] -> FastBool
1155 cafRefss _ [] = fastBool False
1156 cafRefss p (e:es) = fastOr (cafRefs p e) (cafRefss p) es
1158 fastOr :: FastBool -> (a -> FastBool) -> a -> FastBool
1159 -- hack for lazy-or over FastBool.
1160 fastOr a f x = fastBool (isFastTrue a || isFastTrue (f x))