TcInteract.lhs 84.8 KB
 simonpj@microsoft.com committed Sep 13, 2010 1 2 \begin{code} module TcInteract (  3  solveInteract, AtomicInert, tyVarsOfInert,  simonpj@microsoft.com committed Nov 12, 2010 4  InertSet, emptyInert, updInertSet, extractUnsolved, solveOne, foldISEqCts  simonpj@microsoft.com committed Sep 13, 2010 5 6 7 8  ) where #include "HsVersions.h"  dimitris@microsoft.com committed Oct 04, 2010 9   simonpj@microsoft.com committed Sep 13, 2010 10 11 12 13 14 15 16 17 18 import BasicTypes import TcCanonical import VarSet import Type import Id import Var import TcType  simonpj@microsoft.com committed Nov 12, 2010 19 import HsBinds  simonpj@microsoft.com committed Sep 13, 2010 20   simonpj@microsoft.com committed Nov 12, 2010 21 22 23 import InstEnv import Class import TyCon  simonpj@microsoft.com committed Sep 13, 2010 24 25 26 27 28 29 30 31 32 import Name import FunDeps import Control.Monad ( when ) import Coercion import Outputable  simonpj@microsoft.com committed Nov 12, 2010 33 import TcRnTypes  simonpj@microsoft.com committed Sep 13, 2010 34 import TcErrors  simonpj@microsoft.com committed Nov 12, 2010 35 import TcSMonad  simonpj@microsoft.com committed Oct 07, 2010 36 import Bag  simonpj@microsoft.com committed Nov 12, 2010 37 import qualified Data.Map as Map  dimitris@microsoft.com committed Oct 04, 2010 38   simonpj@microsoft.com committed Sep 13, 2010 39 40 41 42 43 import Control.Monad( zipWithM, unless ) import FastString ( sLit ) import DynFlags \end{code}  dimitris@microsoft.com committed Oct 06, 2010 44 Note [InertSet invariants]  simonpj@microsoft.com committed Sep 13, 2010 45 46 47 48 49 50 51 52 53 54 55 56 57 58 ~~~~~~~~~~~~~~~~~~~~~~~~~~~ An InertSet is a bag of canonical constraints, with the following invariants: 1 No two constraints react with each other. A tricky case is when there exists a given (solved) dictionary constraint and a wanted identical constraint in the inert set, but do not react because reaction would create loopy dictionary evidence for the wanted. See note [Recursive dictionaries] 2 Given equalities form an idempotent substitution [none of the given LHS's occur in any of the given RHS's or reactant parts] 3 Wanted equalities also form an idempotent substitution  simonpj@microsoft.com committed Nov 12, 2010 59   simonpj@microsoft.com committed Sep 13, 2010 60 61 62 63 64 65  4 The entire set of equalities is acyclic. 5 Wanted dictionaries are inert with the top-level axiom set 6 Equalities of the form tv1 ~ tv2 always have a touchable variable on the left (if possible).  simonpj@microsoft.com committed Nov 12, 2010 66 67  7 No wanted constraints tv1 ~ tv2 with tv1 touchable. Such constraints  simonpj@microsoft.com committed Sep 13, 2010 68 69  will be marked as solved right before being pushed into the inert set. See note [Touchables and givens].  simonpj@microsoft.com committed Nov 12, 2010 70 71 72  8 No Given constraint mentions a touchable unification variable, except if the  simonpj@microsoft.com committed Sep 13, 2010 73 74 75 76 77 78 79 80 81 82 83 84 85 86  Note that 6 and 7 are /not/ enforced by canonicalization but rather by insertion in the inert list, ie by TcInteract. During the process of solving, the inert set will contain some previously given constraints, some wanted constraints, and some given constraints which have arisen from solving wanted constraints. For now we do not distinguish between given and solved constraints. Note that we must switch wanted inert items to given when going under an implication constraint (when in top-level inference mode). \begin{code}  87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 data CCanMap a = CCanMap { cts_givder :: Map.Map a CanonicalCts -- Invariant: all Given or Derived , cts_wanted :: Map.Map a CanonicalCts } -- Invariant: all Wanted cCanMapToBag :: Ord a => CCanMap a -> CanonicalCts cCanMapToBag cmap = Map.fold unionBags rest_cts (cts_givder cmap) where rest_cts = Map.fold unionBags emptyCCan (cts_wanted cmap) emptyCCanMap :: CCanMap a emptyCCanMap = CCanMap { cts_givder = Map.empty, cts_wanted = Map.empty } updCCanMap:: Ord a => (a,CanonicalCt) -> CCanMap a -> CCanMap a updCCanMap (a,ct) cmap = case cc_flavor ct of Wanted {} -> cmap { cts_wanted = Map.insertWith unionBags a this_ct (cts_wanted cmap) } _ -> cmap { cts_givder = Map.insertWith unionBags a this_ct (cts_givder cmap) } where this_ct = singleCCan ct getRelevantCts :: Ord a => a -> CCanMap a -> (CanonicalCts, CCanMap a) -- Gets the relevant constraints and returns the rest of the CCanMap getRelevantCts a cmap = let relevant = unionBags (Map.findWithDefault emptyCCan a (cts_wanted cmap)) (Map.findWithDefault emptyCCan a (cts_givder cmap)) residual_map = cmap { cts_wanted = Map.delete a (cts_wanted cmap) , cts_givder = Map.delete a (cts_givder cmap) } in (relevant, residual_map) extractUnsolvedCMap :: Ord a => CCanMap a -> (CanonicalCts, CCanMap a) -- Gets the wanted constraints and returns a residual CCanMap extractUnsolvedCMap cmap = let unsolved = Map.fold unionBags emptyCCan (cts_wanted cmap) in (unsolved, cmap { cts_wanted = Map.empty})  simonpj@microsoft.com committed Sep 13, 2010 122 -- See Note [InertSet invariants]  dimitris@microsoft.com committed Oct 04, 2010 123 data InertSet  124 125 126 127 128 129 130 131 132  = IS { inert_eqs :: CanonicalCts -- Equalities only (CTyEqCan) , inert_dicts :: CCanMap Class -- Dictionaries only , inert_ips :: CCanMap (IPName Name) -- Implicit parameters , inert_funeqs :: CCanMap TyCon -- Type family equalities only -- This representation allows us to quickly get to the relevant -- inert constraints when interacting a work item with the inert set.  133 134  , inert_fds :: FDImprovements -- List of pairwise improvements that have kicked in already -- and reside either in the worklist or in the inerts  simonpj@microsoft.com committed Nov 12, 2010 135  }  dimitris@microsoft.com committed Oct 04, 2010 136   137 138 139 140 141 142 143 144 tyVarsOfInert :: InertSet -> TcTyVarSet tyVarsOfInert (IS { inert_eqs = eqs , inert_dicts = dictmap , inert_ips = ipmap , inert_funeqs = funeqmap }) = tyVarsOfCanonicals cts where cts = eqs andCCan cCanMapToBag dictmap andCCan cCanMapToBag ipmap andCCan cCanMapToBag funeqmap  145 146 147 type FDImprovement = (PredType,PredType) type FDImprovements = [(PredType,PredType)]  simonpj@microsoft.com committed Sep 13, 2010 148 instance Outputable InertSet where  149  ppr is = vcat [ vcat (map ppr (Bag.bagToList $inert_eqs is))  150 151 152  , vcat (map ppr (Bag.bagToList$ cCanMapToBag (inert_dicts is))) , vcat (map ppr (Bag.bagToList $cCanMapToBag (inert_ips is))) , vcat (map ppr (Bag.bagToList$ cCanMapToBag (inert_funeqs is)))  dimitris@microsoft.com committed Oct 04, 2010 153 154  ]  simonpj@microsoft.com committed Sep 13, 2010 155 emptyInert :: InertSet  156 157 158 159 emptyInert = IS { inert_eqs = Bag.emptyBag , inert_dicts = emptyCCanMap , inert_ips = emptyCCanMap , inert_funeqs = emptyCCanMap  simonpj@microsoft.com committed Nov 12, 2010 160  , inert_fds = [] }  dimitris@microsoft.com committed Oct 04, 2010 161 162  updInertSet :: InertSet -> AtomicInert -> InertSet  163 164 165 166 167 168 169 170 171 172 updInertSet is item | isCTyEqCan item -- Other equality = let eqs' = inert_eqs is Bag.snocBag item in is { inert_eqs = eqs' } | Just cls <- isCDictCan_Maybe item -- Dictionary = is { inert_dicts = updCCanMap (cls,item) (inert_dicts is) } | Just x <- isCIPCan_Maybe item -- IP = is { inert_ips = updCCanMap (x,item) (inert_ips is) } | Just tc <- isCFunEqCan_Maybe item -- Function equality = is { inert_funeqs = updCCanMap (tc,item) (inert_funeqs is) }  173  | otherwise  174  = pprPanic "Unknown form of constraint!" (ppr item)  175 176 177 178  updInertSetFDImprs :: InertSet -> Maybe FDImprovement -> InertSet updInertSetFDImprs is (Just fdi) = is { inert_fds = fdi : inert_fds is } updInertSetFDImprs is Nothing = is  dimitris@microsoft.com committed Oct 04, 2010 179   180 181 182 183 184 foldISEqCtsM :: Monad m => (a -> AtomicInert -> m a) -> a -> InertSet -> m a -- Fold over the equalities of the inerts foldISEqCtsM k z IS { inert_eqs = eqs } = Bag.foldlBagM k z eqs  simonpj@microsoft.com committed Nov 12, 2010 185 186 187 188 foldISEqCts :: (a -> AtomicInert -> a) -> a -> InertSet -> a foldISEqCts k z IS { inert_eqs = eqs } = Bag.foldlBag k z eqs  simonpj@microsoft.com committed Sep 13, 2010 189 extractUnsolved :: InertSet -> (InertSet, CanonicalCts)  190 extractUnsolved is@(IS {inert_eqs = eqs})  simonpj@microsoft.com committed Nov 12, 2010 191 192 193 194 195  = let is_solved = is { inert_eqs = solved_eqs , inert_dicts = solved_dicts , inert_ips = solved_ips , inert_funeqs = solved_funeqs } in (is_solved, unsolved)  196 197 198 199 200  where (unsolved_eqs, solved_eqs) = Bag.partitionBag isWantedCt eqs (unsolved_ips, solved_ips) = extractUnsolvedCMap (inert_ips is) (unsolved_dicts, solved_dicts) = extractUnsolvedCMap (inert_dicts is) (unsolved_funeqs, solved_funeqs) = extractUnsolvedCMap (inert_funeqs is)  simonpj@microsoft.com committed Sep 13, 2010 201   202 203  unsolved = unsolved_eqs unionBags unsolved_ips unionBags unsolved_dicts unionBags unsolved_funeqs  dimitris@microsoft.com committed Oct 04, 2010 204   205 206 207 208 haveBeenImproved :: FDImprovements -> PredType -> PredType -> Bool haveBeenImproved [] _ _ = False haveBeenImproved ((pty1,pty2):fdimprs) pty1' pty2' | tcEqPred pty1 pty1' && tcEqPred pty2 pty2'  209  = True  210  | tcEqPred pty1 pty2' && tcEqPred pty2 pty1'  211 212 213  = True | otherwise = haveBeenImproved fdimprs pty1' pty2'  214   215 getFDImprovements :: InertSet -> FDImprovements  216 -- Return a list of the improvements that have kicked in so far  217 getFDImprovements = inert_fds  dimitris@microsoft.com committed Oct 04, 2010 218   simonpj@microsoft.com committed Sep 13, 2010 219 220 \end{code}  simonpj@microsoft.com committed Nov 12, 2010 221 222 {-- DV: This note will go away!  simonpj@microsoft.com committed Sep 13, 2010 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 Note [Touchables and givens] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Touchable variables will never show up in givens which are inputs to the solver. However, touchables may show up in givens generated by the flattener. For example, axioms: G Int ~ Char F Char ~ Int wanted: F (G alpha) ~w Int canonicalises to G alpha ~g b F b ~w Int which can be put in the inert set. Suppose we also have a wanted alpha ~w Int We cannot rewrite the given G alpha ~g b using the wanted alpha ~w Int. Instead, after reacting alpha ~w Int with the whole inert set, we observe that we can solve it by unifying alpha with Int, so we mark it as solved and put it back in the *work list*. [We also immediately unify alpha := Int, without telling anyone, see trySpontaneousSolve function, to avoid doing this in the end.] Later, because it is solved (given, in effect), we can use it to rewrite G alpha ~g b to G Int ~g b, which gets put back in the work list. Eventually, we will dispatch the remaining wanted constraints using the top-level axioms. Finally, note that after reacting a wanted equality with the entire inert set we may end up with something like b ~w alpha which we should flip around to generate the solved constraint alpha ~s b.  simonpj@microsoft.com committed Nov 12, 2010 263 264 265 266 -}  simonpj@microsoft.com committed Sep 13, 2010 267 268 269 270 271 272 273 274 275 276 277 278 %********************************************************************* %* * * Main Interaction Solver * * * ********************************************************************** Note [Basic plan] ~~~~~~~~~~~~~~~~~ 1. Canonicalise (unary) 2. Pairwise interaction (binary) * Take one from work list * Try all pair-wise interactions with each constraint in inert  279 280 281 282  As an optimisation, we prioritize the equalities both in the worklist and in the inerts.  simonpj@microsoft.com committed Sep 13, 2010 283 284 285 286 287 288 289 290 3. Try to solve spontaneously for equalities involving touchables 4. Top-level interaction (binary wrt top-level) Superclass decomposition belongs in (4), see note [Superclasses] \begin{code} type AtomicInert = CanonicalCt -- constraint pulled from InertSet type WorkItem = CanonicalCt -- constraint pulled from WorkList  291 292 -- A mixture of Given, Wanted, and Derived constraints. -- We split between equalities and the rest to process equalities first.  simonpj@microsoft.com committed Oct 20, 2010 293 type WorkList = CanonicalCts  simonpj@microsoft.com committed Sep 13, 2010 294 295  unionWorkLists :: WorkList -> WorkList -> WorkList  simonpj@microsoft.com committed Oct 20, 2010 296 unionWorkLists = andCCan  simonpj@microsoft.com committed Sep 13, 2010 297 298  isEmptyWorkList :: WorkList -> Bool  simonpj@microsoft.com committed Oct 20, 2010 299 isEmptyWorkList = isEmptyCCan  simonpj@microsoft.com committed Sep 13, 2010 300 301  emptyWorkList :: WorkList  simonpj@microsoft.com committed Oct 20, 2010 302 emptyWorkList = emptyCCan  303   304 workListFromCCan :: CanonicalCt -> WorkList  simonpj@microsoft.com committed Oct 20, 2010 305 workListFromCCan = singleCCan  dimitris@microsoft.com committed Oct 04, 2010 306   simonpj@microsoft.com committed Oct 20, 2010 307 ------------------------  simonpj@microsoft.com committed Sep 13, 2010 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 data StopOrContinue = Stop -- Work item is consumed | ContinueWith WorkItem -- Not consumed instance Outputable StopOrContinue where ppr Stop = ptext (sLit "Stop") ppr (ContinueWith w) = ptext (sLit "ContinueWith") <+> ppr w -- Results after interacting a WorkItem as far as possible with an InertSet data StageResult = SR { sr_inerts :: InertSet -- The new InertSet to use (REPLACES the old InertSet) , sr_new_work :: WorkList -- Any new work items generated (should be ADDED to the old WorkList) -- Invariant: -- sr_stop = Just workitem => workitem is *not* in sr_inerts and -- workitem is inert wrt to sr_inerts , sr_stop :: StopOrContinue } instance Outputable StageResult where ppr (SR { sr_inerts = inerts, sr_new_work = work, sr_stop = stop }) = ptext (sLit "SR") <+> braces (sep [ ptext (sLit "inerts =") <+> ppr inerts <> comma , ptext (sLit "new work =") <+> ppr work <> comma , ptext (sLit "stop =") <+> ppr stop]) type SimplifierStage = WorkItem -> InertSet -> TcS StageResult -- Combine a sequence of simplifier 'stages' to create a pipeline runSolverPipeline :: [(String, SimplifierStage)] -> InertSet -> WorkItem -> TcS (InertSet, WorkList) -- Precondition: non-empty list of stages runSolverPipeline pipeline inerts workItem = do { traceTcS "Start solver pipeline" $vcat [ ptext (sLit "work item =") <+> ppr workItem , ptext (sLit "inerts =") <+> ppr inerts] ; let itr_in = SR { sr_inerts = inerts , sr_new_work = emptyWorkList , sr_stop = ContinueWith workItem } ; itr_out <- run_pipeline pipeline itr_in ; let new_inert = case sr_stop itr_out of Stop -> sr_inerts itr_out  dimitris@microsoft.com committed Oct 04, 2010 354  ContinueWith item -> sr_inerts itr_out updInertSet item  simonpj@microsoft.com committed Sep 13, 2010 355 356 357 358 359 360 361 362 363 364 365 366 367  ; return (new_inert, sr_new_work itr_out) } where run_pipeline :: [(String, SimplifierStage)] -> StageResult -> TcS StageResult run_pipeline [] itr = return itr run_pipeline _ itr@(SR { sr_stop = Stop }) = return itr run_pipeline ((name,stage):stages) (SR { sr_new_work = accum_work , sr_inerts = inerts , sr_stop = ContinueWith work_item }) = do { itr <- stage work_item inerts ; traceTcS ("Stage result (" ++ name ++ ")") (ppr itr)  368  ; let itr' = itr { sr_new_work = accum_work unionWorkLists sr_new_work itr }  simonpj@microsoft.com committed Sep 13, 2010 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399  ; run_pipeline stages itr' } \end{code} Example 1: Inert: {c ~ d, F a ~ t, b ~ Int, a ~ ty} (all given) Reagent: a ~ [b] (given) React with (c~d) ==> IR (ContinueWith (a~[b])) True [] React with (F a ~ t) ==> IR (ContinueWith (a~[b])) False [F [b] ~ t] React with (b ~ Int) ==> IR (ContinueWith (a~[Int]) True [] Example 2: Inert: {c ~w d, F a ~g t, b ~w Int, a ~w ty} Reagent: a ~w [b] React with (c ~w d) ==> IR (ContinueWith (a~[b])) True [] React with (F a ~g t) ==> IR (ContinueWith (a~[b])) True [] (can't rewrite given with wanted!) etc. Example 3: Inert: {a ~ Int, F Int ~ b} (given) Reagent: F a ~ b (wanted) React with (a ~ Int) ==> IR (ContinueWith (F Int ~ b)) True [] React with (F Int ~ b) ==> IR Stop True [] -- after substituting we re-canonicalize and get nothing \begin{code} -- Main interaction solver: we fully solve the worklist 'in one go', -- returning an extended inert set. -- -- See Note [Touchables and givens].  400 solveInteract :: InertSet -> CanonicalCts -> TcS InertSet  simonpj@microsoft.com committed Sep 13, 2010 401 402 solveInteract inert ws = do { dyn_flags <- getDynFlags  simonpj@microsoft.com committed Oct 20, 2010 403  ; solveInteractWithDepth (ctxtStkDepth dyn_flags,0,[]) inert ws  simonpj@microsoft.com committed Sep 13, 2010 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422  } solveOne :: InertSet -> WorkItem -> TcS InertSet solveOne inerts workItem = do { dyn_flags <- getDynFlags ; solveOneWithDepth (ctxtStkDepth dyn_flags,0,[]) inerts workItem } ----------------- solveInteractWithDepth :: (Int, Int, [WorkItem]) -> InertSet -> WorkList -> TcS InertSet solveInteractWithDepth ctxt@(max_depth,n,stack) inert ws | isEmptyWorkList ws = return inert | n > max_depth = solverDepthErrorTcS n stack | otherwise = do { traceTcS "solveInteractWithDepth"$  simonpj@microsoft.com committed Oct 20, 2010 423 424 425 426  vcat [ text "Current depth =" <+> ppr n , text "Max depth =" <+> ppr max_depth ] -- Solve equalities first  427  ; let (eqs, non_eqs) = Bag.partitionBag isCTyEqCan ws  simonpj@microsoft.com committed Oct 20, 2010 428 429  ; is_from_eqs <- Bag.foldlBagM (solveOneWithDepth ctxt) inert eqs ; Bag.foldlBagM (solveOneWithDepth ctxt) is_from_eqs non_eqs }  simonpj@microsoft.com committed Sep 13, 2010 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452  ------------------ -- Fully interact the given work item with an inert set, and return a -- new inert set which has assimilated the new information. solveOneWithDepth :: (Int, Int, [WorkItem]) -> InertSet -> WorkItem -> TcS InertSet solveOneWithDepth (max_depth, n, stack) inert work = do { traceTcS0 (indent ++ "Solving {") (ppr work) ; (new_inert, new_work) <- runSolverPipeline thePipeline inert work ; traceTcS0 (indent ++ "Subgoals:") (ppr new_work) -- Recursively solve the new work generated -- from workItem, with a greater depth ; res_inert <- solveInteractWithDepth (max_depth, n+1, work:stack) new_inert new_work ; traceTcS0 (indent ++ "Done }") (ppr work) ; return res_inert } where indent = replicate (2*n) ' ' thePipeline :: [(String,SimplifierStage)]  453 454 455 456 thePipeline = [ ("interact with inert eqs", interactWithInertEqsStage) , ("interact with inerts", interactWithInertsStage) , ("spontaneous solve", spontaneousSolveStage) , ("top-level reactions", topReactionsStage) ]  simonpj@microsoft.com committed Sep 13, 2010 457 458 459 460 461 462 463 464 \end{code} ********************************************************************************* * * The spontaneous-solve Stage * * *********************************************************************************  465 466 467 468 469 470 Note [Efficient Orientation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ There are two cases where we have to be careful about orienting equalities to get better efficiency.  simonpj@microsoft.com committed Nov 12, 2010 471 Case 1: In Rewriting Equalities (function rewriteEqLHS)  472   473 474 475 476 477 478 479 480 481 482  When rewriting two equalities with the same LHS: (a) (tv ~ xi1) (b) (tv ~ xi2) We have a choice of producing work (xi1 ~ xi2) (up-to the canonicalization invariants) However, to prevent the inert items from getting kicked out of the inerts first, we prefer to canonicalize (xi1 ~ xi2) if (b) comes from the inert set, or (xi2 ~ xi1) if (a) comes from the inert set. This choice is implemented using the WhichComesFromInert flag.  483   simonpj@microsoft.com committed Nov 12, 2010 484 485 486 487 488 Case 2: Functional Dependencies Again, we should prefer, if possible, the inert variables on the RHS Case 3: IP improvement work We must always rewrite so that the inert type is on the right.  489   simonpj@microsoft.com committed Sep 13, 2010 490 491 492 \begin{code} spontaneousSolveStage :: SimplifierStage spontaneousSolveStage workItem inerts  simonpj@microsoft.com committed Nov 12, 2010 493 494  = do { mSolve <- trySpontaneousSolve workItem  simonpj@microsoft.com committed Sep 13, 2010 495  ; case mSolve of  simonpj@microsoft.com committed Nov 12, 2010 496  SPCantSolve -> -- No spontaneous solution for him, keep going  497 498  return $SR { sr_new_work = emptyWorkList , sr_inerts = inerts  simonpj@microsoft.com committed Sep 13, 2010 499 500  , sr_stop = ContinueWith workItem }  simonpj@microsoft.com committed Nov 12, 2010 501  SPSolved workItem'  simonpj@microsoft.com committed Oct 20, 2010 502 503 504 505 506 507 508 509 510 511  | not (isGivenCt workItem) -- Original was wanted or derived but we have now made him -- given so we have to interact him with the inerts due to -- its status change. This in turn may produce more work. -- We do this *right now* (rather than just putting workItem' -- back into the work-list) because we've solved -> do { (new_inert, new_work) <- runSolverPipeline [ ("recursive interact with inert eqs", interactWithInertEqsStage) , ("recursive interact with inerts", interactWithInertsStage) ] inerts workItem'  simonpj@microsoft.com committed Nov 12, 2010 512 513 514  ; return$ SR { sr_new_work = new_work , sr_inerts = new_inert -- will include workItem' , sr_stop = Stop }  simonpj@microsoft.com committed Oct 20, 2010 515  }  516 517 518  | otherwise -> -- Original was given; he must then be inert all right, and -- workList' are all givens from flattening  simonpj@microsoft.com committed Nov 12, 2010 519  return $SR { sr_new_work = emptyWorkList  520 521  , sr_inerts = inerts updInertSet workItem' , sr_stop = Stop }  simonpj@microsoft.com committed Nov 12, 2010 522 523 524 525  SPError -> -- Return with no new work return$ SR { sr_new_work = emptyWorkList , sr_inerts = inerts , sr_stop = Stop }  dimitris@microsoft.com committed Oct 04, 2010 526  }  dimitris@microsoft.com committed Oct 06, 2010 527   simonpj@microsoft.com committed Nov 12, 2010 528 529 530 531 532 533 data SPSolveResult = SPCantSolve | SPSolved WorkItem | SPError -- SPCantSolve means that we can't do the unification because e.g. the variable is untouchable -- SPSolved workItem' gives us a new *given* to go on -- SPError means that it's completely impossible to solve this equality, eg due to a kind error  simonpj@microsoft.com committed Sep 13, 2010 534 -- @trySpontaneousSolve wi@ solves equalities where one side is a  simonpj@microsoft.com committed Nov 12, 2010 535 -- touchable unification variable.  simonpj@microsoft.com committed Sep 13, 2010 536 -- See Note [Touchables and givens]  simonpj@microsoft.com committed Nov 12, 2010 537 538 trySpontaneousSolve :: WorkItem -> TcS SPSolveResult trySpontaneousSolve workItem@(CTyEqCan { cc_id = cv, cc_flavor = gw, cc_tyvar = tv1, cc_rhs = xi })  simonpj@microsoft.com committed Oct 07, 2010 539  | isGiven gw  simonpj@microsoft.com committed Nov 12, 2010 540  = return SPCantSolve  simonpj@microsoft.com committed Sep 13, 2010 541 542 543 544  | Just tv2 <- tcGetTyVar_maybe xi = do { tch1 <- isTouchableMetaTyVar tv1 ; tch2 <- isTouchableMetaTyVar tv2 ; case (tch1, tch2) of  simonpj@microsoft.com committed Nov 12, 2010 545 546 547 548  (True, True) -> trySpontaneousEqTwoWay cv gw tv1 tv2 (True, False) -> trySpontaneousEqOneWay cv gw tv1 xi (False, True) -> trySpontaneousEqOneWay cv gw tv2 (mkTyVarTy tv1) _ -> return SPCantSolve }  simonpj@microsoft.com committed Sep 13, 2010 549 550  | otherwise = do { tch1 <- isTouchableMetaTyVar tv1  simonpj@microsoft.com committed Nov 12, 2010 551  ; if tch1 then trySpontaneousEqOneWay cv gw tv1 xi  552  else do { traceTcS "Untouchable LHS, can't spontaneously solve workitem:" (ppr workItem)  simonpj@microsoft.com committed Nov 12, 2010 553  ; return SPCantSolve }  554  }  simonpj@microsoft.com committed Sep 13, 2010 555 556 557 558  -- No need for -- trySpontaneousSolve (CFunEqCan ...) = ... -- See Note [No touchables as FunEq RHS] in TcSMonad  simonpj@microsoft.com committed Nov 12, 2010 559 trySpontaneousSolve _ = return SPCantSolve  simonpj@microsoft.com committed Sep 13, 2010 560 561  ----------------  simonpj@microsoft.com committed Nov 12, 2010 562 trySpontaneousEqOneWay :: CoVar -> CtFlavor -> TcTyVar -> Xi -> TcS SPSolveResult  563 -- tv is a MetaTyVar, not untouchable  simonpj@microsoft.com committed Nov 12, 2010 564 trySpontaneousEqOneWay cv gw tv xi  565  | not (isSigTyVar tv) || isTyVarTy xi  simonpj@microsoft.com committed Nov 12, 2010 566 567  = do { let kxi = typeKind xi -- NB: 'xi' is fully rewritten according to the inerts -- so we have its more specific kind in our hands  568  ; if kxi isSubKind tyVarKind tv then  simonpj@microsoft.com committed Nov 12, 2010 569  solveWithIdentity cv gw tv xi  570  else if tyVarKind tv isSubKind kxi then  simonpj@microsoft.com committed Nov 12, 2010 571 572 573 574 575 576  return SPCantSolve -- kinds are compatible but we can't solveWithIdentity this way -- This case covers the a_touchable :: * ~ b_untouchable :: ?? -- which has to be deferred or floated out for someone else to solve -- it in a scope where 'b' is no longer untouchable. else do { addErrorTcS KindError gw (mkTyVarTy tv) xi -- See Note [Kind errors] ; return SPError }  577  }  578  | otherwise -- Still can't solve, sig tyvar and non-variable rhs  simonpj@microsoft.com committed Nov 12, 2010 579  = return SPCantSolve  simonpj@microsoft.com committed Sep 13, 2010 580 581  ----------------  simonpj@microsoft.com committed Nov 12, 2010 582 trySpontaneousEqTwoWay :: CoVar -> CtFlavor -> TcTyVar -> TcTyVar -> TcS SPSolveResult  583 -- Both tyvars are *touchable* MetaTyvars so there is only a chance for kind error here  simonpj@microsoft.com committed Nov 12, 2010 584 trySpontaneousEqTwoWay cv gw tv1 tv2  585  | k1 isSubKind k2  simonpj@microsoft.com committed Nov 12, 2010 586  , nicer_to_update_tv2 = solveWithIdentity cv gw tv2 (mkTyVarTy tv1)  587  | k2 isSubKind k1  simonpj@microsoft.com committed Nov 12, 2010 588  = solveWithIdentity cv gw tv1 (mkTyVarTy tv2)  589  | otherwise -- None is a subkind of the other, but they are both touchable!  simonpj@microsoft.com committed Nov 12, 2010 590 591  = do { addErrorTcS KindError gw (mkTyVarTy tv1) (mkTyVarTy tv2) ; return SPError }  simonpj@microsoft.com committed Sep 13, 2010 592 593 594 595 596 597  where k1 = tyVarKind tv1 k2 = tyVarKind tv2 nicer_to_update_tv2 = isSigTyVar tv1 || isSystemName (Var.varName tv2) \end{code}  598 599 600 601 602 603 Note [Kind errors] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider the wanted problem: alpha ~ (# Int, Int #) where alpha :: ?? and (# Int, Int #) :: (#). We can't spontaneously solve this constraint, but we should rather reject the program that give rise to it. If 'trySpontaneousEqTwoWay'  simonpj@microsoft.com committed Nov 12, 2010 604 simply returns @CantSolve@ then that wanted constraint is going to propagate all the way and  605 get quantified over in inference mode. That's bad because we do know at this point that the  simonpj@microsoft.com committed Nov 12, 2010 606 constraint is insoluble. Instead, we call 'recKindErrorTcS' here, which will fail later on.  607 608  The same applies in canonicalization code in case of kind errors in the givens.  609   610 However, when we canonicalize givens we only check for compatibility (@compatKind@).  simonpj@microsoft.com committed Nov 12, 2010 611 If there were a kind error in the givens, this means some form of inconsistency or dead code.  612   simonpj@microsoft.com committed Nov 12, 2010 613 614 615 616 617 You may think that when we spontaneously solve wanteds we may have to look through the bindings to determine the right kind of the RHS type. E.g one may be worried that xi is @alpha@ where alpha :: ? and a previous spontaneous solving has set (alpha := f) with (f :: *). But we orient our constraints so that spontaneously solved ones can rewrite all other constraint so this situation can't happen.  618   619 620 621 622 623 624 625 626 627 628 Note [Spontaneous solving and kind compatibility] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Note that our canonical constraints insist that only *given* equalities (tv ~ xi) or (F xis ~ rhs) require the LHS and the RHS to have exactly the same kinds. - We have to require this because: Given equalities can be freely used to rewrite inside other types or constraints. - We do not have to do the same for wanteds because:  629 630 631 632 633 634 635 636  First, wanted equations (tv ~ xi) where tv is a touchable unification variable may have kinds that do not agree (the kind of xi must be a sub kind of the kind of tv). Second, any potential kind mismatch will result in the constraint not being soluble, which will be reported anyway. This is the reason that @trySpontaneousOneWay@ and @trySpontaneousTwoWay@ will perform a kind compatibility check, and only then will they proceed to @solveWithIdentity@.  637 638 639 640 641 642 643 644 645 646 647  Caveat: - Givens from higher-rank, such as: type family T b :: * -> * -> * type instance T Bool = (->) f :: forall a. ((T a ~ (->)) => ...) -> a -> ... flop = f (...) True Whereas we would be able to apply the type instance, we would not be able to use the given (T Bool ~ (->)) in the body of 'flop'  simonpj@microsoft.com committed Sep 13, 2010 648 649 650 651 652 653 654  Note [Avoid double unifications] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The spontaneous solver has to return a given which mentions the unified unification variable *on the left* of the equality. Here is what happens if not: Original wanted: (a ~ alpha), (alpha ~ Int) We spontaneously solve the first wanted, without changing the order!  simonpj@microsoft.com committed Oct 07, 2010 655  given : a ~ alpha [having unified alpha := a]  simonpj@microsoft.com committed Sep 13, 2010 656 657 658 Now the second wanted comes along, but he cannot rewrite the given, so we simply continue. At the end we spontaneously solve that guy, *reunifying* [alpha := Int]  simonpj@microsoft.com committed Nov 12, 2010 659 We avoid this problem by orienting the resulting given so that the unification  simonpj@microsoft.com committed Oct 07, 2010 660 661 variable is on the left. [Note that alternatively we could attempt to enforce this at canonicalization]  simonpj@microsoft.com committed Sep 13, 2010 662   simonpj@microsoft.com committed Oct 07, 2010 663 664 665 See also Note [No touchables as FunEq RHS] in TcSMonad; avoiding double unifications is the main reason we disallow touchable unification variables as RHS of type family equations: F xis ~ alpha.  simonpj@microsoft.com committed Sep 13, 2010 666 667 668  \begin{code} ----------------  simonpj@microsoft.com committed Nov 12, 2010 669 670  solveWithIdentity :: CoVar -> CtFlavor -> TcTyVar -> Xi -> TcS SPSolveResult  simonpj@microsoft.com committed Sep 13, 2010 671 672 -- Solve with the identity coercion -- Precondition: kind(xi) is a sub-kind of kind(tv)  simonpj@microsoft.com committed Oct 07, 2010 673 674 675 -- Precondition: CtFlavor is Wanted or Derived -- See [New Wanted Superclass Work] to see why solveWithIdentity -- must work for Derived as well as Wanted  simonpj@microsoft.com committed Nov 12, 2010 676 -- Returns: workItem where  677 -- workItem = the new Given constraint  simonpj@microsoft.com committed Nov 12, 2010 678 679 680 solveWithIdentity cv wd tv xi = do { traceTcS "Sneaky unification:" $vcat [text "Coercion variable: " <+> ppr wd,  dimitris@microsoft.com committed Oct 04, 2010 681 682 683  text "Coercion: " <+> pprEq (mkTyVarTy tv) xi, text "Left Kind is : " <+> ppr (typeKind (mkTyVarTy tv)), text "Right Kind is : " <+> ppr (typeKind xi)  simonpj@microsoft.com committed Oct 07, 2010 684  ]  685   simonpj@microsoft.com committed Nov 12, 2010 686 687 688 689 690 691 692 693 694 695 696 697  ; setWantedTyBind tv xi -- Set tv := xi_unflat ; cv_given <- newGivOrDerCoVar (mkTyVarTy tv) xi xi ; case wd of Wanted {} -> setWantedCoBind cv xi Derived {} -> setDerivedCoBind cv xi _ -> pprPanic "Can't spontaneously solve given!" empty ; return$ SPSolved (CTyEqCan { cc_id = cv_given , cc_flavor = mkGivenFlavor wd UnkSkol , cc_tyvar = tv, cc_rhs = xi }) }  simonpj@microsoft.com committed Sep 13, 2010 698 699 700 \end{code}  simonpj@microsoft.com committed Nov 12, 2010 701 702   simonpj@microsoft.com committed Sep 13, 2010 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 ********************************************************************************* * * The interact-with-inert Stage * * ********************************************************************************* \begin{code} -- Interaction result of WorkItem <~> AtomicInert data InteractResult = IR { ir_stop :: StopOrContinue -- Stop -- => Reagent (work item) consumed. -- ContinueWith new_reagent -- => Reagent transformed but keep gathering interactions. -- The transformed item remains inert with respect -- to any previously encountered inerts. , ir_inert_action :: InertAction -- Whether the inert item should remain in the InertSet. , ir_new_work :: WorkList -- new work items to add to the WorkList  725 726  , ir_improvement :: Maybe FDImprovement -- In case improvement kicked in  simonpj@microsoft.com committed Sep 13, 2010 727 728 729  } -- What to do with the inert reactant.  simonpj@microsoft.com committed Nov 12, 2010 730 731 732 data InertAction = KeepInert | DropInert | KeepTransformedInert CanonicalCt -- Keep a slightly transformed inert  simonpj@microsoft.com committed Sep 13, 2010 733 734  mkIRContinue :: Monad m => WorkItem -> InertAction -> WorkList -> m InteractResult  735 mkIRContinue wi keep newWork = return $IR (ContinueWith wi) keep newWork Nothing  simonpj@microsoft.com committed Sep 13, 2010 736 737  mkIRStop :: Monad m => InertAction -> WorkList -> m InteractResult  738 739 740 741 742 mkIRStop keep newWork = return$ IR Stop keep newWork Nothing mkIRStop_RecordImprovement :: Monad m => InertAction -> WorkList -> FDImprovement -> m InteractResult mkIRStop_RecordImprovement keep newWork fdimpr = return $IR Stop keep newWork (Just fdimpr)  simonpj@microsoft.com committed Sep 13, 2010 743 dischargeWorkItem :: Monad m => m InteractResult  744 dischargeWorkItem = mkIRStop KeepInert emptyWorkList  simonpj@microsoft.com committed Sep 13, 2010 745 746  noInteraction :: Monad m => WorkItem -> m InteractResult  747 noInteraction workItem = mkIRContinue workItem KeepInert emptyWorkList  simonpj@microsoft.com committed Sep 13, 2010 748   dimitris@microsoft.com committed Sep 23, 2010 749 data WhichComesFromInert = LeftComesFromInert | RightComesFromInert  simonpj@microsoft.com committed Nov 12, 2010 750  -- See Note [Efficient Orientation]  simonpj@microsoft.com committed Sep 13, 2010 751   752   simonpj@microsoft.com committed Sep 13, 2010 753 ---------------------------------------------------  754 755 756 757 758 759 760 -- Interact a single WorkItem with the equalities of an inert set as far as possible, i.e. until we -- get a Stop result from an individual reaction (i.e. when the WorkItem is consumed), or until we've -- interact the WorkItem with the entire equalities of the InertSet interactWithInertEqsStage :: SimplifierStage interactWithInertEqsStage workItem inert = foldISEqCtsM interactNext initITR inert  simonpj@microsoft.com committed Nov 12, 2010 761  where initITR = SR { sr_inerts = IS { inert_eqs = emptyCCan -- Will fold over equalities  762 763 764 765  , inert_dicts = inert_dicts inert , inert_ips = inert_ips inert , inert_funeqs = inert_funeqs inert , inert_fds = inert_fds inert  766 767 768 769  } , sr_new_work = emptyWorkList , sr_stop = ContinueWith workItem }  simonpj@microsoft.com committed Sep 13, 2010 770   771 772 773 774 --------------------------------------------------- -- Interact a single WorkItem with *non-equality* constraints in the inert set. -- Precondition: equality interactions must have already happened, hence we have -- to pick up some information from the incoming inert, before folding over the  775 776 -- "Other" constraints it contains!  simonpj@microsoft.com committed Sep 13, 2010 777 778 interactWithInertsStage :: SimplifierStage interactWithInertsStage workItem inert  779 780 781 782 783  = let (relevant, inert_residual) = getISRelevant workItem inert initITR = SR { sr_inerts = inert_residual , sr_new_work = emptyWorkList , sr_stop = ContinueWith workItem } in Bag.foldlBagM interactNext initITR relevant  simonpj@microsoft.com committed Sep 13, 2010 784  where  785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800  getISRelevant :: CanonicalCt -> InertSet -> (CanonicalCts, InertSet) getISRelevant (CDictCan { cc_class = cls } ) is = let (relevant, residual_map) = getRelevantCts cls (inert_dicts is) in (relevant, is { inert_dicts = residual_map }) getISRelevant (CFunEqCan { cc_fun = tc } ) is = let (relevant, residual_map) = getRelevantCts tc (inert_funeqs is) in (relevant, is { inert_funeqs = residual_map }) getISRelevant (CIPCan { cc_ip_nm = nm }) is = let (relevant, residual_map) = getRelevantCts nm (inert_ips is) in (relevant, is { inert_ips = residual_map }) -- An equality, finally, may kick everything except equalities out -- because we have already interacted the equalities in interactWithInertEqsStage getISRelevant _eq_ct is -- Equality, everything is relevant for this one -- TODO: if we were caching variables, we'd know that only -- some are relevant. Experiment with this for now. = let cts = cCanMapToBag (inert_ips is) unionBags  simonpj@microsoft.com committed Nov 12, 2010 801 802 803  cCanMapToBag (inert_dicts is) unionBags cCanMapToBag (inert_funeqs is) in (cts, is { inert_dicts = emptyCCanMap , inert_ips = emptyCCanMap  804  , inert_funeqs = emptyCCanMap })  simonpj@microsoft.com committed Sep 13, 2010 805   806 807 808 interactNext :: StageResult -> AtomicInert -> TcS StageResult interactNext it inert | ContinueWith workItem <- sr_stop it  simonpj@microsoft.com committed Oct 19, 2010 809 810 811 812 813  = do { let inerts = sr_inerts it fdimprs_old = getFDImprovements inerts ; ir <- interactWithInert fdimprs_old inert workItem  simonpj@microsoft.com committed Nov 12, 2010 814 815 816 817 818 819 820  -- New inerts depend on whether we KeepInert or not and must -- be updated with FD improvement information from the interaction result (ir) ; let inerts_new = updInertSetFDImprs upd_inert (ir_improvement ir) upd_inert = case ir_inert_action ir of KeepInert -> inerts updInertSet inert DropInert -> inerts KeepTransformedInert inert' -> inerts updInertSet inert'  simonpj@microsoft.com committed Oct 19, 2010 821 822 823 824 825 826  ; return$ SR { sr_inerts = inerts_new , sr_new_work = sr_new_work it unionWorkLists ir_new_work ir , sr_stop = ir_stop ir } } | otherwise = return $it { sr_inerts = (sr_inerts it) updInertSet inert }  simonpj@microsoft.com committed Sep 13, 2010 827 828  -- Do a single interaction of two constraints.  829 830 interactWithInert :: FDImprovements -> AtomicInert -> WorkItem -> TcS InteractResult interactWithInert fdimprs inert workitem  simonpj@microsoft.com committed Sep 13, 2010 831 832 833 834 835  = do { ctxt <- getTcSContext ; let is_allowed = allowedInteraction (simplEqsOnly ctxt) inert workitem inert_ev = cc_id inert work_ev = cc_id workitem  simonpj@microsoft.com committed Nov 12, 2010 836 837 838 839 840  -- Never interact a wanted and a derived where the derived's evidence -- mentions the wanted evidence in an unguarded way. -- See Note [Superclasses and recursive dictionaries] -- and Note [New Wanted Superclass Work] -- We don't have to do this for givens, as we fully know the evidence for them.  simonpj@microsoft.com committed Sep 13, 2010 841 842  ; rec_ev_ok <- case (cc_flavor inert, cc_flavor workitem) of  843 844 845  (Wanted loc, Derived {}) -> isGoodRecEv work_ev (WantedEvVar inert_ev loc) (Derived {}, Wanted loc) -> isGoodRecEv inert_ev (WantedEvVar work_ev loc) _ -> return True  simonpj@microsoft.com committed Sep 13, 2010 846 847  ; if is_allowed && rec_ev_ok then  848  doInteractWithInert fdimprs inert workitem  simonpj@microsoft.com committed Sep 13, 2010 849 850 851 852 853 854 855 856 857 858 859  else noInteraction workitem } allowedInteraction :: Bool -> AtomicInert -> WorkItem -> Bool -- Allowed interactions allowedInteraction eqs_only (CDictCan {}) (CDictCan {}) = not eqs_only allowedInteraction eqs_only (CIPCan {}) (CIPCan {}) = not eqs_only allowedInteraction _ _ _ = True --------------------------------------------  860 doInteractWithInert :: FDImprovements -> CanonicalCt -> CanonicalCt -> TcS InteractResult  simonpj@microsoft.com committed Sep 13, 2010 861 862 -- Identical class constraints.  863 doInteractWithInert fdimprs  simonpj@microsoft.com committed Sep 13, 2010 864  (CDictCan { cc_id = d1, cc_flavor = fl1, cc_class = cls1, cc_tyargs = tys1 })  simonpj@microsoft.com committed Nov 12, 2010 865  workItem@(CDictCan { cc_flavor = fl2, cc_class = cls2, cc_tyargs = tys2 })  simonpj@microsoft.com committed Sep 13, 2010 866 867 868 869 870  | cls1 == cls2 && (and$ zipWith tcEqType tys1 tys2) = solveOneFromTheOther (d1,fl1) workItem | cls1 == cls2 && (not (isGiven fl1 && isGiven fl2)) = -- See Note [When improvement happens]  simonpj@microsoft.com committed Nov 12, 2010 871 872 873 874  do { let pty1 = ClassP cls1 tys1 pty2 = ClassP cls2 tys2 work_item_pred_loc = (pty2, pprFlavorArising fl2) inert_pred_loc = (pty1, pprFlavorArising fl1)  simonpj@microsoft.com committed Sep 13, 2010 875  loc = combineCtLoc fl1 fl2  simonpj@microsoft.com committed Nov 12, 2010 876 877  eqn_pred_locs = improveFromAnother work_item_pred_loc inert_pred_loc -- See Note [Efficient Orientation]  878   simonpj@microsoft.com committed Sep 13, 2010 879  ; wevvars <- mkWantedFunDepEqns loc eqn_pred_locs  simonpj@microsoft.com committed Oct 20, 2010 880  ; fd_work <- canWanteds wevvars  simonpj@microsoft.com committed Sep 13, 2010 881  -- See Note [Generating extra equalities]  882  ; traceTcS "Checking if improvements existed." (ppr fdimprs)  simonpj@microsoft.com committed Oct 20, 2010 883  ; if isEmptyWorkList fd_work || haveBeenImproved fdimprs pty1 pty2 then  884  -- Must keep going  885  mkIRContinue workItem KeepInert fd_work  886 887  else do { traceTcS "Recording improvement and throwing item back in worklist." (ppr (pty1,pty2)) ; mkIRStop_RecordImprovement KeepInert  888  (fd_work unionWorkLists workListFromCCan workItem) (pty1,pty2)  889  }  890  -- See Note [FunDep Reactions]  simonpj@microsoft.com committed Sep 13, 2010 891 892 893 894  } -- Class constraint and given equality: use the equality to rewrite -- the class constraint.  895 896 doInteractWithInert _fdimprs (CTyEqCan { cc_id = cv, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi })  simonpj@microsoft.com committed Sep 13, 2010 897 898 899  (CDictCan { cc_id = dv, cc_flavor = wfl, cc_class = cl, cc_tyargs = xis }) | ifl canRewrite wfl , tv elemVarSet tyVarsOfTypes xis  900 901 902 903 904 905  = if isDerivedSC wfl then mkIRStop KeepInert $emptyWorkList -- See Note [Adding Derived Superclasses] else do { rewritten_dict <- rewriteDict (cv,tv,xi) (dv,wfl,cl,xis) -- Continue with rewritten Dictionary because we can only be in the -- interactWithEqsStage, so the dictionary is inert. ; mkIRContinue rewritten_dict KeepInert emptyWorkList }  simonpj@microsoft.com committed Sep 13, 2010 906   907 908 doInteractWithInert _fdimprs (CDictCan { cc_id = dv, cc_flavor = ifl, cc_class = cl, cc_tyargs = xis })  simonpj@microsoft.com committed Sep 13, 2010 909 910 911  workItem@(CTyEqCan { cc_id = cv, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi }) | wfl canRewrite ifl , tv elemVarSet tyVarsOfTypes xis  912 913 914 915 916  = if isDerivedSC ifl then mkIRContinue workItem DropInert emptyWorkList -- No need to do any rewriting, -- see Note [Adding Derived Superclasses] else do { rewritten_dict <- rewriteDict (cv,tv,xi) (dv,ifl,cl,xis) ; mkIRContinue workItem DropInert (workListFromCCan rewritten_dict) }  simonpj@microsoft.com committed Sep 13, 2010 917 918 919  -- Class constraint and given equality: use the equality to rewrite -- the class constraint.  920 921 doInteractWithInert _fdimprs (CTyEqCan { cc_id = cv, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi })  simonpj@microsoft.com committed Sep 13, 2010 922 923 924 925  (CIPCan { cc_id = ipid, cc_flavor = wfl, cc_ip_nm = nm, cc_ip_ty = ty }) | ifl canRewrite wfl , tv elemVarSet tyVarsOfType ty = do { rewritten_ip <- rewriteIP (cv,tv,xi) (ipid,wfl,nm,ty)  926  ; mkIRContinue rewritten_ip KeepInert emptyWorkList }  simonpj@microsoft.com committed Sep 13, 2010 927   928 929 doInteractWithInert _fdimprs (CIPCan { cc_id = ipid, cc_flavor = ifl, cc_ip_nm = nm, cc_ip_ty = ty })  simonpj@microsoft.com committed Sep 13, 2010 930 931 932 933  workItem@(CTyEqCan { cc_id = cv, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi }) | wfl canRewrite ifl , tv elemVarSet tyVarsOfType ty = do { rewritten_ip <- rewriteIP (cv,tv,xi) (ipid,ifl,nm,ty)  934  ; mkIRContinue workItem DropInert (workListFromCCan rewritten_ip) }  simonpj@microsoft.com committed Sep 13, 2010 935 936 937 938 939 940  -- Two implicit parameter constraints. If the names are the same, -- but their types are not, we generate a wanted type equality -- that equates the type (this is "improvement"). -- However, we don't actually need the coercion evidence, -- so we just generate a fresh coercion variable that isn't used anywhere.  941 942 doInteractWithInert _fdimprs (CIPCan { cc_id = id1, cc_flavor = ifl, cc_ip_nm = nm1, cc_ip_ty = ty1 })  simonpj@microsoft.com committed Sep 13, 2010 943  workItem@(CIPCan { cc_flavor = wfl, cc_ip_nm = nm2, cc_ip_ty = ty2 })  simonpj@microsoft.com committed Sep 17, 2010 944 945 946 947  | nm1 == nm2 && isGiven wfl && isGiven ifl = -- See Note [Overriding implicit parameters] -- Dump the inert item, override totally with the new one -- Do not require type equality  948  mkIRContinue workItem DropInert emptyWorkList  simonpj@microsoft.com committed Sep 17, 2010 949   simonpj@microsoft.com committed Sep 13, 2010 950 951 952  | nm1 == nm2 && ty1 tcEqType ty2 = solveOneFromTheOther (id1,ifl) workItem  simonpj@microsoft.com committed Sep 17, 2010 953  | nm1 == nm2  simonpj@microsoft.com committed Sep 13, 2010 954  = -- See Note [When improvement happens]  simonpj@microsoft.com committed Nov 12, 2010 955  do { co_var <- newWantedCoVar ty2 ty1 -- See Note [Efficient Orientation]  simonpj@microsoft.com committed Sep 13, 2010 956  ; let flav = Wanted (combineCtLoc ifl wfl)  957  ; cans <- mkCanonical flav co_var  simonpj@microsoft.com committed Oct 20, 2010 958  ; mkIRContinue workItem KeepInert cans }  simonpj@microsoft.com committed Sep 13, 2010 959 960 961 962  -- Never rewrite a given with a wanted equality, and a type function  simonpj@microsoft.com committed Nov 12, 2010 963 964 965 -- equality can never rewrite an equality. We rewrite LHS *and* RHS -- of function equalities so that our inert set exposes everything that -- we know about equalities.  simonpj@microsoft.com committed Sep 13, 2010 966   simonpj@microsoft.com committed Nov 12, 2010 967 -- Inert: equality, work item: function equality  968 969 doInteractWithInert _fdimprs (CTyEqCan { cc_id = cv1, cc_flavor = ifl, cc_tyvar = tv, cc_rhs = xi1 })  simonpj@microsoft.com committed Sep 13, 2010 970 971 972  (CFunEqCan { cc_id = cv2, cc_flavor = wfl, cc_fun = tc , cc_tyargs = args, cc_rhs = xi2 }) | ifl canRewrite wfl  simonpj@microsoft.com committed Nov 12, 2010 973  , tv elemVarSet tyVarsOfTypes (xi2:args) -- Rewrite RHS as well  simonpj@microsoft.com committed Sep 13, 2010 974  = do { rewritten_funeq <- rewriteFunEq (cv1,tv,xi1) (cv2,wfl,tc,args,xi2)  975  ; mkIRStop KeepInert (workListFromCCan rewritten_funeq) }  simonpj@microsoft.com committed Nov 12, 2010 976  -- Must Stop here, because we may no longer be inert after the rewritting.  simonpj@microsoft.com committed Sep 13, 2010 977 978  -- Inert: function equality, work item: equality  979 980 doInteractWithInert _fdimprs (CFunEqCan {cc_id = cv1, cc_flavor = ifl, cc_fun = tc  simonpj@microsoft.com committed Sep 13, 2010 981 982 983  , cc_tyargs = args, cc_rhs = xi1 }) workItem@(CTyEqCan { cc_id = cv2, cc_flavor = wfl, cc_tyvar = tv, cc_rhs = xi2 }) | wfl canRewrite ifl  simonpj@microsoft.com committed Nov 12, 2010 984  , tv elemVarSet tyVarsOfTypes (xi1:args) -- Rewrite RHS as well  simonpj@microsoft.com committed Sep 13, 2010 985  = do { rewritten_funeq <- rewriteFunEq (cv2,tv,xi2) (cv1,ifl,tc,args,xi1)  986  ; mkIRContinue workItem DropInert (workListFromCCan rewritten_funeq) }  simonpj@microsoft.com committed Nov 12, 2010 987 988 989 990 991 992 993 994  -- One may think that we could (KeepTransformedInert rewritten_funeq) -- but that is wrong, because it may end up not being inert with respect -- to future inerts. Example: -- Original inert = { F xis ~ [a], b ~ Maybe Int } -- Work item comes along = a ~ [b] -- If we keep { F xis ~ [b] } in the inert set we will end up with: -- { F xis ~ [b], b ~ Maybe Int, a ~ [Maybe Int] } -- At the end, which is *not* inert. So we should unfortunately DropInert here.  simonpj@microsoft.com committed Sep 13, 2010 995   996 997 doInteractWithInert _fdimprs (CFunEqCan { cc_id = cv1, cc_flavor = fl1, cc_fun = tc1  simonpj@microsoft.com committed Sep 13, 2010 998 999 1000  , cc_tyargs = args1, cc_rhs = xi1 }) workItem@(CFunEqCan { cc_id = cv2, cc_flavor = fl2, cc_fun = tc2 , cc_tyargs = args2, cc_rhs = xi2 })  1001  | fl1 canSolve fl2 && lhss_match  dimitris@microsoft.com committed Sep 23, 2010 1002  = do { cans <- rewriteEqLHS LeftComesFromInert (mkCoVarCoercion cv1,xi1) (cv2,fl2,xi2)  simonpj@microsoft.com committed Oct 20, 2010 1003  ; mkIRStop KeepInert cans }  1004  | fl2 canSolve fl1 && lhss_match  dimitris@microsoft.com committed Sep 23, 2010 1005  = do { cans <- rewriteEqLHS RightComesFromInert (mkCoVarCoercion cv2,xi2) (cv1,fl1,xi1)  simonpj@microsoft.com committed Oct 20, 2010 1006  ; mkIRContinue workItem DropInert cans }  simonpj@microsoft.com committed Sep 13, 2010 1007 1008 1009  where lhss_match = tc1 == tc2 && and (zipWith tcEqType args1 args2)  1010 doInteractWithInert _fdimprs  simonpj@microsoft.com committed Nov 12, 2010 1011  (CTyEqCan { cc_id = cv1, cc_flavor = fl1, cc_tyvar = tv1, cc_rhs = xi1 })  simonpj@microsoft.com committed Sep 13, 2010 1012 1013  workItem@(CTyEqCan { cc_id = cv2, cc_flavor = fl2, cc_tyvar = tv2, cc_rhs = xi2 }) -- Check for matching LHS  1014  | fl1 canSolve fl2 && tv1 == tv2  dimitris@microsoft.com committed Sep 23, 2010 1015  = do { cans <- rewriteEqLHS LeftComesFromInert (mkCoVarCoercion cv1,xi1) (cv2,fl2,xi2)  simonpj@microsoft.com committed Oct 20, 2010 1016  ; mkIRStop KeepInert cans }  simonpj@microsoft.com committed Sep 13, 2010 1017   1018  | fl2 canSolve fl1 && tv1 == tv2  dimitris@microsoft.com committed Sep 23, 2010 1019  = do { cans <- rewriteEqLHS RightComesFromInert (mkCoVarCoercion cv2,xi2) (cv1,fl1,xi1)  simonpj@microsoft.com committed Nov 12, 2010 1020  ; mkIRContinue workItem DropInert cans }  simonpj@microsoft.com committed Sep 13, 2010 1021 1022 1023 -- Check for rewriting RHS | fl1 canRewrite fl2 && tv1 elemVarSet tyVarsOfType xi2 = do { rewritten_eq <- rewriteEqRHS (cv1,tv1,xi1) (cv2,fl2,tv2,xi2)  simonpj@microsoft.com committed Oct 20, 2010 1024  ; mkIRStop KeepInert rewritten_eq }  simonpj@microsoft.com committed Sep 13, 2010 1025 1026  | fl2 canRewrite fl1 && tv2 elemVarSet tyVarsOfType xi1 = do { rewritten_eq <- rewriteEqRHS (cv2,tv2,xi2) (cv1,fl1,tv1,xi1)  simonpj@microsoft.com committed Oct 20, 2010 1027  ; mkIRContinue workItem DropInert rewritten_eq }  dimitris@microsoft.com committed Oct 06, 2010 1028 1029  -- Fall-through case for all other situations  1030 doInteractWithInert _fdimprs _ workItem = noInteraction workItem  simonpj@microsoft.com committed Sep 13, 2010 1031   simonpj@microsoft.com committed Oct 08, 2010 1032 -------------------------  simonpj@microsoft.com committed Sep 13, 2010 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 -- Equational Rewriting rewriteDict :: (CoVar, TcTyVar, Xi) -> (DictId, CtFlavor, Class, [Xi]) -> TcS CanonicalCt rewriteDict (cv,tv,xi) (dv,gw,cl,xis) = do { let cos = substTysWith [tv] [mkCoVarCoercion cv] xis -- xis[tv] ~ xis[xi] args = substTysWith [tv] [xi] xis con = classTyCon cl dict_co = mkTyConCoercion con cos ; dv' <- newDictVar cl args ; case gw of Wanted {} -> setDictBind dv (EvCast dv' (mkSymCoercion dict_co)) _given_or_derived -> setDictBind dv' (EvCast dv dict_co) ; return (CDictCan { cc_id = dv' , cc_flavor = gw , cc_class = cl , cc_tyargs = args }) } rewriteIP :: (CoVar,TcTyVar,Xi) -> (EvVar,CtFlavor, IPName Name, TcType) -> TcS CanonicalCt rewriteIP (cv,tv,xi) (ipid,gw,nm,ty) = do { let ip_co = substTyWith [tv] [mkCoVarCoercion cv] ty -- ty[tv] ~ t[xi] ty' = substTyWith [tv] [xi] ty ; ipid' <- newIPVar nm ty' ; case gw of Wanted {} -> setIPBind ipid (EvCast ipid' (mkSymCoercion ip_co)) _given_or_derived -> setIPBind ipid' (EvCast ipid ip_co) ; return (CIPCan { cc_id = ipid' , cc_flavor = gw , cc_ip_nm = nm , cc_ip_ty = ty' }) } rewriteFunEq :: (CoVar,TcTyVar,Xi) -> (CoVar,CtFlavor,TyCon, [Xi], Xi) -> TcS CanonicalCt  simonpj@microsoft.com committed Nov 12, 2010 1063 rewriteFunEq (cv1,tv,xi1) (cv2,gw, tc,args,xi2) -- cv2 :: F args ~ xi2  simonpj@microsoft.com committed Sep 13, 2010 1064 1065  = do { let arg_cos = substTysWith [tv] [mkCoVarCoercion cv1] args args' = substTysWith [tv] [xi1] args  simonpj@microsoft.com committed Nov 12, 2010 1066 1067 1068 1069  fun_co = mkTyConCoercion tc arg_cos -- fun_co :: F args ~ F args' xi2' = substTyWith [tv] [xi1] xi2 xi2_co = substTyWith [tv] [mkCoVarCoercion cv1] xi2 -- xi2_co :: xi2 ~ xi2'  simonpj@microsoft.com committed Sep 13, 2010 1070  ; cv2' <- case gw of  simonpj@microsoft.com committed Nov 12, 2010 1071  Wanted {} -> do { cv2' <- newWantedCoVar (mkTyConApp tc args') xi2'  simonpj@microsoft.com committed Sep 13, 2010 1072  ; setWantedCoBind cv2$  simonpj@microsoft.com committed Nov 12, 2010 1073 1074  fun_co mkTransCoercion mkCoVarCoercion cv2' mkTransCoercion mkSymCoercion xi2_co  simonpj@microsoft.com committed Sep 13, 2010 1075  ; return cv2' }  simonpj@microsoft.com committed Nov 12, 2010 1076 1077 1078  _giv_or_der -> newGivOrDerCoVar (mkTyConApp tc args') xi2' $mkSymCoercion fun_co mkTransCoercion mkCoVarCoercion cv2 mkTransCoercion xi2_co  simonpj@microsoft.com committed Sep 13, 2010 1079 1080 1081 1082  ; return (CFunEqCan { cc_id = cv2' , cc_flavor = gw , cc_tyargs = args' , cc_fun = tc  simonpj@microsoft.com committed Nov 12, 2010 1083  , cc_rhs = xi2' }) }  simonpj@microsoft.com committed Sep 13, 2010 1084 1085   simonpj@microsoft.com committed Oct 20, 2010 1086 rewriteEqRHS :: (CoVar,TcTyVar,Xi) -> (CoVar,CtFlavor,TcTyVar,Xi) -> TcS WorkList  simonpj@microsoft.com committed Sep 13, 2010 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 -- Use the first equality to rewrite the second, flavors already checked. -- E.g. c1 : tv1 ~ xi1 c2 : tv2 ~ xi2 -- rewrites c2 to give -- c2' : tv2 ~ xi2[xi1/tv1] -- We must do an occurs check to sure the new constraint is canonical -- So we might return an empty bag rewriteEqRHS (cv1,tv1,xi1) (cv2,gw,tv2,xi2) | Just tv2' <- tcGetTyVar_maybe xi2' , tv2 == tv2' -- In this case xi2[xi1/tv1] = tv2, so we have tv2~tv2 = do { when (isWanted gw) (setWantedCoBind cv2 (mkSymCoercion co2')) ; return emptyCCan }  simonpj@microsoft.com committed Nov 12, 2010 1098 1099 1100 1101 1102 1103  | otherwise = do { cv2' <- case gw of Wanted {} -> do { cv2' <- newWantedCoVar (mkTyVarTy tv2) xi2' ; setWantedCoBind cv2$  simonpj@microsoft.com committed Sep 13, 2010 1104  mkCoVarCoercion cv2' mkTransCoercion mkSymCoercion co2'  simonpj@microsoft.com committed Nov 12, 2010 1105  ; return cv2' }  simonpj@microsoft.com committed Sep 13, 2010 1106 1107 1108 1109  _giv_or_der -> newGivOrDerCoVar (mkTyVarTy tv2) xi2' $mkCoVarCoercion cv2 mkTransCoercion co2'  simonpj@microsoft.com committed Nov 12, 2010 1110  ; canEq gw cv2' (mkTyVarTy tv2) xi2'  simonpj@microsoft.com committed Sep 13, 2010 1111 1112 1113 1114 1115  } where xi2' = substTyWith [tv1] [xi1] xi2 co2' = substTyWith [tv1] [mkCoVarCoercion cv1] xi2 -- xi2 ~ xi2[xi1/tv1]  dimitris@microsoft.com committed Sep 23, 2010 1116   simonpj@microsoft.com committed Oct 20, 2010 1117 rewriteEqLHS :: WhichComesFromInert -> (Coercion,Xi) -> (CoVar,CtFlavor,Xi) -> TcS WorkList  1118 -- Used to ineract two equalities of the following form:  simonpj@microsoft.com committed Sep 13, 2010 1119 1120 -- First Equality: co1: (XXX ~ xi1) -- Second Equality: cv2: (XXX ~ xi2)  1121 -- Where the cv1 canSolve cv2 equality  1122 1123 -- We have an option of creating new work (xi1 ~ xi2) OR (xi2 ~ xi1), -- See Note [Efficient Orientation] for that  dimitris@microsoft.com committed Sep 23, 2010 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 rewriteEqLHS which (co1,xi1) (cv2,gw,xi2) = do { cv2' <- case (isWanted gw, which) of (True,LeftComesFromInert) -> do { cv2' <- newWantedCoVar xi2 xi1 ; setWantedCoBind cv2$ co1 mkTransCoercion mkSymCoercion (mkCoVarCoercion cv2') ; return cv2' } (True,RightComesFromInert) -> do { cv2' <- newWantedCoVar xi1 xi2 ; setWantedCoBind cv2 $co1 mkTransCoercion mkCoVarCoercion cv2' ; return cv2' } (False,LeftComesFromInert) -> newGivOrDerCoVar xi2 xi1$ mkSymCoercion (mkCoVarCoercion cv2) mkTransCoercion co1 (False,RightComesFromInert) -> newGivOrDerCoVar xi1 xi2 $mkSymCoercion co1 mkTransCoercion mkCoVarCoercion cv2  1142 1143 1144  ; mkCanonical gw cv2' }  simonpj@microsoft.com committed Sep 13, 2010 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 solveOneFromTheOther :: (EvVar, CtFlavor) -> CanonicalCt -> TcS InteractResult -- First argument inert, second argument workitem. They both represent -- wanted/given/derived evidence for the *same* predicate so we try here to -- discharge one directly from the other. -- -- Precondition: value evidence only (implicit parameters, classes) -- not coercion solveOneFromTheOther (iid,ifl) workItem -- Both derived needs a special case. You might think that we do not need -- two evidence terms for the same claim. But, since the evidence is partial, -- either evidence may do in some cases; see TcSMonad.isGoodRecEv. -- See also Example 3 in Note [Superclasses and recursive dictionaries] | isDerived ifl && isDerived wfl = noInteraction workItem  1160  | ifl canSolve wfl  simonpj@microsoft.com committed Sep 17, 2010 1161 1162 1163 1164 1165  = do { unless (isGiven wfl)$ setEvBind wid (EvId iid) -- Overwrite the binding, if one exists -- For Givens, which are lambda-bound, nothing to overwrite, ; dischargeWorkItem }  1166  | otherwise -- wfl canSolve ifl  simonpj@microsoft.com committed Sep 13, 2010 1167  = do { unless (isGiven ifl) \$ setEvBind iid (EvId wid)  1168  ; mkIRContinue workItem DropInert emptyWorkList }  simonpj@microsoft.com committed Sep 13, 2010 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311