Tidy.hs 62.4 KB
Newer Older
Austin Seipp's avatar
Austin Seipp committed
1 2
{-
(c) The GRASP/AQUA Project, Glasgow University, 1992-1998
Simon Marlow's avatar
Simon Marlow committed
3

4
\section{Tidying up Core}
Austin Seipp's avatar
Austin Seipp committed
5
-}
6

7
{-# LANGUAGE CPP, DeriveFunctor, ViewPatterns #-}
8

9 10
{-# OPTIONS_GHC -Wno-incomplete-uni-patterns #-}

11
module GHC.Iface.Tidy (
12
       mkBootModDetailsTc, tidyProgram
13
   ) where
14 15 16

#include "HsVersions.h"

17 18
import GhcPrelude

Simon Marlow's avatar
Simon Marlow committed
19
import TcRnTypes
Simon Marlow's avatar
Simon Marlow committed
20
import DynFlags
21
import CoreSyn
Simon Marlow's avatar
Simon Marlow committed
22 23 24
import CoreUnfold
import CoreFVs
import CoreTidy
25
import CoreMonad
Sylvain Henry's avatar
Sylvain Henry committed
26
import GHC.CoreToStg.Prep
27
import CoreUtils        (rhsIsStatic)
28
import CoreStats        (coreBindsStats, CoreStats(..))
29
import CoreSeq          (seqBinds)
30
import CoreLint
31
import Literal
32
import Rules
33 34
import PatSyn
import ConLike
35
import CoreArity        ( exprArity, exprBotStrictness_maybe )
36
import StaticPtrTable
37 38
import VarEnv
import VarSet
39
import Var
Simon Marlow's avatar
Simon Marlow committed
40
import Id
41
import MkId             ( mkDictSelRhs )
Simon Marlow's avatar
Simon Marlow committed
42 43
import IdInfo
import InstEnv
Simon Peyton Jones's avatar
Simon Peyton Jones committed
44
import Type             ( tidyTopType )
Joachim Breitner's avatar
Joachim Breitner committed
45
import Demand           ( appIsBottom, isTopSig, isBottomingSig )
Simon Marlow's avatar
Simon Marlow committed
46
import BasicTypes
47
import Name hiding (varName)
Simon Marlow's avatar
Simon Marlow committed
48
import NameSet
49
import NameCache
50
import Avail
51
import GHC.Iface.Env
52 53
import TcEnv
import TcRnMonad
Simon Marlow's avatar
Simon Marlow committed
54 55
import DataCon
import TyCon
56
import Class
57
import Module
58
import Packages( isDllName )
59
import HscTypes
Simon Marlow's avatar
Simon Marlow committed
60 61
import Maybes
import UniqSupply
62
import Outputable
63
import Util( filterOut )
64
import qualified ErrUtils as Err
Simon Marlow's avatar
Simon Marlow committed
65

66
import Control.Monad
Ian Lynagh's avatar
Ian Lynagh committed
67
import Data.Function
68
import Data.List        ( sortBy, mapAccumL )
69
import Data.IORef       ( atomicModifyIORef' )
70

Austin Seipp's avatar
Austin Seipp committed
71
{-
72
Constructing the TypeEnv, Instances, Rules from which the
73 74
ModIface is constructed, and which goes on to subsequent modules in
--make mode.
75 76 77 78 79 80 81 82

Most of the interface file is obtained simply by serialising the
TypeEnv.  One important consequence is that if the *interface file*
has pragma info if and only if the final TypeEnv does. This is not so
important for *this* module, but it's essential for ghc --make:
subsequent compilations must not see (e.g.) the arity if the interface
file does not contain arity If they do, they'll exploit the arity;
then the arity might change, but the iface file doesn't change =>
83
recompilation does not happen => disaster.
84 85 86 87 88 89

For data types, the final TypeEnv will have a TyThing for the TyCon,
plus one for each DataCon; the interface file will contain just one
data type declaration, but it is de-serialised back into a collection
of TyThings.

Austin Seipp's avatar
Austin Seipp committed
90 91
************************************************************************
*                                                                      *
92
                Plan A: simpleTidyPgm
Austin Seipp's avatar
Austin Seipp committed
93 94
*                                                                      *
************************************************************************
95 96


97
Plan A: mkBootModDetails: omit pragmas, make interfaces small
98 99 100
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* Ignore the bindings

101 102
* Drop all WiredIn things from the TypeEnv
        (we never want them in interface files)
103 104

* Retain all TyCons and Classes in the TypeEnv, to avoid
105 106
        having to find which ones are mentioned in the
        types of exported Ids
107 108

* Trim off the constructors of non-exported TyCons, both
109
        from the TyCon and from the TypeEnv
110 111 112

* Drop non-exported Ids from the TypeEnv

113
* Tidy the types of the DFunIds of Instances,
114 115 116 117 118 119 120 121
  make them into GlobalIds, (they already have External Names)
  and add them to the TypeEnv

* Tidy the types of the (exported) Ids in the TypeEnv,
  make them into GlobalIds (they already have External Names)

* Drop rules altogether

122
* Tidy the bindings, to ensure that the Caf and Arity
123
  information is correct for each top-level binder; the
124 125
  code generator needs it. And to ensure that local names have
  distinct OccNames in case of object-file splitting
126

127 128
* If this an hsig file, drop the instances altogether too (they'll
  get pulled in by the implicit module import.
Austin Seipp's avatar
Austin Seipp committed
129
-}
130

131 132
-- This is Plan A: make a small type env when typechecking only,
-- or when compiling a hs-boot file, or simply when not using -O
133 134 135
--
-- We don't look at the bindings at all -- there aren't any
-- for hs-boot files
136

Simon Marlow's avatar
Simon Marlow committed
137
mkBootModDetailsTc :: HscEnv -> TcGblEnv -> IO ModDetails
138
mkBootModDetailsTc hsc_env
139 140 141 142 143 144 145 146
        TcGblEnv{ tcg_exports          = exports,
                  tcg_type_env         = type_env, -- just for the Ids
                  tcg_tcs              = tcs,
                  tcg_patsyns          = pat_syns,
                  tcg_insts            = insts,
                  tcg_fam_insts        = fam_insts,
                  tcg_complete_matches = complete_sigs,
                  tcg_mod              = this_mod
Simon Marlow's avatar
Simon Marlow committed
147
                }
148 149
  = -- This timing isn't terribly useful since the result isn't forced, but
    -- the message is useful to locating oneself in the compilation process.
150
    Err.withTiming dflags
151 152
                   (text "CoreTidy"<+>brackets (ppr this_mod))
                   (const ()) $
153 154 155 156 157 158 159 160
    return (ModDetails { md_types         = type_env'
                       , md_insts         = insts'
                       , md_fam_insts     = fam_insts
                       , md_rules         = []
                       , md_anns          = []
                       , md_exports       = exports
                       , md_complete_sigs = complete_sigs
                       })
161
  where
162
    dflags = hsc_dflags hsc_env
163

164 165 166 167 168 169 170 171 172 173 174 175 176
    -- Find the LocalIds in the type env that are exported
    -- Make them into GlobalIds, and tidy their types
    --
    -- It's very important to remove the non-exported ones
    -- because we don't tidy the OccNames, and if we don't remove
    -- the non-exported ones we'll get many things with the
    -- same name in the interface file, giving chaos.
    --
    -- Do make sure that we keep Ids that are already Global.
    -- When typechecking an .hs-boot file, the Ids come through as
    -- GlobalIds.
    final_ids = [ globaliseAndTidyBootId id
                | id <- typeEnvIds type_env
177
                , keep_it id ]
178

179
    final_tcs  = filterOut isWiredIn tcs
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
                 -- See Note [Drop wired-in things]
    type_env1  = typeEnvFromEntities final_ids final_tcs fam_insts
    insts'     = mkFinalClsInsts type_env1 insts
    pat_syns'  = mkFinalPatSyns  type_env1 pat_syns
    type_env'  = extendTypeEnvWithPatSyns pat_syns' type_env1

    -- Default methods have their export flag set (isExportedId),
    -- but everything else doesn't (yet), because this is
    -- pre-desugaring, so we must test against the exports too.
    keep_it id | isWiredInName id_name           = False
                 -- See Note [Drop wired-in things]
               | isExportedId id                 = True
               | id_name `elemNameSet` exp_names = True
               | otherwise                       = False
               where
                 id_name = idName id

    exp_names = availsToNameSet exports

lookupFinalId :: TypeEnv -> Id -> Id
lookupFinalId type_env id
  = case lookupTypeEnv type_env (idName id) of
      Just (AnId id') -> id'
      _ -> pprPanic "lookup_final_id" (ppr id)

mkFinalClsInsts :: TypeEnv -> [ClsInst] -> [ClsInst]
mkFinalClsInsts env = map (updateClsInstDFun (lookupFinalId env))

mkFinalPatSyns :: TypeEnv -> [PatSyn] -> [PatSyn]
mkFinalPatSyns env = map (updatePatSynIds (lookupFinalId env))
210

211 212 213
extendTypeEnvWithPatSyns :: [PatSyn] -> TypeEnv -> TypeEnv
extendTypeEnvWithPatSyns tidy_patsyns type_env
  = extendTypeEnvList type_env [AConLike (PatSynCon ps) | ps <- tidy_patsyns ]
214

215 216
globaliseAndTidyBootId :: Id -> Id
-- For a LocalId with an External Name,
217
-- makes it into a GlobalId
218 219
--     * unchanged Name (might be Internal or External)
--     * unchanged details
220
--     * VanillaIdInfo (makes a conservative assumption about Caf-hood and arity)
221
--     * BootUnfolding (see Note [Inlining and hs-boot files] in GHC.CoreToIface)
222 223 224
globaliseAndTidyBootId id
  = globaliseId id `setIdType`      tidyTopType (idType id)
                   `setIdUnfolding` BootUnfolding
225

Austin Seipp's avatar
Austin Seipp committed
226 227 228
{-
************************************************************************
*                                                                      *
229
        Plan B: tidy bindings, make TypeEnv full of IdInfo
Austin Seipp's avatar
Austin Seipp committed
230 231
*                                                                      *
************************************************************************
232

233
Plan B: include pragmas, make interfaces
234
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
235 236
* Step 1: Figure out which Ids are externally visible
          See Note [Choosing external Ids]
237

Gabor Greif's avatar
Gabor Greif committed
238
* Step 2: Gather the externally visible rules, separately from
239 240 241 242 243
          the top-level bindings.
          See Note [Finding external rules]

* Step 3: Tidy the bindings, externalising appropriate Ids
          See Note [Tidy the top-level bindings]
244

245
* Drop all Ids from the TypeEnv, and add all the External Ids from
246 247
  the bindings.  (This adds their IdInfo to the TypeEnv; and adds
  floated-out Ids that weren't even in the TypeEnv before.)
248

249 250
Note [Choosing external Ids]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Simon Marlow's avatar
Simon Marlow committed
251
See also the section "Interface stability" in the
252 253
recompilation-avoidance commentary:
  https://gitlab.haskell.org/ghc/ghc/wikis/commentary/compiler/recompilation-avoidance
Simon Marlow's avatar
Simon Marlow committed
254

255 256 257
First we figure out which Ids are "external" Ids.  An
"external" Id is one that is visible from outside the compilation
unit.  These are
258
  a) the user exported ones
259
  b) the ones bound to static forms
260 261
  c) ones mentioned in the unfoldings, workers, or
     rules of externally-visible ones
262

263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
While figuring out which Ids are external, we pick a "tidy" OccName
for each one.  That is, we make its OccName distinct from the other
external OccNames in this module, so that in interface files and
object code we can refer to it unambiguously by its OccName.  The
OccName for each binder is prefixed by the name of the exported Id
that references it; e.g. if "f" references "x" in its unfolding, then
"x" is renamed to "f_x".  This helps distinguish the different "x"s
from each other, and means that if "f" is later removed, things that
depend on the other "x"s will not need to be recompiled.  Of course,
if there are multiple "f_x"s, then we have to disambiguate somehow; we
use "f_x0", "f_x1" etc.

As far as possible we should assign names in a deterministic fashion.
Each time this module is compiled with the same options, we should end
up with the same set of external names with the same types.  That is,
the ABI hash in the interface should not change.  This turns out to be
quite tricky, since the order of the bindings going into the tidy
phase is already non-deterministic, as it is based on the ordering of
Uniques, which are assigned unpredictably.

To name things in a stable way, we do a depth-first-search of the
bindings, starting from the exports sorted by name.  This way, as long
as the bindings themselves are deterministic (they sometimes aren't!),
the order in which they are presented to the tidying phase does not
affect the names we assign.
288

289 290
Note [Tidy the top-level bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
291 292 293
Next we traverse the bindings top to bottom.  For each *top-level*
binder

294 295
 1. Make it into a GlobalId; its IdDetails becomes VanillaGlobal,
    reflecting the fact that from now on we regard it as a global,
296
    not local, Id
297 298 299 300 301

 2. Give it a system-wide Unique.
    [Even non-exported things need system-wide Uniques because the
    byte-code generator builds a single Name->BCO symbol table.]

302
    We use the NameCache kept in the HscEnv as the
303 304
    source of such system-wide uniques.

305
    For external Ids, use the original-name cache in the NameCache
306
    to ensure that the unique assigned is the same as the Id had
307
    in any previous compilation run.
308 309 310 311 312 313

 3. Rename top-level Ids according to the names we chose in step 1.
    If it's an external Id, make it have a External Name, otherwise
    make it have an Internal Name.  This is used by the code generator
    to decide whether to make the label externally visible

314 315 316 317 318 319
 4. Give it its UTTERLY FINAL IdInfo; in ptic,
        * its unfolding, if it should have one

        * its arity, computed from the number of visible lambdas

        * its CAF info, computed from what is free in its RHS
320 321 322 323 324


Finally, substitute these new top-level binders consistently
throughout, including in unfoldings.  We also tidy binders in
RHSs, so that they print nicely in interfaces.
Austin Seipp's avatar
Austin Seipp committed
325
-}
326

327
tidyProgram :: HscEnv -> ModGuts -> IO (CgGuts, ModDetails)
328 329
tidyProgram hsc_env  (ModGuts { mg_module    = mod
                              , mg_exports   = exports
330
                              , mg_rdr_env   = rdr_env
331
                              , mg_tcs       = tcs
332
                              , mg_insts     = cls_insts
333 334
                              , mg_fam_insts = fam_insts
                              , mg_binds     = binds
cactus's avatar
cactus committed
335
                              , mg_patsyns   = patsyns
336 337
                              , mg_rules     = imp_rules
                              , mg_anns      = anns
338
                              , mg_complete_sigs = complete_sigs
339
                              , mg_deps      = deps
340
                              , mg_foreign   = foreign_stubs
341
                              , mg_foreign_files = foreign_files
342
                              , mg_hpc_info  = hpc_info
343
                              , mg_modBreaks = modBreaks
344 345
                              })

346
  = Err.withTiming dflags
347 348 349
                   (text "CoreTidy"<+>brackets (ppr mod))
                   (const ()) $
    do  { let { omit_prags = gopt Opt_OmitInterfacePragmas dflags
ian@well-typed.com's avatar
ian@well-typed.com committed
350
              ; expose_all = gopt Opt_ExposeAllUnfoldings  dflags
351
              ; print_unqual = mkPrintUnqualified dflags rdr_env
352
              ; implicit_binds = concatMap getImplicitBinds tcs
353
              }
354

355
        ; (unfold_env, tidy_occ_env)
356
              <- chooseExternalIds hsc_env mod omit_prags expose_all
357
                                   binds implicit_binds imp_rules
Austin Seipp's avatar
Austin Seipp committed
358
        ; let { (trimmed_binds, trimmed_rules)
359
                    = findExternalRules omit_prags binds imp_rules unfold_env }
360

361
        ; (tidy_env, tidy_binds)
362
                 <- tidyTopBinds hsc_env mod unfold_env tidy_occ_env trimmed_binds
363

364
          -- See Note [Grand plan for static forms] in StaticPtrTable.
365
        ; (spt_entries, tidy_binds') <-
366
             sptCreateStaticBinds hsc_env mod tidy_binds
367 368 369 370 371 372 373 374 375
        ; let { spt_init_code = sptModuleInitCode mod spt_entries
              ; add_spt_init_code =
                  case hscTarget dflags of
                    -- If we are compiling for the interpreter we will insert
                    -- any necessary SPT entries dynamically
                    HscInterpreted -> id
                    -- otherwise add a C stub to do so
                    _              -> (`appendStubC` spt_init_code)

376 377 378 379 380 381 382 383 384 385 386 387 388 389
              -- The completed type environment is gotten from
              --      a) the types and classes defined here (plus implicit things)
              --      b) adding Ids with correct IdInfo, including unfoldings,
              --              gotten from the bindings
              -- From (b) we keep only those Ids with External names;
              --          the CoreTidy pass makes sure these are all and only
              --          the externally-accessible ones
              -- This truncates the type environment to include only the
              -- exported Ids and things needed from them, which saves space
              --
              -- See Note [Don't attempt to trim data types]
              ; final_ids  = [ if omit_prags then trimId id else id
                             | id <- bindersOfBinds tidy_binds
                             , isExternalName (idName id)
390
                             , not (isWiredIn id)
391 392
                             ]   -- See Note [Drop wired-in things]

393
              ; final_tcs      = filterOut isWiredIn tcs
394 395 396 397 398 399 400 401
                                 -- See Note [Drop wired-in things]
              ; type_env       = typeEnvFromEntities final_ids final_tcs fam_insts
              ; tidy_cls_insts = mkFinalClsInsts type_env cls_insts
              ; tidy_patsyns   = mkFinalPatSyns  type_env patsyns
              ; tidy_type_env  = extendTypeEnvWithPatSyns tidy_patsyns type_env
              ; tidy_rules     = tidyRules tidy_env trimmed_rules

              ; -- See Note [Injecting implicit bindings]
402
                all_tidy_binds = implicit_binds ++ tidy_binds'
403 404

              -- Get the TyCons to generate code for.  Careful!  We must use
405
              -- the untidied TyCons here, because we need
406 407 408 409 410 411
              --  (a) implicit TyCons arising from types and classes defined
              --      in this module
              --  (b) wired-in TyCons, which are normally removed from the
              --      TypeEnv we put in the ModDetails
              --  (c) Constructors even if they are not exported (the
              --      tidied TypeEnv has trimmed these away)
412
              ; alg_tycons = filter isAlgTyCon tcs
413
              }
414

415
        ; endPassIO hsc_env print_unqual CoreTidy all_tidy_binds tidy_rules
416

417 418
          -- If the endPass didn't print the rules, but ddump-rules is
          -- on, print now
419
        ; unless (dopt Opt_D_dump_simpl dflags) $
420
            Err.dumpIfSet_dyn dflags Opt_D_dump_rules
421
              (showSDoc dflags (ppr CoreTidy <+> text "rules"))
Sylvain Henry's avatar
Sylvain Henry committed
422
              Err.FormatText
Sylvain Henry's avatar
Sylvain Henry committed
423
              (pprRulesForUser dflags tidy_rules)
424

425 426
          -- Print one-line size info
        ; let cs = coreBindsStats tidy_binds
427
        ; Err.dumpIfSet_dyn dflags Opt_D_dump_core_stats "Core Stats"
Sylvain Henry's avatar
Sylvain Henry committed
428
            Err.FormatText
429 430 431 432 433
            (text "Tidy size (terms,types,coercions)"
             <+> ppr (moduleName mod) <> colon
             <+> int (cs_tm cs)
             <+> int (cs_ty cs)
             <+> int (cs_co cs) )
434

435 436 437
        ; return (CgGuts { cg_module   = mod,
                           cg_tycons   = alg_tycons,
                           cg_binds    = all_tidy_binds,
438
                           cg_foreign  = add_spt_init_code foreign_stubs,
439
                           cg_foreign_files = foreign_files,
440 441
                           cg_dep_pkgs = map fst $ dep_pkgs deps,
                           cg_hpc_info = hpc_info,
442 443
                           cg_modBreaks = modBreaks,
                           cg_spt_entries = spt_entries },
444

445
                   ModDetails { md_types     = tidy_type_env,
446
                                md_rules     = tidy_rules,
447
                                md_insts     = tidy_cls_insts,
448
                                md_fam_insts = fam_insts,
449
                                md_exports   = exports,
450 451
                                md_anns      = anns,      -- are already tidy
                                md_complete_sigs = complete_sigs
chak@cse.unsw.edu.au.'s avatar
chak@cse.unsw.edu.au. committed
452
                              })
453
        }
454 455
  where
    dflags = hsc_dflags hsc_env
456

457
--------------------------
458 459 460 461 462 463 464 465 466 467 468
trimId :: Id -> Id
trimId id
  | not (isImplicitId id)
  = id `setIdInfo` vanillaIdInfo
  | otherwise
  = id

{- Note [Drop wired-in things]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We never put wired-in TyCons or Ids in an interface file.
They are wired-in, so the compiler knows about them already.
469

470 471 472
Note [Don't attempt to trim data types]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For some time GHC tried to avoid exporting the data constructors
473
of a data type if it wasn't strictly necessary to do so; see #835.
Austin Seipp's avatar
Austin Seipp committed
474
But "strictly necessary" accumulated a longer and longer list
Gabor Greif's avatar
typos  
Gabor Greif committed
475
of exceptions, and finally I gave up the battle:
476 477 478 479 480 481

    commit 9a20e540754fc2af74c2e7392f2786a81d8d5f11
    Author: Simon Peyton Jones <simonpj@microsoft.com>
    Date:   Thu Dec 6 16:03:16 2012 +0000

    Stop attempting to "trim" data types in interface files
Austin Seipp's avatar
Austin Seipp committed
482

483 484 485 486
    Without -O, we previously tried to make interface files smaller
    by not including the data constructors of data types.  But
    there are a lot of exceptions, notably when Template Haskell is
    involved or, more recently, DataKinds.
Austin Seipp's avatar
Austin Seipp committed
487

488
    However #7445 shows that even without TemplateHaskell, using
489 490
    the Data class and invoking Language.Haskell.TH.Quote.dataToExpQ
    is enough to require us to expose the data constructors.
Austin Seipp's avatar
Austin Seipp committed
491

492 493 494 495 496
    So I've given up on this "optimisation" -- it's probably not
    important anyway.  Now I'm simply not attempting to trim off
    the data constructors.  The gain in simplicity is worth the
    modest cost in interface file growth, which is limited to the
    bits reqd to describe those data constructors.
497

Austin Seipp's avatar
Austin Seipp committed
498 499
************************************************************************
*                                                                      *
500
        Implicit bindings
Austin Seipp's avatar
Austin Seipp committed
501 502
*                                                                      *
************************************************************************
503 504 505

Note [Injecting implicit bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Edward Z. Yang's avatar
Edward Z. Yang committed
506
We inject the implicit bindings right at the end, in CoreTidy.
507 508
Some of these bindings, notably record selectors, are not
constructed in an optimised form.  E.g. record selector for
509
        data T = MkT { x :: {-# UNPACK #-} !Int }
510
Then the unfolding looks like
511
        x = \t. case t of MkT x1 -> let x = I# x1 in x
512
This generates bad code unless it's first simplified a bit.  That is
513
why CoreUnfold.mkImplicitUnfolding uses simpleOptExpr to do a bit of
514
optimisation first.  (Only matters when the selector is used curried;
515
eg map x ys.)  See #2070.
516

simonpj@microsoft.com's avatar
simonpj@microsoft.com committed
517 518 519 520 521
[Oct 09: in fact, record selectors are no longer implicit Ids at all,
because we really do want to optimise them properly. They are treated
much like any other Id.  But doing "light" optimisation on an implicit
Id still makes sense.]

522 523
At one time I tried injecting the implicit bindings *early*, at the
beginning of SimplCore.  But that gave rise to real difficulty,
Gabor Greif's avatar
typos  
Gabor Greif committed
524
because GlobalIds are supposed to have *fixed* IdInfo, but the
525 526 527
simplifier and other core-to-core passes mess with IdInfo all the
time.  The straw that broke the camels back was when a class selector
got the wrong arity -- ie the simplifier gave it arity 2, whereas
528
importing modules were expecting it to have arity 1 (#2844).
529 530
It's much safer just to inject them right at the end, after tidying.

531
Oh: two other reasons for injecting them late:
simonpj@microsoft.com's avatar
simonpj@microsoft.com committed
532

533
  - If implicit Ids are already in the bindings when we start tidying,
534
    we'd have to be careful not to treat them as external Ids (in
535
    the sense of chooseExternalIds); else the Ids mentioned in *their*
536
    RHSs will be treated as external and you get an interface file
537
    saying      a18 = <blah>
538
    but nothing referring to a18 (because the implicit Id is the
simonpj@microsoft.com's avatar
simonpj@microsoft.com committed
539
    one that does, and implicit Ids don't appear in interface files).
540 541 542 543

  - More seriously, the tidied type-envt will include the implicit
    Id replete with a18 in its unfolding; but we won't take account
    of a18 when computing a fingerprint for the class; result chaos.
544

simonpj@microsoft.com's avatar
simonpj@microsoft.com committed
545 546 547
There is one sort of implicit binding that is injected still later,
namely those for data constructor workers. Reason (I think): it's
really just a code generation trick.... binding itself makes no sense.
548
See Note [Data constructor workers] in CorePrep.
Austin Seipp's avatar
Austin Seipp committed
549
-}
550

551 552 553 554 555
getImplicitBinds :: TyCon -> [CoreBind]
getImplicitBinds tc = cls_binds ++ getTyConImplicitBinds tc
  where
    cls_binds = maybe [] getClassImplicitBinds (tyConClass_maybe tc)

556
getTyConImplicitBinds :: TyCon -> [CoreBind]
557 558 559
getTyConImplicitBinds tc
  | isNewTyCon tc = []  -- See Note [Compulsory newtype unfolding] in MkId
  | otherwise     = map get_defn (mapMaybe dataConWrapId_maybe (tyConDataCons tc))
560 561

getClassImplicitBinds :: Class -> [CoreBind]
562 563 564
getClassImplicitBinds cls
  = [ NonRec op (mkDictSelRhs cls val_index)
    | (op, val_index) <- classAllSelIds cls `zip` [0..] ]
565 566 567

get_defn :: Id -> CoreBind
get_defn id = NonRec id (unfoldingTemplate (realIdUnfolding id))
568

Austin Seipp's avatar
Austin Seipp committed
569 570 571
{-
************************************************************************
*                                                                      *
572
\subsection{Step 1: finding externals}
Austin Seipp's avatar
Austin Seipp committed
573 574
*                                                                      *
************************************************************************
575

576
See Note [Choosing external Ids].
Austin Seipp's avatar
Austin Seipp committed
577
-}
578 579

type UnfoldEnv  = IdEnv (Name{-new name-}, Bool {-show unfolding-})
580 581
  -- Maps each top-level Id to its new Name (the Id is tidied in step 2)
  -- The Unique is unchanged.  If the new Name is external, it will be
582
  -- visible in the interface file.
Simon Marlow's avatar
Simon Marlow committed
583 584
  --
  -- Bool => expose unfolding or not.
585 586 587

chooseExternalIds :: HscEnv
                  -> Module
588
                  -> Bool -> Bool
589
                  -> [CoreBind]
590 591
                  -> [CoreBind]
                  -> [CoreRule]
592
                  -> IO (UnfoldEnv, TidyOccEnv)
593
                  -- Step 1 from the notes above
594

595
chooseExternalIds hsc_env mod omit_prags expose_all binds implicit_binds imp_id_rules
596 597 598
  = do { (unfold_env1,occ_env1) <- search init_work_list emptyVarEnv init_occ_env
       ; let internal_ids = filter (not . (`elemVarEnv` unfold_env1)) binders
       ; tidy_internal internal_ids unfold_env1 occ_env1 }
599
 where
600
  nc_var = hsc_NC hsc_env
601

602
  -- init_ext_ids is the initial list of Ids that should be
603
  -- externalised.  It serves as the starting point for finding a
604 605
  -- deterministic, tidy, renaming for all external Ids in this
  -- module.
606
  --
607
  -- It is sorted, so that it has a deterministic order (i.e. it's the
608 609 610
  -- same list every time this module is compiled), in contrast to the
  -- bindings, which are ordered non-deterministically.
  init_work_list = zip init_ext_ids init_ext_ids
611
  init_ext_ids   = sortBy (compare `on` getOccName) $ filter is_external binders
612

613 614
  -- An Id should be external if either (a) it is exported,
  -- (b) it appears in the RHS of a local rule for an imported Id, or
615
  -- See Note [Which rules to expose]
616
  is_external id = isExportedId id || id `elemVarSet` rule_rhs_vars
617

618
  rule_rhs_vars  = mapUnionVarSet ruleRhsFreeVars imp_id_rules
619

620
  binders          = map fst $ flattenBinds binds
621
  implicit_binders = bindersOfBinds implicit_binds
622
  binder_set       = mkVarSet binders
623

624
  avoids   = [getOccName name | bndr <- binders ++ implicit_binders,
625
                                let name = idName bndr,
626
                                isExternalName name ]
627 628 629 630 631 632
                -- In computing our "avoids" list, we must include
                --      all implicit Ids
                --      all things with global names (assigned once and for
                --                                      all by the renamer)
                -- since their names are "taken".
                -- The type environment is a convenient source of such things.
633 634
                -- In particular, the set of binders doesn't include
                -- implicit Ids at this stage.
635

636 637 638 639 640 641 642 643
        -- We also make sure to avoid any exported binders.  Consider
        --      f{-u1-} = 1     -- Local decl
        --      ...
        --      f{-u2-} = 2     -- Exported decl
        --
        -- The second exported decl must 'get' the name 'f', so we
        -- have to put 'f' in the avoids list before we get to the first
        -- decl.  tidyTopId then does a no-op on exported binders.
644 645 646
  init_occ_env = initTidyOccEnv avoids


Gabor Greif's avatar
Gabor Greif committed
647
  search :: [(Id,Id)]    -- The work-list: (external id, referring id)
648
                         -- Make a tidy, external Name for the external id,
649 650
                         --   add it to the UnfoldEnv, and do the same for the
                         --   transitive closure of Ids it refers to
651 652
                         -- The referring id is used to generate a tidy
                         ---  name for the external id
653 654 655 656 657 658
         -> UnfoldEnv    -- id -> (new Name, show_unfold)
         -> TidyOccEnv   -- occ env for choosing new Names
         -> IO (UnfoldEnv, TidyOccEnv)

  search [] unfold_env occ_env = return (unfold_env, occ_env)

659 660
  search ((idocc,referrer) : rest) unfold_env occ_env
    | idocc `elemVarEnv` unfold_env = search rest unfold_env occ_env
661
    | otherwise = do
662
      (occ_env', name') <- tidyTopName mod nc_var (Just referrer) occ_env idocc
663
      let
664 665
          (new_ids, show_unfold)
                | omit_prags = ([], False)
666
                | otherwise  = addExternal expose_all refined_id
667

668 669
                -- 'idocc' is an *occurrence*, but we need to see the
                -- unfolding in the *definition*; so look up in binder_set
670 671 672 673 674 675 676
          refined_id = case lookupVarSet binder_set idocc of
                         Just id -> id
                         Nothing -> WARN( True, ppr idocc ) idocc

          unfold_env' = extendVarEnv unfold_env idocc (name',show_unfold)
          referrer' | isExportedId refined_id = refined_id
                    | otherwise               = referrer
677
      --
678
      search (zip new_ids (repeat referrer') ++ rest) unfold_env' occ_env'
679 680 681 682 683 684 685 686 687

  tidy_internal :: [Id] -> UnfoldEnv -> TidyOccEnv
                -> IO (UnfoldEnv, TidyOccEnv)
  tidy_internal []       unfold_env occ_env = return (unfold_env,occ_env)
  tidy_internal (id:ids) unfold_env occ_env = do
      (occ_env', name') <- tidyTopName mod nc_var Nothing occ_env id
      let unfold_env' = extendVarEnv unfold_env id (name',False)
      tidy_internal ids unfold_env' occ_env'

688
addExternal :: Bool -> Id -> ([Id], Bool)
689
addExternal expose_all id = (new_needed_ids, show_unfold)
690
  where
691
    new_needed_ids = bndrFvsInOrder show_unfold id
692
    idinfo         = idInfo id
693
    show_unfold    = show_unfolding (unfoldingInfo idinfo)
694
    never_active   = isNeverActive (inlinePragmaActivation (inlinePragInfo idinfo))
695
    loop_breaker   = isStrongLoopBreaker (occInfo idinfo)
696
    bottoming_fn   = isBottomingSig (strictnessInfo idinfo)
697

698 699 700
        -- Stuff to do with the Id's unfolding
        -- We leave the unfolding there even if there is a worker
        -- In GHCi the unfolding is used by importers
701 702

    show_unfolding (CoreUnfolding { uf_src = src, uf_guidance = guidance })
703 704
       =  expose_all         -- 'expose_all' says to expose all
                             -- unfoldings willy-nilly
705

706 707
       || isStableSource src     -- Always expose things whose
                                 -- source is an inline rule
708

709 710 711 712
       || not (bottoming_fn      -- No need to inline bottom functions
           || never_active       -- Or ones that say not to
           || loop_breaker       -- Or that are loop breakers
           || neverUnfoldGuidance guidance)
713 714
    show_unfolding (DFunUnfolding {}) = True
    show_unfolding _                  = False
715

Austin Seipp's avatar
Austin Seipp committed
716 717 718
{-
************************************************************************
*                                                                      *
719
               Deterministic free variables
Austin Seipp's avatar
Austin Seipp committed
720 721
*                                                                      *
************************************************************************
722

723 724 725 726
We want a deterministic free-variable list.  exprFreeVars gives us
a VarSet, which is in a non-deterministic order when converted to a
list.  Hence, here we define a free-variable finder that returns
the free variables in the order that they are encountered.
727

728
See Note [Choosing external Ids]
Austin Seipp's avatar
Austin Seipp committed
729
-}
730

731 732 733
bndrFvsInOrder :: Bool -> Id -> [Id]
bndrFvsInOrder show_unfold id
  = run (dffvLetBndr show_unfold id)
734

735 736 737 738
run :: DFFV () -> [Id]
run (DFFV m) = case m emptyVarSet (emptyVarSet, []) of
                 ((_,ids),_) -> ids

739 740
newtype DFFV a
  = DFFV (VarSet              -- Envt: non-top-level things that are in scope
741
                              -- we don't want to record these as free vars
742
      -> (VarSet, [Var])      -- Input State: (set, list) of free vars so far
743
      -> ((VarSet,[Var]),a))  -- Output state
744
    deriving (Functor)
Austin Seipp's avatar
Austin Seipp committed
745 746

instance Applicative DFFV where
747
    pure a = DFFV $ \_ st -> (st, a)
Austin Seipp's avatar
Austin Seipp committed
748 749
    (<*>) = ap

750
instance Monad DFFV where
751 752 753 754 755 756 757 758 759 760
  (DFFV m) >>= k = DFFV $ \env st ->
    case m env st of
       (st',a) -> case k a of
                     DFFV f -> f env st'

extendScope :: Var -> DFFV a -> DFFV a
extendScope v (DFFV f) = DFFV (\env st -> f (extendVarSet env v) st)

extendScopeList :: [Var] -> DFFV a -> DFFV a
extendScopeList vs (DFFV f) = DFFV (\env st -> f (extendVarSetList env vs) st)
761 762

insert :: Var -> DFFV ()
763 764
insert v = DFFV $ \ env (set, ids) ->
           let keep_me = isLocalId v &&
765
                         not (v `elemVarSet` env) &&
766 767
                           not (v `elemVarSet` set)
           in if keep_me
768 769 770
              then ((extendVarSet set v, v:ids), ())
              else ((set,                ids),   ())

771 772

dffvExpr :: CoreExpr -> DFFV ()
773
dffvExpr (Var v)              = insert v
774 775
dffvExpr (App e1 e2)          = dffvExpr e1 >> dffvExpr e2
dffvExpr (Lam v e)            = extendScope v (dffvExpr e)
776 777
dffvExpr (Tick (Breakpoint _ ids) e) = mapM_ insert ids >> dffvExpr e
dffvExpr (Tick _other e)    = dffvExpr e
778
dffvExpr (Cast e _)           = dffvExpr e
779 780 781 782 783 784 785 786 787 788
dffvExpr (Let (NonRec x r) e) = dffvBind (x,r) >> extendScope x (dffvExpr e)
dffvExpr (Let (Rec prs) e)    = extendScopeList (map fst prs) $
                                (mapM_ dffvBind prs >> dffvExpr e)
dffvExpr (Case e b _ as)      = dffvExpr e >> extendScope b (mapM_ dffvAlt as)
dffvExpr _other               = return ()

dffvAlt :: (t, [Var], CoreExpr) -> DFFV ()
dffvAlt (_,xs,r) = extendScopeList xs (dffvExpr r)

dffvBind :: (Id, CoreExpr) -> DFFV ()
789
dffvBind(x,r)
790 791
  | not (isId x) = dffvExpr r
  | otherwise    = dffvLetBndr False x >> dffvExpr r
792 793
                -- Pass False because we are doing the RHS right here
                -- If you say True you'll get *exponential* behaviour!
794 795

dffvLetBndr :: Bool -> Id -> DFFV ()
796 797 798 799 800 801 802 803 804 805 806
-- Gather the free vars of the RULES and unfolding of a binder
-- We always get the free vars of a *stable* unfolding, but
-- for a *vanilla* one (InlineRhs), the flag controls what happens:
--   True <=> get fvs of even a *vanilla* unfolding
--   False <=> ignore an InlineRhs
-- For nested bindings (call from dffvBind) we always say "False" because
--       we are taking the fvs of the RHS anyway
-- For top-level bindings (call from addExternal, via bndrFvsInOrder)
--       we say "True" if we are exposing that unfolding
dffvLetBndr vanilla_unfold id
  = do { go_unf (unfoldingInfo idinfo)
807
       ; mapM_ go_rule (ruleInfoRules (ruleInfo idinfo)) }
808
  where
809 810 811 812
    idinfo = idInfo id

    go_unf (CoreUnfolding { uf_tmpl = rhs, uf_src = src })
       = case src of
813 814
           InlineRhs | vanilla_unfold -> dffvExpr rhs
                     | otherwise      -> return ()
815
           _                          -> dffvExpr rhs
816

Austin Seipp's avatar
Austin Seipp committed
817
    go_unf (DFunUnfolding { df_bndrs = bndrs, df_args = args })
818
             = extendScopeList bndrs $ mapM_ dffvExpr args
819 820 821 822 823
    go_unf _ = return ()

    go_rule (BuiltinRule {}) = return ()
    go_rule (Rule { ru_bndrs = bndrs, ru_rhs = rhs })
      = extendScopeList bndrs (dffvExpr rhs)
824

Austin Seipp's avatar
Austin Seipp committed
825 826 827
{-
************************************************************************
*                                                                      *
828
               findExternalRules
Austin Seipp's avatar
Austin Seipp committed
829 830
*                                                                      *
************************************************************************
831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863

Note [Finding external rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The complete rules are gotten by combining
   a) local rules for imported Ids
   b) rules embedded in the top-level Ids

There are two complications:
  * Note [Which rules to expose]
  * Note [Trimming auto-rules]

Note [Which rules to expose]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The function 'expose_rule' filters out rules that mention, on the LHS,
Ids that aren't externally visible; these rules can't fire in a client
module.

The externally-visible binders are computed (by chooseExternalIds)
assuming that all orphan rules are externalised (see init_ext_ids in
function 'search'). So in fact it's a bit conservative and we may
export more than we need.  (It's a sort of mutual recursion.)

Note [Trimming auto-rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Second, with auto-specialisation we may specialise local or imported
dfuns or INLINE functions, and then later inline them.  That may leave
behind something like
   RULE "foo" forall d. f @ Int d = f_spec
where f is either local or imported, and there is no remaining
reference to f_spec except from the RULE.

Now that RULE *might* be useful to an importing module, but that is
purely speculative, and meanwhile the code is taking up space and
Simon Peyton Jones's avatar
Simon Peyton Jones committed
864 865 866 867
codegen time.  I found that binary sizes jumped by 6-10% when I
started to specialise INLINE functions (again, Note [Inline
specialisations] in Specialise).

868
So it seems better to drop the binding for f_spec, and the rule
Simon Peyton Jones's avatar
Simon Peyton Jones committed
869 870
itself, if the auto-generated rule is the *only* reason that it is
being kept alive.
871 872 873 874 875 876 877 878 879 880 881 882

(The RULE still might have been useful in the past; that is, it was
the right thing to have generated it in the first place.  See Note
[Inline specialisations] in Specialise.  But now it has served its
purpose, and can be discarded.)

So findExternalRules does this:
  * Remove all bindings that are kept alive *only* by isAutoRule rules
      (this is done in trim_binds)
  * Remove all auto rules that mention bindings that have been removed
      (this is done by filtering by keep_rule)

Simon Peyton Jones's avatar
Simon Peyton Jones committed
883
NB: if a binding is kept alive for some *other* reason (e.g. f_spec is
884 885
called in the final code), we keep the rule too.

Simon Peyton Jones's avatar
Simon Peyton Jones committed
886
This stuff is the only reason for the ru_auto field in a Rule.
Austin Seipp's avatar
Austin Seipp committed
887
-}
888 889 890 891 892 893 894 895 896 897 898

findExternalRules