2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section{Tidying up Core}
7 module TidyPgm( tidyCorePgm, tidyCoreExpr ) where
9 #include "HsVersions.h"
11 import CmdLineOpts ( DynFlags, DynFlag(..), opt_OmitInterfacePragmas )
13 import CoreUnfold ( noUnfolding, mkTopUnfolding, okToUnfoldInHiFile )
14 import CoreFVs ( ruleLhsFreeIds, ruleRhsFreeVars, exprSomeFreeVars )
15 import CoreTidy ( tidyExpr, tidyVarOcc, tidyIdRules )
16 import PprCore ( pprIdRules )
17 import CoreLint ( showPass, endPass )
18 import CoreUtils ( exprArity )
21 import Var ( Id, Var )
22 import Id ( idType, idInfo, idName, idCoreRules,
23 isExportedId, idUnique, mkVanillaGlobal, isLocalId,
24 isImplicitId, mkUserLocal, setIdInfo
26 import IdInfo {- loads of stuff -}
27 import NewDemand ( isBottomingSig, topSig )
28 import BasicTypes ( isNeverActive )
29 import Name ( getOccName, nameOccName, mkInternalName, mkExternalName,
30 localiseName, isExternalName, nameSrcLoc
32 import NameEnv ( filterNameEnv )
33 import OccName ( TidyOccEnv, initTidyOccEnv, tidyOccName )
34 import Type ( tidyTopType )
35 import Module ( Module, moduleName )
36 import HscTypes ( PersistentCompilerState( pcs_PRS ),
37 PersistentRenamerState( prsOrig ),
38 NameSupply( nsNames, nsUniqs ),
39 TypeEnv, extendTypeEnvList, typeEnvIds,
40 ModDetails(..), TyThing(..)
42 import FiniteMap ( lookupFM, addToFM )
43 import Maybes ( orElse )
44 import ErrUtils ( showPass, dumpIfSet_core )
45 import SrcLoc ( noSrcLoc )
46 import UniqFM ( mapUFM )
47 import UniqSupply ( splitUniqSupply, uniqFromSupply )
48 import List ( partition )
49 import Util ( mapAccumL )
50 import Maybe ( isJust )
56 %************************************************************************
58 \subsection{What goes on}
60 %************************************************************************
66 Step 1: Figure out external Ids
67 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
68 First we figure out which Ids are "external" Ids. An
69 "external" Id is one that is visible from outside the compilation
71 a) the user exported ones
72 b) ones mentioned in the unfoldings, workers,
73 or rules of externally-visible ones
74 This exercise takes a sweep of the bindings bottom to top. Actually,
75 in Step 2 we're also going to need to know which Ids should be
76 exported with their unfoldings, so we produce not an IdSet but an
80 Step 2: Tidy the program
81 ~~~~~~~~~~~~~~~~~~~~~~~~
82 Next we traverse the bindings top to bottom. For each *top-level*
85 1. Make it into a GlobalId
87 2. Give it a system-wide Unique.
88 [Even non-exported things need system-wide Uniques because the
89 byte-code generator builds a single Name->BCO symbol table.]
91 We use the NameSupply kept in the PersistentRenamerState as the
92 source of such system-wide uniques.
94 For external Ids, use the original-name cache in the NameSupply
95 to ensure that the unique assigned is the same as the Id had
96 in any previous compilation run.
98 3. If it's an external Id, make it have a global Name, otherwise
99 make it have a local Name.
100 This is used by the code generator to decide whether
101 to make the label externally visible
103 4. Give external Ids a "tidy" occurrence name. This means
104 we can print them in interface files without confusing
105 "x" (unique 5) with "x" (unique 10).
107 5. Give it its UTTERLY FINAL IdInfo; in ptic,
108 * Its IdDetails becomes VanillaGlobal, reflecting the fact that
109 from now on we regard it as a global, not local, Id
111 * its unfolding, if it should have one
113 * its arity, computed from the number of visible lambdas
115 * its CAF info, computed from what is free in its RHS
118 Finally, substitute these new top-level binders consistently
119 throughout, including in unfoldings. We also tidy binders in
120 RHSs, so that they print nicely in interfaces.
123 tidyCorePgm :: DynFlags -> Module
124 -> PersistentCompilerState
125 -> CgInfoEnv -- Information from the back end,
126 -- to be splatted into the IdInfo
128 -> IO (PersistentCompilerState, ModDetails)
130 tidyCorePgm dflags mod pcs cg_info_env
131 (ModDetails { md_types = env_tc, md_insts = insts_tc,
132 md_binds = binds_in, md_rules = orphans_in })
133 = do { showPass dflags "Tidy Core"
135 ; let ext_ids = findExternalSet binds_in orphans_in
136 ; let ext_rules = findExternalRules binds_in orphans_in ext_ids
137 -- findExternalRules filters ext_rules to avoid binders that
138 -- aren't externally visible; but the externally-visible binders
139 -- are computed (by findExternalSet) assuming that all orphan
140 -- rules are exported. So in fact we may export more than we
141 -- need. (It's a sort of mutual recursion.)
143 -- We also make sure to avoid any exported binders. Consider
144 -- f{-u1-} = 1 -- Local decl
146 -- f{-u2-} = 2 -- Exported decl
148 -- The second exported decl must 'get' the name 'f', so we
149 -- have to put 'f' in the avoids list before we get to the first
150 -- decl. tidyTopId then does a no-op on exported binders.
151 ; let prs = pcs_PRS pcs
152 orig_ns = prsOrig prs
154 init_tidy_env = (orig_ns, initTidyOccEnv avoids, emptyVarEnv)
155 avoids = [getOccName name | bndr <- typeEnvIds env_tc,
156 let name = idName bndr,
158 -- In computing our "avoids" list, we must include
160 -- all things with global names (assigned once and for
161 -- all by the renamer)
162 -- since their names are "taken".
163 -- The type environment is a convenient source of such things.
165 ; let ((orig_ns', occ_env, subst_env), tidy_binds)
166 = mapAccumL (tidyTopBind mod ext_ids cg_info_env)
167 init_tidy_env binds_in
169 ; let tidy_rules = tidyIdRules (occ_env,subst_env) ext_rules
171 ; let prs' = prs { prsOrig = orig_ns' }
172 pcs' = pcs { pcs_PRS = prs' }
174 ; let final_ids = [ id
176 , id <- bindersOf bind
177 , isExternalName (idName id)]
179 -- Dfuns are local Ids that might have
180 -- changed their unique during tidying
181 ; let lookup_dfun_id id = lookupVarEnv subst_env id `orElse`
182 pprPanic "lookup_dfun_id" (ppr id)
185 ; let tidy_type_env = mkFinalTypeEnv env_tc final_ids
186 tidy_dfun_ids = map lookup_dfun_id insts_tc
188 ; let tidy_details = ModDetails { md_types = tidy_type_env,
189 md_rules = tidy_rules,
190 md_insts = tidy_dfun_ids,
191 md_binds = tidy_binds }
193 ; endPass dflags "Tidy Core" Opt_D_dump_simpl tidy_binds
194 ; dumpIfSet_core dflags Opt_D_dump_simpl
196 (pprIdRules tidy_rules)
198 ; return (pcs', tidy_details)
201 tidyCoreExpr :: CoreExpr -> IO CoreExpr
202 tidyCoreExpr expr = return (tidyExpr emptyTidyEnv expr)
206 %************************************************************************
208 \subsection{Write a new interface file}
210 %************************************************************************
213 mkFinalTypeEnv :: TypeEnv -- From typechecker
217 mkFinalTypeEnv type_env final_ids
218 = extendTypeEnvList (filterNameEnv keep_it type_env)
221 -- The competed type environment is gotten from
222 -- a) keeping the types and classes
223 -- b) removing all Ids,
224 -- c) adding Ids with correct IdInfo, including unfoldings,
225 -- gotten from the bindings
226 -- From (c) we keep only those Ids with Global names;
227 -- the CoreTidy pass makes sure these are all and only
228 -- the externally-accessible ones
229 -- This truncates the type environment to include only the
230 -- exported Ids and things needed from them, which saves space
232 -- However, we do keep things like constructors, which should not appear
233 -- in interface files, because they are needed by importing modules when
234 -- using the compilation manager
236 -- We keep implicit Ids, because they won't appear
237 -- in the bindings from which final_ids are derived!
238 keep_it (AnId id) = isImplicitId id -- Remove all Ids except implicit ones
239 keep_it other = True -- Keep all TyCons and Classes
243 findExternalRules :: [CoreBind]
244 -> [IdCoreRule] -- Orphan rules
245 -> IdEnv a -- Ids that are exported, so we need their rules
247 -- The complete rules are gotten by combining
248 -- a) the orphan rules
249 -- b) rules embedded in the top-level Ids
250 findExternalRules binds orphan_rules ext_ids
251 | opt_OmitInterfacePragmas = []
253 = filter needed_rule (orphan_rules ++ local_rules)
256 | id <- bindersOfBinds binds,
257 id `elemVarEnv` ext_ids,
258 rule <- idCoreRules id
260 needed_rule (id, rule)
261 = not (isBuiltinRule rule)
262 -- We can't print builtin rules in interface files
263 -- Since they are built in, an importing module
264 -- will have access to them anyway
266 && not (any internal_id (varSetElems (ruleLhsFreeIds rule)))
267 -- Don't export a rule whose LHS mentions an Id that
268 -- is completely internal (i.e. not visible to an
271 internal_id id = isLocalId id && not (id `elemVarEnv` ext_ids)
274 %************************************************************************
276 \subsection{Step 1: finding externals}
278 %************************************************************************
281 findExternalSet :: [CoreBind] -> [IdCoreRule]
282 -> IdEnv Bool -- In domain => external
283 -- Range = True <=> show unfolding
284 -- Step 1 from the notes above
285 findExternalSet binds orphan_rules
286 = foldr find init_needed binds
288 orphan_rule_ids :: IdSet
289 orphan_rule_ids = unionVarSets [ ruleRhsFreeVars rule
290 | (_, rule) <- orphan_rules]
291 init_needed :: IdEnv Bool
292 init_needed = mapUFM (\_ -> False) orphan_rule_ids
293 -- The mapUFM is a bit cheesy. It is a cheap way
294 -- to turn the set of orphan_rule_ids, which we use to initialise
295 -- the sweep, into a mapping saying 'don't expose unfolding'
296 -- (When we come to the binding site we may change our mind, of course.)
298 find (NonRec id rhs) needed
299 | need_id needed id = addExternal (id,rhs) needed
301 find (Rec prs) needed = find_prs prs needed
303 -- For a recursive group we have to look for a fixed point
305 | null needed_prs = needed
306 | otherwise = find_prs other_prs new_needed
308 (needed_prs, other_prs) = partition (need_pr needed) prs
309 new_needed = foldr addExternal needed needed_prs
311 -- The 'needed' set contains the Ids that are needed by earlier
312 -- interface file emissions. If the Id isn't in this set, and isn't
313 -- exported, there's no need to emit anything
314 need_id needed_set id = id `elemVarEnv` needed_set || isExportedId id
315 need_pr needed_set (id,rhs) = need_id needed_set id
317 addExternal :: (Id,CoreExpr) -> IdEnv Bool -> IdEnv Bool
318 -- The Id is needed; extend the needed set
319 -- with it and its dependents (free vars etc)
320 addExternal (id,rhs) needed
321 = extendVarEnv (foldVarSet add_occ needed new_needed_ids)
324 add_occ id needed = extendVarEnv needed id False
325 -- "False" because we don't know we need the Id's unfolding
326 -- We'll override it later when we find the binding site
328 new_needed_ids | opt_OmitInterfacePragmas = emptyVarSet
329 | otherwise = worker_ids `unionVarSet`
330 unfold_ids `unionVarSet`
334 dont_inline = isNeverActive (inlinePragInfo idinfo)
335 loop_breaker = isLoopBreaker (occInfo idinfo)
336 bottoming_fn = isBottomingSig (newStrictnessInfo idinfo `orElse` topSig)
337 spec_ids = rulesRhsFreeVars (specInfo idinfo)
338 worker_info = workerInfo idinfo
340 -- Stuff to do with the Id's unfolding
341 -- The simplifier has put an up-to-date unfolding
342 -- in the IdInfo, but the RHS will do just as well
343 unfolding = unfoldingInfo idinfo
344 rhs_is_small = not (neverUnfold unfolding)
346 -- We leave the unfolding there even if there is a worker
347 -- In GHCI the unfolding is used by importers
348 -- When writing an interface file, we omit the unfolding
349 -- if there is a worker
350 show_unfold = not bottoming_fn && -- Not necessary
353 rhs_is_small && -- Small enough
354 okToUnfoldInHiFile rhs -- No casms etc
356 unfold_ids | show_unfold = exprSomeFreeVars isLocalId rhs
357 | otherwise = emptyVarSet
359 worker_ids = case worker_info of
360 HasWorker work_id _ -> unitVarSet work_id
361 otherwise -> emptyVarSet
365 %************************************************************************
367 \subsection{Step 2: top-level tidying}
369 %************************************************************************
373 type TopTidyEnv = (NameSupply, TidyOccEnv, VarEnv Var)
375 -- TopTidyEnv: when tidying we need to know
376 -- * ns: The NameSupply, containing a unique supply and any pre-ordained Names.
377 -- These may have arisen because the
378 -- renamer read in an interface file mentioning M.$wf, say,
379 -- and assigned it unique r77. If, on this compilation, we've
380 -- invented an Id whose name is $wf (but with a different unique)
381 -- we want to rename it to have unique r77, so that we can do easy
382 -- comparisons with stuff from the interface file
384 -- * occ_env: The TidyOccEnv, which tells us which local occurrences
387 -- * subst_env: A Var->Var mapping that substitutes the new Var for the old
392 tidyTopBind :: Module
393 -> IdEnv Bool -- Domain = Ids that should be external
394 -- True <=> their unfolding is external too
396 -> TopTidyEnv -> CoreBind
397 -> (TopTidyEnv, CoreBind)
399 tidyTopBind mod ext_ids cg_info_env top_tidy_env (NonRec bndr rhs)
400 = ((orig,occ,subst) , NonRec bndr' rhs')
402 ((orig,occ,subst), bndr')
403 = tidyTopBinder mod ext_ids cg_info_env
404 rec_tidy_env rhs rhs' top_tidy_env bndr
405 rec_tidy_env = (occ,subst)
406 rhs' = tidyExpr rec_tidy_env rhs
408 tidyTopBind mod ext_ids cg_info_env top_tidy_env (Rec prs)
409 = (final_env, Rec prs')
411 (final_env@(_,occ,subst), prs') = mapAccumL do_one top_tidy_env prs
412 rec_tidy_env = (occ,subst)
414 do_one top_tidy_env (bndr,rhs)
415 = ((orig,occ,subst), (bndr',rhs'))
417 ((orig,occ,subst), bndr')
418 = tidyTopBinder mod ext_ids cg_info_env
419 rec_tidy_env rhs rhs' top_tidy_env bndr
421 rhs' = tidyExpr rec_tidy_env rhs
423 tidyTopBinder :: Module -> IdEnv Bool -> CgInfoEnv
424 -> TidyEnv -- The TidyEnv is used to tidy the IdInfo
425 -> CoreExpr -- RHS *before* tidying
426 -> CoreExpr -- RHS *after* tidying
427 -- The TidyEnv and the after-tidying RHS are
428 -- both are knot-tied: don't look at them!
429 -> TopTidyEnv -> Id -> (TopTidyEnv, Id)
430 -- NB: tidyTopBinder doesn't affect the unique supply
432 tidyTopBinder mod ext_ids cg_info_env rec_tidy_env rhs tidy_rhs
433 env@(ns2, occ_env2, subst_env2) id
434 -- This function is the heart of Step 2
435 -- The rec_tidy_env is the one to use for the IdInfo
436 -- It's necessary because when we are dealing with a recursive
437 -- group, a variable late in the group might be mentioned
438 -- in the IdInfo of one early in the group
440 -- The rhs is already tidied
442 = ((orig_env', occ_env', subst_env'), id')
444 (orig_env', occ_env', name') = tidyTopName mod ns2 occ_env2
447 ty' = tidyTopType (idType id)
448 idinfo = tidyTopIdInfo rec_tidy_env is_external
449 (idInfo id) unfold_info arity
450 (lookupCgInfo cg_info_env name')
452 id' = mkVanillaGlobal name' ty' idinfo
454 subst_env' = extendVarEnv subst_env2 id id'
456 maybe_external = lookupVarEnv ext_ids id
457 is_external = isJust maybe_external
459 -- Expose an unfolding if ext_ids tells us to
460 -- Remember that ext_ids maps an Id to a Bool:
461 -- True to show the unfolding, False to hide it
462 show_unfold = maybe_external `orElse` False
463 unfold_info | show_unfold = mkTopUnfolding tidy_rhs
464 | otherwise = noUnfolding
466 -- Usually the Id will have an accurate arity on it, because
467 -- the simplifier has just run, but not always.
468 -- One case I found was when the last thing the simplifier
469 -- did was to let-bind a non-atomic argument and then float
470 -- it to the top level. So it seems more robust just to
472 arity = exprArity rhs
476 -- tidyTopIdInfo creates the final IdInfo for top-level
477 -- binders. There are two delicate pieces:
479 -- * Arity. After CoreTidy, this arity must not change any more.
480 -- Indeed, CorePrep must eta expand where necessary to make
481 -- the manifest arity equal to the claimed arity.
483 -- * CAF info, which comes from the CoreToStg pass via a knot.
484 -- The CAF info will not be looked at by the downstream stuff:
485 -- it *generates* it, and knot-ties it back. It will only be
486 -- looked at by (a) MkIface when generating an interface file
487 -- (b) In GHCi, importing modules
488 -- Nevertheless, we add the info here so that it propagates to all
489 -- occurrences of the binders in RHSs, and hence to occurrences in
490 -- unfoldings, which are inside Ids imported by GHCi. Ditto RULES.
492 -- An alterative would be to do a second pass over the unfoldings
493 -- of Ids, and rules, right at the top, but that would be a pain.
495 tidyTopIdInfo tidy_env is_external idinfo unfold_info arity cg_info
496 | opt_OmitInterfacePragmas || not is_external
497 -- Only basic info if the Id isn't external, or if we don't have -O
500 | otherwise -- Add extra optimisation info
502 `setInlinePragInfo` inlinePragInfo idinfo
503 `setUnfoldingInfo` unfold_info
504 `setWorkerInfo` tidyWorker tidy_env (workerInfo idinfo)
505 -- NB: we throw away the Rules
506 -- They have already been extracted by findExternalRules
509 -- baasic_info is attached to every top-level binder
510 basic_info = vanillaIdInfo
513 `setAllStrictnessInfo` newStrictnessInfo idinfo
515 -- This is where we set names to local/global based on whether they really are
516 -- externally visible (see comment at the top of this module). If the name
517 -- was previously local, we have to give it a unique occurrence name if
518 -- we intend to externalise it.
519 tidyTopName mod ns occ_env external name
520 | global && internal = (ns, occ_env, localiseName name)
522 | global && external = (ns, occ_env, name)
523 -- Global names are assumed to have been allocated by the renamer,
524 -- so they already have the "right" unique
525 -- And it's a system-wide unique too
527 | local && internal = (ns_w_local, occ_env', new_local_name)
528 -- Even local, internal names must get a unique occurrence, because
529 -- if we do -split-objs we externalise the name later, in the code generator
531 -- Similarly, we must make sure it has a system-wide Unique, because
532 -- the byte-code generator builds a system-wide Name->BCO symbol table
534 | local && external = case lookupFM ns_names key of
535 Just orig -> (ns, occ_env', orig)
536 Nothing -> (ns_w_global, occ_env', new_external_name)
537 -- If we want to externalise a currently-local name, check
538 -- whether we have already assigned a unique for it.
539 -- If so, use it; if not, extend the table (ns_w_global).
540 -- This is needed when *re*-compiling a module in GHCi; we want to
541 -- use the same name for externally-visible things as we did before.
544 global = isExternalName name
546 internal = not external
548 (occ_env', occ') = tidyOccName occ_env (nameOccName name)
549 key = (moduleName mod, occ')
550 ns_names = nsNames ns
551 ns_uniqs = nsUniqs ns
552 (us1, us2) = splitUniqSupply ns_uniqs
553 uniq = uniqFromSupply us1
554 loc = nameSrcLoc name
556 new_local_name = mkInternalName uniq occ' loc
557 new_external_name = mkExternalName uniq mod occ' loc
559 ns_w_local = ns { nsUniqs = us2 }
560 ns_w_global = ns { nsUniqs = us2, nsNames = addToFM ns_names key new_external_name }
563 ------------ Worker --------------
564 tidyWorker tidy_env (HasWorker work_id wrap_arity)
565 = HasWorker (tidyVarOcc tidy_env work_id) wrap_arity
566 tidyWorker tidy_env other