2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section{Tidying up Core}
8 tidyCorePgm, tidyExpr, tidyCoreExpr,
12 #include "HsVersions.h"
14 import CmdLineOpts ( DynFlags, DynFlag(..), opt_OmitInterfacePragmas )
16 import CoreUnfold ( noUnfolding, mkTopUnfolding, okToUnfoldInHiFile )
17 import CoreFVs ( ruleSomeFreeVars, exprSomeFreeVars )
18 import CoreLint ( showPass, endPass )
21 import Var ( Id, Var )
22 import Id ( idType, idInfo, idName, isExportedId,
23 idSpecialisation, idUnique, isDataConWrapId,
24 mkVanillaGlobal, isLocalId, isRecordSelector,
25 setIdUnfolding, hasNoBinding, mkUserLocal
27 import IdInfo {- loads of stuff -}
28 import Name ( getOccName, nameOccName, globaliseName, setNameOcc,
29 localiseName, isGlobalName, setNameUnique
31 import NameEnv ( filterNameEnv )
32 import OccName ( TidyOccEnv, initTidyOccEnv, tidyOccName )
33 import Type ( tidyTopType, tidyType, tidyTyVar )
34 import Module ( Module, moduleName )
35 import HscTypes ( PersistentCompilerState( pcs_PRS ),
36 PersistentRenamerState( prsOrig ),
37 NameSupply( nsNames, nsUniqs ),
38 TypeEnv, extendTypeEnvList,
39 ModDetails(..), TyThing(..)
41 import FiniteMap ( lookupFM, addToFM )
42 import Maybes ( maybeToBool, orElse )
43 import ErrUtils ( showPass )
44 import SrcLoc ( noSrcLoc )
45 import UniqFM ( mapUFM )
46 import UniqSupply ( splitUniqSupply, uniqFromSupply )
47 import List ( partition )
48 import Util ( mapAccumL )
54 %************************************************************************
56 \subsection{What goes on}
58 %************************************************************************
64 Step 1: Figure out external Ids
65 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
66 First we figure out which Ids are "external" Ids. An
67 "external" Id is one that is visible from outside the compilation
69 a) the user exported ones
70 b) ones mentioned in the unfoldings, workers,
71 or rules of externally-visible ones
72 This exercise takes a sweep of the bindings bottom to top. Actually,
73 in Step 2 we're also going to need to know which Ids should be
74 exported with their unfoldings, so we produce not an IdSet but an
78 Step 2: Tidy the program
79 ~~~~~~~~~~~~~~~~~~~~~~~~
80 Next we traverse the bindings top to bottom. For each *top-level*
83 1. Make it into a GlobalId
85 2. Give it a system-wide Unique.
86 [Even non-exported things need system-wide Uniques because the
87 byte-code generator builds a single Name->BCO symbol table.]
89 We use the NameSupply kept in the PersistentRenamerState as the
90 source of such system-wide uniques.
92 For external Ids, use the original-name cache in the NameSupply
93 to ensure that the unique assigned is the same as the Id had
94 in any previous compilation run.
96 3. If it's an external Id, make it have a global Name and vice versa.
97 This is used by the code generator to decide whether
98 to make the label externally visible
100 4. Give external Ids a "tidy" occurrence name. This means
101 we can print them in interface files without confusing
102 "x" (unique 5) with "x" (unique 10).
104 5. Give it its UTTERLY FINAL IdInfo; in ptic,
105 * Its IdDetails becomes VanillaGlobal, reflecting the fact that
106 from now on we regard it as a global, not local, Id
108 * its unfolding, if it should have one
110 * its arity, computed from the number of visible lambdas
112 * its CAF info, computed from what is free in its RHS
115 Finally, substitute these new top-level binders consistently
116 throughout, including in unfoldings. We also tidy binders in
117 RHSs, so that they print nicely in interfaces.
120 tidyCorePgm :: DynFlags -> Module
121 -> PersistentCompilerState
122 -> CgInfoEnv -- Information from the back end,
123 -- to be splatted into the IdInfo
125 -> IO (PersistentCompilerState, ModDetails)
127 tidyCorePgm dflags mod pcs cg_info_env
128 (ModDetails { md_types = env_tc, md_insts = insts_tc,
129 md_binds = binds_in, md_rules = orphans_in })
130 = do { showPass dflags "Tidy Core"
132 ; let ext_ids = findExternalSet binds_in orphans_in
133 ; let ext_rules = findExternalRules binds_in orphans_in ext_ids
135 -- We also make sure to avoid any exported binders. Consider
136 -- f{-u1-} = 1 -- Local decl
138 -- f{-u2-} = 2 -- Exported decl
140 -- The second exported decl must 'get' the name 'f', so we
141 -- have to put 'f' in the avoids list before we get to the first
142 -- decl. tidyTopId then does a no-op on exported binders.
143 ; let prs = pcs_PRS pcs
144 orig_ns = prsOrig prs
146 init_tidy_env = (orig_ns, initTidyOccEnv avoids, emptyVarEnv)
147 avoids = [getOccName bndr | bndr <- bindersOfBinds binds_in,
148 isGlobalName (idName bndr)]
150 ; let ((orig_ns', occ_env, subst_env), tidy_binds)
151 = mapAccumL (tidyTopBind mod ext_ids cg_info_env)
152 init_tidy_env binds_in
154 ; let tidy_rules = tidyIdRules (occ_env,subst_env) ext_rules
156 ; let prs' = prs { prsOrig = orig_ns' }
157 pcs' = pcs { pcs_PRS = prs' }
159 ; let final_ids = [ id | bind <- tidy_binds
160 , id <- bindersOf bind
161 , isGlobalName (idName id)]
163 -- Dfuns are local Ids that might have
164 -- changed their unique during tidying
165 ; let lookup_dfun_id id = lookupVarEnv subst_env id `orElse`
166 pprPanic "lookup_dfun_id" (ppr id)
169 ; let tidy_type_env = mkFinalTypeEnv env_tc final_ids
170 tidy_dfun_ids = map lookup_dfun_id insts_tc
172 ; let tidy_details = ModDetails { md_types = tidy_type_env,
173 md_rules = tidy_rules,
174 md_insts = tidy_dfun_ids,
175 md_binds = tidy_binds }
177 ; endPass dflags "Tidy Core" Opt_D_dump_simpl tidy_binds
179 ; return (pcs', tidy_details)
182 tidyCoreExpr :: CoreExpr -> IO CoreExpr
183 tidyCoreExpr expr = return (tidyExpr emptyTidyEnv expr)
187 %************************************************************************
189 \subsection{Write a new interface file}
191 %************************************************************************
194 mkFinalTypeEnv :: TypeEnv -- From typechecker
198 mkFinalTypeEnv type_env final_ids
199 = extendTypeEnvList (filterNameEnv keep_it type_env)
202 -- The competed type environment is gotten from
203 -- a) keeping the types and classes
204 -- b) removing all Ids,
205 -- c) adding Ids with correct IdInfo, including unfoldings,
206 -- gotten from the bindings
207 -- From (c) we keep only those Ids with Global names;
208 -- the CoreTidy pass makes sure these are all and only
209 -- the externally-accessible ones
210 -- This truncates the type environment to include only the
211 -- exported Ids and things needed from them, which saves space
213 -- However, we do keep things like constructors, which should not appear
214 -- in interface files, because they are needed by importing modules when
215 -- using the compilation manager
217 -- We keep "hasNoBinding" Ids, notably constructor workers,
218 -- because they won't appear in the bindings from which final_ids are derived!
219 keep_it (AnId id) = hasNoBinding id -- Remove all Ids except constructor workers
220 keep_it other = True -- Keep all TyCons and Classes
224 findExternalRules :: [CoreBind]
225 -> [IdCoreRule] -- Orphan rules
226 -> IdEnv a -- Ids that are exported, so we need their rules
228 -- The complete rules are gotten by combining
229 -- a) the orphan rules
230 -- b) rules embedded in the top-level Ids
231 findExternalRules binds orphan_rules ext_ids
232 | opt_OmitInterfacePragmas = []
234 = orphan_rules ++ local_rules
236 local_rules = [ (id, rule)
237 | id <- bindersOfBinds binds,
238 id `elemVarEnv` ext_ids,
239 rule <- rulesRules (idSpecialisation id),
240 not (isBuiltinRule rule)
241 -- We can't print builtin rules in interface files
242 -- Since they are built in, an importing module
243 -- will have access to them anyway
247 %************************************************************************
249 \subsection{Step 1: finding externals}
251 %************************************************************************
254 findExternalSet :: [CoreBind] -> [IdCoreRule]
255 -> IdEnv Bool -- In domain => external
256 -- Range = True <=> show unfolding
257 -- Step 1 from the notes above
258 findExternalSet binds orphan_rules
259 = foldr find init_needed binds
261 orphan_rule_ids :: IdSet
262 orphan_rule_ids = unionVarSets [ ruleSomeFreeVars isLocalId rule
263 | (_, rule) <- orphan_rules]
264 init_needed :: IdEnv Bool
265 init_needed = mapUFM (\_ -> False) orphan_rule_ids
266 -- The mapUFM is a bit cheesy. It is a cheap way
267 -- to turn the set of orphan_rule_ids, which we use to initialise
268 -- the sweep, into a mapping saying 'don't expose unfolding'
269 -- (When we come to the binding site we may change our mind, of course.)
271 find (NonRec id rhs) needed
272 | need_id needed id = addExternal (id,rhs) needed
274 find (Rec prs) needed = find_prs prs needed
276 -- For a recursive group we have to look for a fixed point
278 | null needed_prs = needed
279 | otherwise = find_prs other_prs new_needed
281 (needed_prs, other_prs) = partition (need_pr needed) prs
282 new_needed = foldr addExternal needed needed_prs
284 -- The 'needed' set contains the Ids that are needed by earlier
285 -- interface file emissions. If the Id isn't in this set, and isn't
286 -- exported, there's no need to emit anything
287 need_id needed_set id = id `elemVarEnv` needed_set || isExportedId id
288 need_pr needed_set (id,rhs) = need_id needed_set id
290 addExternal :: (Id,CoreExpr) -> IdEnv Bool -> IdEnv Bool
291 -- The Id is needed; extend the needed set
292 -- with it and its dependents (free vars etc)
293 addExternal (id,rhs) needed
294 = extendVarEnv (foldVarSet add_occ needed new_needed_ids)
297 add_occ id needed = extendVarEnv needed id False
298 -- "False" because we don't know we need the Id's unfolding
299 -- We'll override it later when we find the binding site
301 new_needed_ids | opt_OmitInterfacePragmas = emptyVarSet
302 | otherwise = worker_ids `unionVarSet`
303 unfold_ids `unionVarSet`
307 dont_inline = isNeverInlinePrag (inlinePragInfo idinfo)
308 loop_breaker = isLoopBreaker (occInfo idinfo)
309 bottoming_fn = isBottomingStrictness (strictnessInfo idinfo)
310 spec_ids = rulesRhsFreeVars (specInfo idinfo)
311 worker_info = workerInfo idinfo
313 -- Stuff to do with the Id's unfolding
314 -- The simplifier has put an up-to-date unfolding
315 -- in the IdInfo, but the RHS will do just as well
316 unfolding = unfoldingInfo idinfo
317 rhs_is_small = not (neverUnfold unfolding)
319 -- We leave the unfolding there even if there is a worker
320 -- In GHCI the unfolding is used by importers
321 -- When writing an interface file, we omit the unfolding
322 -- if there is a worker
323 show_unfold = not bottoming_fn && -- Not necessary
326 rhs_is_small && -- Small enough
327 okToUnfoldInHiFile rhs -- No casms etc
329 unfold_ids | show_unfold = exprSomeFreeVars isLocalId rhs
330 | otherwise = emptyVarSet
332 worker_ids = case worker_info of
333 HasWorker work_id _ -> unitVarSet work_id
334 otherwise -> emptyVarSet
338 %************************************************************************
340 \subsection{Step 2: top-level tidying}
342 %************************************************************************
346 type TopTidyEnv = (NameSupply, TidyOccEnv, VarEnv Var)
348 -- TopTidyEnv: when tidying we need to know
349 -- * ns: The NameSupply, containing a unique supply and any pre-ordained Names.
350 -- These may have arisen because the
351 -- renamer read in an interface file mentioning M.$wf, say,
352 -- and assigned it unique r77. If, on this compilation, we've
353 -- invented an Id whose name is $wf (but with a different unique)
354 -- we want to rename it to have unique r77, so that we can do easy
355 -- comparisons with stuff from the interface file
357 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
360 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
365 tidyTopBind :: Module
366 -> IdEnv Bool -- Domain = Ids that should be external
367 -- True <=> their unfolding is external too
369 -> TopTidyEnv -> CoreBind
370 -> (TopTidyEnv, CoreBind)
372 tidyTopBind mod ext_ids cg_info_env top_tidy_env (NonRec bndr rhs)
373 = ((orig,occ,subst) , NonRec bndr' rhs')
375 ((orig,occ,subst), bndr')
376 = tidyTopBinder mod ext_ids cg_info_env rec_tidy_env rhs' top_tidy_env bndr
377 rec_tidy_env = (occ,subst)
378 rhs' = tidyExpr rec_tidy_env rhs
380 tidyTopBind mod ext_ids cg_info_env top_tidy_env (Rec prs)
381 = (final_env, Rec prs')
383 (final_env@(_,occ,subst), prs') = mapAccumL do_one top_tidy_env prs
384 rec_tidy_env = (occ,subst)
386 do_one top_tidy_env (bndr,rhs)
387 = ((orig,occ,subst), (bndr',rhs'))
389 ((orig,occ,subst), bndr')
390 = tidyTopBinder mod ext_ids cg_info_env
391 rec_tidy_env rhs' top_tidy_env bndr
393 rhs' = tidyExpr rec_tidy_env rhs
395 tidyTopBinder :: Module -> IdEnv Bool
397 -> TidyEnv -> CoreExpr
398 -- The TidyEnv is used to tidy the IdInfo
399 -- The expr is the already-tided RHS
400 -- Both are knot-tied: don't look at them!
401 -> TopTidyEnv -> Id -> (TopTidyEnv, Id)
402 -- NB: tidyTopBinder doesn't affect the unique supply
404 tidyTopBinder mod ext_ids cg_info_env tidy_env rhs
405 env@(ns2, occ_env2, subst_env2) id
407 | isDataConWrapId id -- Don't tidy constructor wrappers
408 = (env, id) -- The Id is stored in the TyCon, so it would be bad
409 -- if anything changed
411 -- HACK ALERT: we *do* tidy record selectors. Reason: they mention error
412 -- messages, which may be floated out:
413 -- x_field pt = case pt of
415 -- Pol _ _ -> error "buggle wuggle"
416 -- The error message will be floated out so we'll get
417 -- lvl5 = error "buggle wuggle"
418 -- x_field pt = case pt of
422 -- When this happens, it's vital that the Id exposed to importing modules
423 -- (by ghci) mentions lvl5 in its unfolding, not the un-tidied version.
425 -- What about the Id in the TyCon? It probably shouldn't be in the TyCon at
426 -- all, but in any case it will have the error message inline so it won't matter.
429 | isRecordSelector id -- We can't use the "otherwise" case, because that
430 -- forgets the IdDetails, which forgets that this is
431 -- a record selector, which confuses an importing module
432 = (env, id `setIdUnfolding` unfold_info)
435 -- This function is the heart of Step 2
436 -- The second env is the one to use for the IdInfo
437 -- It's necessary because when we are dealing with a recursive
438 -- group, a variable late in the group might be mentioned
439 -- in the IdInfo of one early in the group
441 -- The rhs is already tidied
443 = ((orig_env', occ_env', subst_env'), id')
445 (orig_env', occ_env', name') = tidyTopName mod ns2 occ_env2
448 ty' = tidyTopType (idType id)
449 cg_info = lookupCgInfo cg_info_env name'
450 idinfo' = tidyIdInfo tidy_env is_external unfold_info cg_info id
452 id' = mkVanillaGlobal name' ty' idinfo'
453 subst_env' = extendVarEnv subst_env2 id id'
455 maybe_external = lookupVarEnv ext_ids id
456 is_external = maybeToBool maybe_external
458 -- Expose an unfolding if ext_ids tells us to
459 show_unfold = maybe_external `orElse` False
460 unfold_info | show_unfold = mkTopUnfolding rhs
461 | otherwise = noUnfolding
464 tidyIdInfo tidy_env is_external unfold_info cg_info id
465 | opt_OmitInterfacePragmas || not is_external
466 -- No IdInfo if the Id isn't external, or if we don't have -O
469 `setStrictnessInfo` strictnessInfo core_idinfo
470 -- Keep strictness; it's used by CorePrep
475 `setCprInfo` cprInfo core_idinfo
476 `setStrictnessInfo` strictnessInfo core_idinfo
477 `setInlinePragInfo` inlinePragInfo core_idinfo
478 `setUnfoldingInfo` unfold_info
479 `setWorkerInfo` tidyWorker tidy_env (workerInfo core_idinfo)
480 -- NB: we throw away the Rules
481 -- They have already been extracted by findExternalRules
483 core_idinfo = idInfo id
486 -- This is where we set names to local/global based on whether they really are
487 -- externally visible (see comment at the top of this module). If the name
488 -- was previously local, we have to give it a unique occurrence name if
489 -- we intend to globalise it.
490 tidyTopName mod ns occ_env external name
491 | global && internal = (ns, occ_env, localiseName name)
493 | global && external = (ns, occ_env, name)
494 -- Global names are assumed to have been allocated by the renamer,
495 -- so they already have the "right" unique
496 -- And it's a system-wide unique too
498 | local && internal = (ns { nsUniqs = us2 }, occ_env', unique_name)
499 -- Even local, internal names must get a unique occurrence, because
500 -- if we do -split-objs we globalise the name later, in the code generator
502 -- Similarly, we must make sure it has a system-wide Unique, because
503 -- the byte-code generator builds a system-wide Name->BCO symbol table
505 | local && external = case lookupFM ns_names key of
506 Just orig -> (ns, occ_env', orig)
507 Nothing -> (ns { nsUniqs = us2, nsNames = ns_names' }, occ_env', global_name)
508 -- If we want to globalise a currently-local name, check
509 -- whether we have already assigned a unique for it.
510 -- If so, use it; if not, extend the table
513 global = isGlobalName name
515 internal = not external
517 (occ_env', occ') = tidyOccName occ_env (nameOccName name)
518 key = (moduleName mod, occ')
519 ns_names = nsNames ns
520 ns_uniqs = nsUniqs ns
521 (us1, us2) = splitUniqSupply ns_uniqs
522 unique_name = setNameUnique (setNameOcc name occ') (uniqFromSupply us1)
523 global_name = globaliseName unique_name mod
524 ns_names' = addToFM ns_names key global_name
527 ------------ Worker --------------
528 tidyWorker tidy_env (HasWorker work_id wrap_arity)
529 = HasWorker (tidyVarOcc tidy_env work_id) wrap_arity
530 tidyWorker tidy_env other
533 ------------ Rules --------------
534 tidyIdRules :: TidyEnv -> [IdCoreRule] -> [IdCoreRule]
535 tidyIdRules env [] = []
536 tidyIdRules env ((fn,rule) : rules)
537 = tidyRule env rule =: \ rule ->
538 tidyIdRules env rules =: \ rules ->
539 ((tidyVarOcc env fn, rule) : rules)
541 tidyRule :: TidyEnv -> CoreRule -> CoreRule
542 tidyRule env rule@(BuiltinRule _) = rule
543 tidyRule env (Rule name vars tpl_args rhs)
544 = tidyBndrs env vars =: \ (env', vars) ->
545 map (tidyExpr env') tpl_args =: \ tpl_args ->
546 (Rule name vars tpl_args (tidyExpr env' rhs))
549 %************************************************************************
551 \subsection{Step 2: inner tidying
553 %************************************************************************
558 -> (TidyEnv, CoreBind)
560 tidyBind env (NonRec bndr rhs)
561 = tidyBndrWithRhs env (bndr,rhs) =: \ (env', bndr') ->
562 (env', NonRec bndr' (tidyExpr env' rhs))
564 tidyBind env (Rec prs)
565 = mapAccumL tidyBndrWithRhs env prs =: \ (env', bndrs') ->
566 map (tidyExpr env') (map snd prs) =: \ rhss' ->
567 (env', Rec (zip bndrs' rhss'))
570 tidyExpr env (Var v) = Var (tidyVarOcc env v)
571 tidyExpr env (Type ty) = Type (tidyType env ty)
572 tidyExpr env (Lit lit) = Lit lit
573 tidyExpr env (App f a) = App (tidyExpr env f) (tidyExpr env a)
574 tidyExpr env (Note n e) = Note (tidyNote env n) (tidyExpr env e)
576 tidyExpr env (Let b e)
577 = tidyBind env b =: \ (env', b') ->
578 Let b' (tidyExpr env' e)
580 tidyExpr env (Case e b alts)
581 = tidyBndr env b =: \ (env', b) ->
582 Case (tidyExpr env e) b (map (tidyAlt env') alts)
584 tidyExpr env (Lam b e)
585 = tidyBndr env b =: \ (env', b) ->
586 Lam b (tidyExpr env' e)
589 tidyAlt env (con, vs, rhs)
590 = tidyBndrs env vs =: \ (env', vs) ->
591 (con, vs, tidyExpr env' rhs)
593 tidyNote env (Coerce t1 t2) = Coerce (tidyType env t1) (tidyType env t2)
594 tidyNote env note = note
598 %************************************************************************
600 \subsection{Tidying up non-top-level binders}
602 %************************************************************************
605 tidyVarOcc (_, var_env) v = case lookupVarEnv var_env v of
609 -- tidyBndr is used for lambda and case binders
610 tidyBndr :: TidyEnv -> Var -> (TidyEnv, Var)
612 | isTyVar var = tidyTyVar env var
613 | otherwise = tidyId env var
615 tidyBndrs :: TidyEnv -> [Var] -> (TidyEnv, [Var])
616 tidyBndrs env vars = mapAccumL tidyBndr env vars
618 -- tidyBndrWithRhs is used for let binders
619 tidyBndrWithRhs :: TidyEnv -> (Id, CoreExpr) -> (TidyEnv, Var)
620 tidyBndrWithRhs env (id,rhs) = tidyId env id
622 tidyId :: TidyEnv -> Id -> (TidyEnv, Id)
623 tidyId env@(tidy_env, var_env) id
624 = -- Non-top-level variables
626 -- Give the Id a fresh print-name, *and* rename its type
627 -- The SrcLoc isn't important now,
628 -- though we could extract it from the Id
630 -- All local Ids now have the same IdInfo, which should save some
632 (tidy_env', occ') = tidyOccName tidy_env (getOccName id)
633 ty' = tidyType (tidy_env,var_env) (idType id)
634 id' = mkUserLocal occ' (idUnique id) ty' noSrcLoc
635 var_env' = extendVarEnv var_env id id'
637 ((tidy_env', var_env'), id')