{-# LANGUAGE CPP, ViewPatterns, BangPatterns #-} module TcFlatten( FlattenMode(..), flatten, flattenKind, flattenArgsNom, unflattenWanteds ) where #include "HsVersions.h" import GhcPrelude import TcRnTypes import TcType import Type import TcEvidence import TyCon import TyCoRep -- performs delicate algorithm on types import Coercion import Var import VarSet import VarEnv import Outputable import TcSMonad as TcS import BasicTypes( SwapFlag(..) ) import Util import Bag import Control.Monad import MonadUtils ( zipWith3M ) import Control.Arrow ( first ) {- Note [The flattening story] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ * A CFunEqCan is either of form [G] <F xis> : F xis ~ fsk -- fsk is a FlatSkolTv [W] x : F xis ~ fmv -- fmv is a FlatMetaTv where x is the witness variable xis are function-free fsk/fmv is a flatten skolem; it is always untouchable (level 0) * CFunEqCans can have any flavour: [G], [W], [WD] or [D] * KEY INSIGHTS: - A given flatten-skolem, fsk, is known a-priori to be equal to F xis (the LHS), with <F xis> evidence. The fsk is still a unification variable, but it is "owned" by its CFunEqCan, and is filled in (unflattened) only by unflattenGivens. - A unification flatten-skolem, fmv, stands for the as-yet-unknown type to which (F xis) will eventually reduce. It is filled in - All fsk/fmv variables are "untouchable". To make it simple to test, we simply give them TcLevel=0. This means that in a CTyVarEq, say, fmv ~ Int we NEVER unify fmv. - A unification flatten-skolem, fmv, ONLY gets unified when either a) The CFunEqCan takes a step, using an axiom b) By unflattenWanteds They are never unified in any other form of equality. For example [W] ffmv ~ Int is stuck; it does not unify with fmv. * We *never* substitute in the RHS (i.e. the fsk/fmv) of a CFunEqCan. That would destroy the invariant about the shape of a CFunEqCan, and it would risk wanted/wanted interactions. The only way we learn information about fsk is when the CFunEqCan takes a step. However we *do* substitute in the LHS of a CFunEqCan (else it would never get to fire!) * Unflattening: - We unflatten Givens when leaving their scope (see unflattenGivens) - We unflatten Wanteds at the end of each attempt to simplify the wanteds; see unflattenWanteds, called from solveSimpleWanteds. * Ownership of fsk/fmv. Each canonical [G], [W], or [WD] CFunEqCan x : F xis ~ fsk/fmv "owns" a distinct evidence variable x, and flatten-skolem fsk/fmv. Why? We make a fresh fsk/fmv when the constraint is born; and we never rewrite the RHS of a CFunEqCan. In contrast a [D] CFunEqCan /shares/ its fmv with its partner [W], but does not "own" it. If we reduce a [D] F Int ~ fmv, where say type instance F Int = ty, then we don't discharge fmv := ty. Rather we simply generate [D] fmv ~ ty (in TcInteract.reduce_top_fun_eq, and dischargeFmv) * Inert set invariant: if F xis1 ~ fsk1, F xis2 ~ fsk2 then xis1 /= xis2 i.e. at most one CFunEqCan with a particular LHS * Function applications can occur in the RHS of a CTyEqCan. No reason not allow this, and it reduces the amount of flattening that must occur. * Flattening a type (F xis): - If we are flattening in a Wanted/Derived constraint then create new [W] x : F xis ~ fmv else create new [G] x : F xis ~ fsk with fresh evidence variable x and flatten-skolem fsk/fmv - Add it to the work list - Replace (F xis) with fsk/fmv in the type you are flattening - You can also add the CFunEqCan to the "flat cache", which simply keeps track of all the function applications you have flattened. - If (F xis) is in the cache already, just use its fsk/fmv and evidence x, and emit nothing. - No need to substitute in the flat-cache. It's not the end of the world if we start with, say (F alpha ~ fmv1) and (F Int ~ fmv2) and then find alpha := Int. Athat will simply give rise to fmv1 := fmv2 via [Interacting rule] below * Canonicalising a CFunEqCan [G/W] x : F xis ~ fsk/fmv - Flatten xis (to substitute any tyvars; there are already no functions) cos :: xis ~ flat_xis - New wanted x2 :: F flat_xis ~ fsk/fmv - Add new wanted to flat cache - Discharge x = F cos ; x2 * [Interacting rule] (inert) [W] x1 : F tys ~ fmv1 (work item) [W] x2 : F tys ~ fmv2 Just solve one from the other: x2 := x1 fmv2 := fmv1 This just unites the two fsks into one. Always solve given from wanted if poss. * For top-level reductions, see Note [Top-level reductions for type functions] in TcInteract Why given-fsks, alone, doesn't work ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Could we get away with only flatten meta-tyvars, with no flatten-skolems? No. [W] w : alpha ~ [F alpha Int] ---> flatten w = ...w'... [W] w' : alpha ~ [fsk] [G] <F alpha Int> : F alpha Int ~ fsk --> unify (no occurs check) alpha := [fsk] But since fsk = F alpha Int, this is really an occurs check error. If that is all we know about alpha, we will succeed in constraint solving, producing a program with an infinite type. Even if we did finally get (g : fsk ~ Bool) by solving (F alpha Int ~ fsk) using axiom, zonking would not see it, so (x::alpha) sitting in the tree will get zonked to an infinite type. (Zonking always only does refl stuff.) Why flatten-meta-vars, alone doesn't work ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Look at Simple13, with unification-fmvs only [G] g : a ~ [F a] ---> Flatten given g' = g;[x] [G] g' : a ~ [fmv] [W] x : F a ~ fmv --> subst a in x g' = g;[x] x = F g' ; x2 [W] x2 : F [fmv] ~ fmv And now we have an evidence cycle between g' and x! If we used a given instead (ie current story) [G] g : a ~ [F a] ---> Flatten given g' = g;[x] [G] g' : a ~ [fsk] [G] <F a> : F a ~ fsk ---> Substitute for a [G] g' : a ~ [fsk] [G] F (sym g'); <F a> : F [fsk] ~ fsk Why is it right to treat fmv's differently to ordinary unification vars? ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ f :: forall a. a -> a -> Bool g :: F Int -> F Int -> Bool Consider f (x:Int) (y:Bool) This gives alpha~Int, alpha~Bool. There is an inconsistency, but really only one error. SherLoc may tell you which location is most likely, based on other occurrences of alpha. Consider g (x:Int) (y:Bool) Here we get (F Int ~ Int, F Int ~ Bool), which flattens to (fmv ~ Int, fmv ~ Bool) But there are really TWO separate errors. ** We must not complain about Int~Bool. ** Moreover these two errors could arise in entirely unrelated parts of the code. (In the alpha case, there must be *some* connection (eg v:alpha in common envt).) Note [Unflattening can force the solver to iterate] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Look at Trac #10340: type family Any :: * -- No instances get :: MonadState s m => m s instance MonadState s (State s) where ... foo :: State Any Any foo = get For 'foo' we instantiate 'get' at types mm ss [WD] MonadState ss mm, [WD] mm ss ~ State Any Any Flatten, and decompose [WD] MonadState ss mm, [WD] Any ~ fmv [WD] mm ~ State fmv, [WD] fmv ~ ss Unify mm := State fmv: [WD] MonadState ss (State fmv) [WD] Any ~ fmv, [WD] fmv ~ ss Now we are stuck; the instance does not match!! So unflatten: fmv := Any ss := Any (*) [WD] MonadState Any (State Any) The unification (*) represents progress, so we must do a second round of solving; this time it succeeds. This is done by the 'go' loop in solveSimpleWanteds. This story does not feel right but it's the best I can do; and the iteration only happens in pretty obscure circumstances. ************************************************************************ * * * Examples Here is a long series of examples I had to work through * * ************************************************************************ Simple20 ~~~~~~~~ axiom F [a] = [F a] [G] F [a] ~ a --> [G] fsk ~ a [G] [F a] ~ fsk (nc) --> [G] F a ~ fsk2 [G] fsk ~ [fsk2] [G] fsk ~ a --> [G] F a ~ fsk2 [G] a ~ [fsk2] [G] fsk ~ a ---------------------------------------- indexed-types/should_compile/T44984 [W] H (F Bool) ~ H alpha [W] alpha ~ F Bool --> F Bool ~ fmv0 H fmv0 ~ fmv1 H alpha ~ fmv2 fmv1 ~ fmv2 fmv0 ~ alpha flatten ~~~~~~~ fmv0 := F Bool fmv1 := H (F Bool) fmv2 := H alpha alpha := F Bool plus fmv1 ~ fmv2 But these two are equal under the above assumptions. Solve by Refl. --- under plan B, namely solve fmv1:=fmv2 eagerly --- [W] H (F Bool) ~ H alpha [W] alpha ~ F Bool --> F Bool ~ fmv0 H fmv0 ~ fmv1 H alpha ~ fmv2 fmv1 ~ fmv2 fmv0 ~ alpha --> F Bool ~ fmv0 H fmv0 ~ fmv1 H alpha ~ fmv2 fmv2 := fmv1 fmv0 ~ alpha flatten fmv0 := F Bool fmv1 := H fmv0 = H (F Bool) retain H alpha ~ fmv2 because fmv2 has been filled alpha := F Bool ---------------------------- indexed-types/should_failt/T4179 after solving [W] fmv_1 ~ fmv_2 [W] A3 (FCon x) ~ fmv_1 (CFunEqCan) [W] A3 (x (aoa -> fmv_2)) ~ fmv_2 (CFunEqCan) ---------------------------------------- indexed-types/should_fail/T7729a a) [W] BasePrimMonad (Rand m) ~ m1 b) [W] tt m1 ~ BasePrimMonad (Rand m) ---> process (b) first BasePrimMonad (Ramd m) ~ fmv_atH fmv_atH ~ tt m1 ---> now process (a) m1 ~ s_atH ~ tt m1 -- An obscure occurs check ---------------------------------------- typecheck/TcTypeNatSimple Original constraint [W] x + y ~ x + alpha (non-canonical) ==> [W] x + y ~ fmv1 (CFunEqCan) [W] x + alpha ~ fmv2 (CFuneqCan) [W] fmv1 ~ fmv2 (CTyEqCan) (sigh) ---------------------------------------- indexed-types/should_fail/GADTwrong1 [G] Const a ~ () ==> flatten [G] fsk ~ () work item: Const a ~ fsk ==> fire top rule [G] fsk ~ () work item fsk ~ () Surely the work item should rewrite to () ~ ()? Well, maybe not; it'a very special case. More generally, our givens look like F a ~ Int, where (F a) is not reducible. ---------------------------------------- indexed_types/should_fail/T8227: Why using a different can-rewrite rule in CFunEqCan heads does not work. Assuming NOT rewriting wanteds with wanteds Inert: [W] fsk_aBh ~ fmv_aBk -> fmv_aBk [W] fmv_aBk ~ fsk_aBh [G] Scalar fsk_aBg ~ fsk_aBh [G] V a ~ f_aBg Worklist includes [W] Scalar fmv_aBi ~ fmv_aBk fmv_aBi, fmv_aBk are flatten unification variables Work item: [W] V fsk_aBh ~ fmv_aBi Note that the inert wanteds are cyclic, because we do not rewrite wanteds with wanteds. Then we go into a loop when normalise the work-item, because we use rewriteOrSame on the argument of V. Conclusion: Don't make canRewrite context specific; instead use [W] a ~ ty to rewrite a wanted iff 'a' is a unification variable. ---------------------------------------- Here is a somewhat similar case: type family G a :: * blah :: (G a ~ Bool, Eq (G a)) => a -> a blah = error "urk" foo x = blah x For foo we get [W] Eq (G a), G a ~ Bool Flattening [W] G a ~ fmv, Eq fmv, fmv ~ Bool We can't simplify away the Eq Bool unless we substitute for fmv. Maybe that doesn't matter: we would still be left with unsolved G a ~ Bool. -------------------------- Trac #9318 has a very simple program leading to [W] F Int ~ Int [W] F Int ~ Bool We don't want to get "Error Int~Bool". But if fmv's can rewrite wanteds, we will [W] fmv ~ Int [W] fmv ~ Bool ---> [W] Int ~ Bool ************************************************************************ * * * FlattenEnv & FlatM * The flattening environment & monad * * ************************************************************************ -} type FlatWorkListRef = TcRef [Ct] -- See Note [The flattening work list] data FlattenEnv = FE { fe_mode :: !FlattenMode , fe_loc :: !CtLoc -- See Note [Flattener CtLoc] , fe_flavour :: !CtFlavour , fe_eq_rel :: !EqRel -- See Note [Flattener EqRels] , fe_work :: !FlatWorkListRef } -- See Note [The flattening work list] data FlattenMode -- Postcondition for all three: inert wrt the type substitution = FM_FlattenAll -- Postcondition: function-free | FM_SubstOnly -- See Note [Flattening under a forall] -- | FM_Avoid TcTyVar Bool -- See Note [Lazy flattening] -- -- Postcondition: -- -- * tyvar is only mentioned in result under a rigid path -- -- e.g. [a] is ok, but F a won't happen -- -- * If flat_top is True, top level is not a function application -- -- (but under type constructors is ok e.g. [F a]) instance Outputable FlattenMode where ppr FM_FlattenAll = text "FM_FlattenAll" ppr FM_SubstOnly = text "FM_SubstOnly" eqFlattenMode :: FlattenMode -> FlattenMode -> Bool eqFlattenMode FM_FlattenAll FM_FlattenAll = True eqFlattenMode FM_SubstOnly FM_SubstOnly = True -- FM_Avoid tv1 b1 `eq` FM_Avoid tv2 b2 = tv1 == tv2 && b1 == b2 eqFlattenMode _ _ = False -- | The 'FlatM' monad is a wrapper around 'TcS' with the following -- extra capabilities: (1) it offers access to a 'FlattenEnv'; -- and (2) it maintains the flattening worklist. -- See Note [The flattening work list]. newtype FlatM a = FlatM { runFlatM :: FlattenEnv -> TcS a } instance Monad FlatM where m >>= k = FlatM $ \env -> do { a <- runFlatM m env ; runFlatM (k a) env } instance Functor FlatM where fmap = liftM instance Applicative FlatM where pure x = FlatM $ const (pure x) (<*>) = ap liftTcS :: TcS a -> FlatM a liftTcS thing_inside = FlatM $ const thing_inside emitFlatWork :: Ct -> FlatM () -- See Note [The flattening work list] emitFlatWork ct = FlatM $ \env -> updTcRef (fe_work env) (ct :) -- convenient wrapper when you have a CtEvidence describing -- the flattening operation runFlattenCtEv :: FlattenMode -> CtEvidence -> FlatM a -> TcS a runFlattenCtEv mode ev = runFlatten mode (ctEvLoc ev) (ctEvFlavour ev) (ctEvEqRel ev) -- Run thing_inside (which does flattening), and put all -- the work it generates onto the main work list -- See Note [The flattening work list] runFlatten :: FlattenMode -> CtLoc -> CtFlavour -> EqRel -> FlatM a -> TcS a runFlatten mode loc flav eq_rel thing_inside = do { flat_ref <- newTcRef [] ; let fmode = FE { fe_mode = mode , fe_loc = loc , fe_flavour = flav , fe_eq_rel = eq_rel , fe_work = flat_ref } ; res <- runFlatM thing_inside fmode ; new_flats <- readTcRef flat_ref ; updWorkListTcS (add_flats new_flats) ; return res } where add_flats new_flats wl = wl { wl_funeqs = add_funeqs new_flats (wl_funeqs wl) } add_funeqs [] wl = wl add_funeqs (f:fs) wl = add_funeqs fs (f:wl) -- add_funeqs fs ws = reverse fs ++ ws -- e.g. add_funeqs [f1,f2,f3] [w1,w2,w3,w4] -- = [f3,f2,f1,w1,w2,w3,w4] traceFlat :: String -> SDoc -> FlatM () traceFlat herald doc = liftTcS $ traceTcS herald doc getFlatEnvField :: (FlattenEnv -> a) -> FlatM a getFlatEnvField accessor = FlatM $ \env -> return (accessor env) getEqRel :: FlatM EqRel getEqRel = getFlatEnvField fe_eq_rel getRole :: FlatM Role getRole = eqRelRole <$> getEqRel getFlavour :: FlatM CtFlavour getFlavour = getFlatEnvField fe_flavour getFlavourRole :: FlatM CtFlavourRole getFlavourRole = do { flavour <- getFlavour ; eq_rel <- getEqRel ; return (flavour, eq_rel) } getMode :: FlatM FlattenMode getMode = getFlatEnvField fe_mode getLoc :: FlatM CtLoc getLoc = getFlatEnvField fe_loc checkStackDepth :: Type -> FlatM () checkStackDepth ty = do { loc <- getLoc ; liftTcS $ checkReductionDepth loc ty } -- | Change the 'EqRel' in a 'FlatM'. setEqRel :: EqRel -> FlatM a -> FlatM a setEqRel new_eq_rel thing_inside = FlatM $ \env -> if new_eq_rel == fe_eq_rel env then runFlatM thing_inside env else runFlatM thing_inside (env { fe_eq_rel = new_eq_rel }) -- | Change the 'FlattenMode' in a 'FlattenEnv'. setMode :: FlattenMode -> FlatM a -> FlatM a setMode new_mode thing_inside = FlatM $ \env -> if new_mode `eqFlattenMode` fe_mode env then runFlatM thing_inside env else runFlatM thing_inside (env { fe_mode = new_mode }) -- | Make sure that flattening actually produces a coercion (in other -- words, make sure our flavour is not Derived) -- Note [No derived kind equalities] noBogusCoercions :: FlatM a -> FlatM a noBogusCoercions thing_inside = FlatM $ \env -> -- No new thunk is made if the flavour hasn't changed (note the bang). let !env' = case fe_flavour env of Derived -> env { fe_flavour = Wanted WDeriv } _ -> env in runFlatM thing_inside env' bumpDepth :: FlatM a -> FlatM a bumpDepth (FlatM thing_inside) = FlatM $ \env -> do -- bumpDepth can be called a lot during flattening so we force the -- new env to avoid accumulating thunks. { let !env' = env { fe_loc = bumpCtLocDepth (fe_loc env) } ; thing_inside env' } {- Note [The flattening work list] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The "flattening work list", held in the fe_work field of FlattenEnv, is a list of CFunEqCans generated during flattening. The key idea is this. Consider flattening (Eq (F (G Int) (H Bool)): * The flattener recursively calls itself on sub-terms before building the main term, so it will encounter the terms in order G Int H Bool F (G Int) (H Bool) flattening to sub-goals w1: G Int ~ fuv0 w2: H Bool ~ fuv1 w3: F fuv0 fuv1 ~ fuv2 * Processing w3 first is BAD, because we can't reduce i t,so it'll get put into the inert set, and later kicked out when w1, w2 are solved. In Trac #9872 this led to inert sets containing hundreds of suspended calls. * So we want to process w1, w2 first. * So you might think that we should just use a FIFO deque for the work-list, so that putting adding goals in order w1,w2,w3 would mean we processed w1 first. * BUT suppose we have 'type instance G Int = H Char'. Then processing w1 leads to a new goal w4: H Char ~ fuv0 We do NOT want to put that on the far end of a deque! Instead we want to put it at the *front* of the work-list so that we continue to work on it. So the work-list structure is this: * The wl_funeqs (in TcS) is a LIFO stack; we push new goals (such as w4) on top (extendWorkListFunEq), and take new work from the top (selectWorkItem). * When flattening, emitFlatWork pushes new flattening goals (like w1,w2,w3) onto the flattening work list, fe_work, another push-down stack. * When we finish flattening, we *reverse* the fe_work stack onto the wl_funeqs stack (which brings w1 to the top). The function runFlatten initialises the fe_work stack, and reverses it onto wl_fun_eqs at the end. Note [Flattener EqRels] ~~~~~~~~~~~~~~~~~~~~~~~ When flattening, we need to know which equality relation -- nominal or representation -- we should be respecting. The only difference is that we rewrite variables by representational equalities when fe_eq_rel is ReprEq, and that we unwrap newtypes when flattening w.r.t. representational equality. Note [Flattener CtLoc] ~~~~~~~~~~~~~~~~~~~~~~ The flattener does eager type-family reduction. Type families might loop, and we don't want GHC to do so. A natural solution is to have a bounded depth to these processes. A central difficulty is that such a solution isn't quite compositional. For example, say it takes F Int 10 steps to get to Bool. How many steps does it take to get from F Int -> F Int to Bool -> Bool? 10? 20? What about getting from Const Char (F Int) to Char? 11? 1? Hard to know and hard to track. So, we punt, essentially. We store a CtLoc in the FlattenEnv and just update the environment when recurring. In the TyConApp case, where there may be multiple type families to flatten, we just copy the current CtLoc into each branch. If any branch hits the stack limit, then the whole thing fails. A consequence of this is that setting the stack limits appropriately will be essentially impossible. So, the official recommendation if a stack limit is hit is to disable the check entirely. Otherwise, there will be baffling, unpredictable errors. Note [Lazy flattening] ~~~~~~~~~~~~~~~~~~~~~~ The idea of FM_Avoid mode is to flatten less aggressively. If we have a ~ [F Int] there seems to be no great merit in lifting out (F Int). But if it was a ~ [G a Int] then we *do* want to lift it out, in case (G a Int) reduces to Bool, say, which gets rid of the occurs-check problem. (For the flat_top Bool, see comments above and at call sites.) HOWEVER, the lazy flattening actually seems to make type inference go *slower*, not faster. perf/compiler/T3064 is a case in point; it gets *dramatically* worse with FM_Avoid. I think it may be because floating the types out means we normalise them, and that often makes them smaller and perhaps allows more re-use of previously solved goals. But to be honest I'm not absolutely certain, so I am leaving FM_Avoid in the code base. What I'm removing is the unique place where it is *used*, namely in TcCanonical.canEqTyVar. See also Note [Conservative unification check] in TcUnify, which gives other examples where lazy flattening caused problems. Bottom line: FM_Avoid is unused for now (Nov 14). Note: T5321Fun got faster when I disabled FM_Avoid T5837 did too, but it's pathalogical anyway Note [Phantoms in the flattener] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Suppose we have data Proxy p = Proxy and we're flattening (Proxy ty) w.r.t. ReprEq. Then, we know that `ty` is really irrelevant -- it will be ignored when solving for representational equality later on. So, we omit flattening `ty` entirely. This may violate the expectation of "xi"s for a bit, but the canonicaliser will soon throw out the phantoms when decomposing a TyConApp. (Or, the canonicaliser will emit an insoluble, in which case the unflattened version yields a better error message anyway.) Note [No derived kind equalities] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ A kind-level coercion can appear in types, via mkCastTy. So, whenever we are generating a coercion in a dependent context (in other words, in a kind) we need to make sure that our flavour is never Derived (as Derived constraints have no evidence). The noBogusCoercions function changes the flavour from Derived just for this purpose. -} {- ********************************************************************* * * * Externally callable flattening functions * * * * They are all wrapped in runFlatten, so their * * flattening work gets put into the work list * * * ********************************************************************* -} flatten :: FlattenMode -> CtEvidence -> TcType -> TcS (Xi, TcCoercion) flatten mode ev ty = do { traceTcS "flatten {" (ppr mode <+> ppr ty) ; (ty', co) <- runFlattenCtEv mode ev (flatten_one ty) ; traceTcS "flatten }" (ppr ty') ; return (ty', co) } -- specialized to flattening kinds: never Derived, always Nominal -- See Note [No derived kind equalities] flattenKind :: CtLoc -> CtFlavour -> TcType -> TcS (Xi, TcCoercionN) flattenKind loc flav ty = do { traceTcS "flattenKind {" (ppr flav <+> ppr ty) ; let flav' = case flav of Derived -> Wanted WDeriv -- the WDeriv/WOnly choice matters not _ -> flav ; (ty', co) <- runFlatten FM_FlattenAll loc flav' NomEq (flatten_one ty) ; traceTcS "flattenKind }" (ppr ty' $$ ppr co) -- co is never a panic ; return (ty', co) } flattenArgsNom :: CtEvidence -> TyCon -> [TcType] -> TcS ([Xi], [TcCoercion], TcCoercionN) -- Externally-callable, hence runFlatten -- Flatten a vector of types all at once; in fact they are -- always the arguments of type family or class, so -- ctEvFlavour ev = Nominal -- and we want to flatten all at nominal role -- The kind passed in is the kind of the type family or class, call it T -- The last coercion returned has type (tcTypeKind(T xis) ~N tcTypeKind(T tys)) -- -- For Derived constraints the returned coercion may be undefined -- because flattening may use a Derived equality ([D] a ~ ty) flattenArgsNom ev tc tys = do { traceTcS "flatten_args {" (vcat (map ppr tys)) ; (tys', cos, kind_co) <- runFlattenCtEv FM_FlattenAll ev (flatten_args_tc tc (repeat Nominal) tys) ; traceTcS "flatten }" (vcat (map ppr tys')) ; return (tys', cos, kind_co) } {- ********************************************************************* * * * The main flattening functions * * ********************************************************************* -} {- Note [Flattening] ~~~~~~~~~~~~~~~~~~~~ flatten ty ==> (xi, co) where xi has no type functions, unless they appear under ForAlls has no skolems that are mapped in the inert set has no filled-in metavariables co :: xi ~ ty Key invariants: (F0) co :: xi ~ zonk(ty) (F1) tcTypeKind(xi) succeeds and returns a fully zonked kind (F2) tcTypeKind(xi) `eqType` zonk(tcTypeKind(ty)) Note that it is flatten's job to flatten *every type function it sees*. flatten is only called on *arguments* to type functions, by canEqGiven. Flattening also: * zonks, removing any metavariables, and * applies the substitution embodied in the inert set Because flattening zonks and the returned coercion ("co" above) is also zonked, it's possible that (co :: xi ~ ty) isn't quite true. So, instead, we can rely on this fact: (F1) tcTypeKind(xi) succeeds and returns a fully zonked kind Note that the left-hand type of co is *always* precisely xi. The right-hand type may or may not be ty, however: if ty has unzonked filled-in metavariables, then the right-hand type of co will be the zonked version of ty. It is for this reason that we occasionally have to explicitly zonk, when (co :: xi ~ ty) is important even before we zonk the whole program. For example, see the FTRNotFollowed case in flattenTyVar. Why have these invariants on flattening? Because we sometimes use tcTypeKind during canonicalisation, and we want this kind to be zonked (e.g., see TcCanonical.canEqTyVar). Flattening is always homogeneous. That is, the kind of the result of flattening is always the same as the kind of the input, modulo zonking. More formally: (F2) tcTypeKind(xi) `eqType` zonk(tcTypeKind(ty)) This invariant means that the kind of a flattened type might not itself be flat. Recall that in comments we use alpha[flat = ty] to represent a flattening skolem variable alpha which has been generated to stand in for ty. ----- Example of flattening a constraint: ------ flatten (List (F (G Int))) ==> (xi, cc) where xi = List alpha cc = { G Int ~ beta[flat = G Int], F beta ~ alpha[flat = F beta] } Here * alpha and beta are 'flattening skolem variables'. * All the constraints in cc are 'given', and all their coercion terms are the identity. NB: Flattening Skolems only occur in canonical constraints, which are never zonked, so we don't need to worry about zonking doing accidental unflattening. Note that we prefer to leave type synonyms unexpanded when possible, so when the flattener encounters one, it first asks whether its transitive expansion contains any type function applications. If so, it expands the synonym and proceeds; if not, it simply returns the unexpanded synonym. Note [flatten_args performance] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ In programs with lots of type-level evaluation, flatten_args becomes part of a tight loop. For example, see test perf/compiler/T9872a, which calls flatten_args a whopping 7,106,808 times. It is thus important that flatten_args be efficient. Performance testing showed that the current implementation is indeed efficient. It's critically important that zipWithAndUnzipM be specialized to TcS, and it's also quite helpful to actually `inline` it. On test T9872a, here are the allocation stats (Dec 16, 2014): * Unspecialized, uninlined: 8,472,613,440 bytes allocated in the heap * Specialized, uninlined: 6,639,253,488 bytes allocated in the heap * Specialized, inlined: 6,281,539,792 bytes allocated in the heap To improve performance even further, flatten_args_nom is split off from flatten_args, as nominal equality is the common case. This would be natural to write using mapAndUnzipM, but even inlined, that function is not as performant as a hand-written loop. * mapAndUnzipM, inlined: 7,463,047,432 bytes allocated in the heap * hand-written recursion: 5,848,602,848 bytes allocated in the heap If you make any change here, pay close attention to the T9872{a,b,c} tests and T5321Fun. If we need to make this yet more performant, a possible way forward is to duplicate the flattener code for the nominal case, and make that case faster. This doesn't seem quite worth it, yet. Note [flatten_exact_fam_app_fully performance] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The refactor of GRefl seems to cause performance trouble for T9872x: the allocation of flatten_exact_fam_app_fully_performance increased. See note [Generalized reflexive coercion] in TyCoRep for more information about GRefl and Trac #15192 for the current state. The explicit pattern match in homogenise_result helps with T9872a, b, c. Still, it increases the expected allocation of T9872d by ~2%. TODO: a step-by-step replay of the refactor to analyze the performance. -} {-# INLINE flatten_args_tc #-} flatten_args_tc :: TyCon -- T -> [Role] -- Role r -> [Type] -- Arg types [t1,..,tn] -> FlatM ( [Xi] -- List of flattened args [x1,..,xn] -- 1-1 corresp with [t1,..,tn] , [Coercion] -- List of arg coercions [co1,..,con] -- 1-1 corresp with [t1,..,tn] -- coi :: xi ~r ti , CoercionN) -- Result coercion, rco -- rco : (T t1..tn) ~N (T (x1 |> co1) .. (xn |> con)) flatten_args_tc tc = flatten_args all_bndrs any_named_bndrs inner_ki emptyVarSet -- NB: TyCon kinds are always closed where (bndrs, named) = ty_con_binders_ty_binders' (tyConBinders tc) -- it's possible that the result kind has arrows (for, e.g., a type family) -- so we must split it (inner_bndrs, inner_ki, inner_named) = split_pi_tys' (tyConResKind tc) !all_bndrs = bndrs `chkAppend` inner_bndrs !any_named_bndrs = named || inner_named -- NB: Those bangs there drop allocations in T9872{a,c,d} by 8%. {-# INLINE flatten_args #-} flatten_args :: [TyCoBinder] -> Bool -- Binders, and True iff any of them are -- named. -> Kind -> TcTyCoVarSet -- function kind; kind's free vars -> [Role] -> [Type] -- these are in 1-to-1 correspondence -> FlatM ([Xi], [Coercion], CoercionN) -- Coercions :: Xi ~ Type, at roles given -- Third coercion :: tcTypeKind(fun xis) ~N tcTypeKind(fun tys) -- That is, the third coercion relates the kind of some function (whose kind is -- passed as the first parameter) instantiated at xis to the kind of that -- function instantiated at the tys. This is useful in keeping flattening -- homoegeneous. The list of roles must be at least as long as the list of -- types. flatten_args orig_binders any_named_bndrs orig_inner_ki orig_fvs orig_roles orig_tys = if any_named_bndrs then flatten_args_slow orig_binders orig_inner_ki orig_fvs orig_roles orig_tys else flatten_args_fast orig_binders orig_inner_ki orig_roles orig_tys {-# INLINE flatten_args_fast #-} -- | fast path flatten_args, in which none of the binders are named and -- therefore we can avoid tracking a lifting context. -- There are many bang patterns in here. It's been observed that they -- greatly improve performance of an optimized build. -- The T9872 test cases are good witnesses of this fact. flatten_args_fast :: [TyCoBinder] -> Kind -> [Role] -> [Type] -> FlatM ([Xi], [Coercion], CoercionN) flatten_args_fast orig_binders orig_inner_ki orig_roles orig_tys = fmap finish (iterate orig_tys orig_roles orig_binders) where iterate :: [Type] -> [Role] -> [TyCoBinder] -> FlatM ([Xi], [Coercion], [TyCoBinder]) iterate (ty:tys) (role:roles) (_:binders) = do (xi, co) <- go role ty (xis, cos, binders) <- iterate tys roles binders pure (xi : xis, co : cos, binders) iterate [] _ binders = pure ([], [], binders) iterate _ _ _ = pprPanic "flatten_args wandered into deeper water than usual" (vcat []) -- This debug information is commented out because leaving it in -- causes a ~2% increase in allocations in T9872{a,c,d}. {- (vcat [ppr orig_binders, ppr orig_inner_ki, ppr (take 10 orig_roles), -- often infinite! ppr orig_tys]) -} {-# INLINE go #-} go :: Role -> Type -> FlatM (Xi, Coercion) go role ty = case role of -- In the slow path we bind the Xi and Coercion from the recursive -- call and then use it such -- -- let kind_co = mkTcSymCo $ mkReflCo Nominal (tyBinderType binder) -- casted_xi = xi `mkCastTy` kind_co -- casted_co = xi |> kind_co ~r xi ; co -- -- but this isn't necessary: -- mkTcSymCo (Refl a b) = Refl a b, -- mkCastTy x (Refl _ _) = x -- mkTcGReflLeftCo _ ty (Refl _ _) `mkTransCo` co = co -- -- Also, no need to check isAnonTyCoBinder or isNamedBinder, since -- we've already established that they're all anonymous. Nominal -> setEqRel NomEq $ flatten_one ty Representational -> setEqRel ReprEq $ flatten_one ty Phantom -> -- See Note [Phantoms in the flattener] do { ty <- liftTcS $ zonkTcType ty ; return (ty, mkReflCo Phantom ty) } {-# INLINE finish #-} finish :: ([Xi], [Coercion], [TyCoBinder]) -> ([Xi], [Coercion], CoercionN) finish (xis, cos, binders) = (xis, cos, kind_co) where final_kind = mkPiTys binders orig_inner_ki kind_co = mkNomReflCo final_kind {-# INLINE flatten_args_slow #-} -- | Slow path, compared to flatten_args_fast, because this one must track -- a lifting context. flatten_args_slow :: [TyCoBinder] -> Kind -> TcTyCoVarSet -> [Role] -> [Type] -> FlatM ([Xi], [Coercion], CoercionN) flatten_args_slow binders inner_ki fvs roles tys -- Arguments used dependently must be flattened with proper coercions, but -- we're not guaranteed to get a proper coercion when flattening with the -- "Derived" flavour. So we must call noBogusCoercions when flattening arguments -- corresponding to binders that are dependent. However, we might legitimately -- have *more* arguments than binders, in the case that the inner_ki is a variable -- that gets instantiated with a Π-type. We conservatively choose not to produce -- bogus coercions for these, too. Note that this might miss an opportunity for -- a Derived rewriting a Derived. The solution would be to generate evidence for -- Deriveds, thus avoiding this whole noBogusCoercions idea. See also -- Note [No derived kind equalities] = do { flattened_args <- zipWith3M fl (map isNamedBinder binders ++ repeat True) roles tys ; return (simplifyArgsWorker binders inner_ki fvs roles flattened_args) } where {-# INLINE fl #-} fl :: Bool -- must we ensure to produce a real coercion here? -- see comment at top of function -> Role -> Type -> FlatM (Xi, Coercion) fl True r ty = noBogusCoercions $ fl1 r ty fl False r ty = fl1 r ty {-# INLINE fl1 #-} fl1 :: Role -> Type -> FlatM (Xi, Coercion) fl1 Nominal ty = setEqRel NomEq $ flatten_one ty fl1 Representational ty = setEqRel ReprEq $ flatten_one ty fl1 Phantom ty -- See Note [Phantoms in the flattener] = do { ty <- liftTcS $ zonkTcType ty ; return (ty, mkReflCo Phantom ty) } ------------------ flatten_one :: TcType -> FlatM (Xi, Coercion) -- Flatten a type to get rid of type function applications, returning -- the new type-function-free type, and a collection of new equality -- constraints. See Note [Flattening] for more detail. -- -- Postcondition: Coercion :: Xi ~ TcType -- The role on the result coercion matches the EqRel in the FlattenEnv flatten_one xi@(LitTy {}) = do { role <- getRole ; return (xi, mkReflCo role xi) } flatten_one (TyVarTy tv) = flattenTyVar tv flatten_one (AppTy ty1 ty2) = flatten_app_tys ty1 [ty2] flatten_one (TyConApp tc tys) -- Expand type synonyms that mention type families -- on the RHS; see Note [Flattening synonyms] | Just (tenv, rhs, tys') <- expandSynTyCon_maybe tc tys , let expanded_ty = mkAppTys (substTy (mkTvSubstPrs tenv) rhs) tys' = do { mode <- getMode ; case mode of FM_FlattenAll | not (isFamFreeTyCon tc) -> flatten_one expanded_ty _ -> flatten_ty_con_app tc tys } -- Otherwise, it's a type function application, and we have to -- flatten it away as well, and generate a new given equality constraint -- between the application and a newly generated flattening skolem variable. | isTypeFamilyTyCon tc = flatten_fam_app tc tys -- For * a normal data type application -- * data family application -- we just recursively flatten the arguments. | otherwise -- FM_Avoid stuff commented out; see Note [Lazy flattening] -- , let fmode' = case fmode of -- Switch off the flat_top bit in FM_Avoid -- FE { fe_mode = FM_Avoid tv _ } -- -> fmode { fe_mode = FM_Avoid tv False } -- _ -> fmode = flatten_ty_con_app tc tys flatten_one (FunTy ty1 ty2) = do { (xi1,co1) <- flatten_one ty1 ; (xi2,co2) <- flatten_one ty2 ; role <- getRole ; return (mkFunTy xi1 xi2, mkFunCo role co1 co2) } flatten_one ty@(ForAllTy {}) -- TODO (RAE): This is inadequate, as it doesn't flatten the kind of -- the bound tyvar. Doing so will require carrying around a substitution -- and the usual substTyVarBndr-like silliness. Argh. -- We allow for-alls when, but only when, no type function -- applications inside the forall involve the bound type variables. = do { let (bndrs, rho) = tcSplitForAllVarBndrs ty tvs = binderVars bndrs ; (rho', co) <- setMode FM_SubstOnly $ flatten_one rho -- Substitute only under a forall -- See Note [Flattening under a forall] ; return (mkForAllTys bndrs rho', mkHomoForAllCos tvs co) } flatten_one (CastTy ty g) = do { (xi, co) <- flatten_one ty ; (g', _) <- flatten_co g ; role <- getRole ; return (mkCastTy xi g', castCoercionKind co role xi ty g' g) } flatten_one (CoercionTy co) = first mkCoercionTy <$> flatten_co co -- | "Flatten" a coercion. Really, just zonk it so we can uphold -- (F1) of Note [Flattening] flatten_co :: Coercion -> FlatM (Coercion, Coercion) flatten_co co = do { co <- liftTcS $ zonkCo co ; env_role <- getRole ; let co' = mkTcReflCo env_role (mkCoercionTy co) ; return (co, co') } -- flatten (nested) AppTys flatten_app_tys :: Type -> [Type] -> FlatM (Xi, Coercion) -- commoning up nested applications allows us to look up the function's kind -- only once. Without commoning up like this, we would spend a quadratic amount -- of time looking up functions' types flatten_app_tys (AppTy ty1 ty2) tys = flatten_app_tys ty1 (ty2:tys) flatten_app_tys fun_ty arg_tys = do { (fun_xi, fun_co) <- flatten_one fun_ty ; flatten_app_ty_args fun_xi fun_co arg_tys } -- Given a flattened function (with the coercion produced by flattening) and -- a bunch of unflattened arguments, flatten the arguments and apply. -- The coercion argument's role matches the role stored in the FlatM monad. -- -- The bang patterns used here were observed to improve performance. If you -- wish to remove them, be sure to check for regeressions in allocations. flatten_app_ty_args :: Xi -> Coercion -> [Type] -> FlatM (Xi, Coercion) flatten_app_ty_args fun_xi fun_co [] -- this will be a common case when called from flatten_fam_app, so shortcut = return (fun_xi, fun_co) flatten_app_ty_args fun_xi fun_co arg_tys = do { (xi, co, kind_co) <- case tcSplitTyConApp_maybe fun_xi of Just (tc, xis) -> do { let tc_roles = tyConRolesRepresentational tc arg_roles = dropList xis tc_roles ; (arg_xis, arg_cos, kind_co) <- flatten_vector (tcTypeKind fun_xi) arg_roles arg_tys -- Here, we have fun_co :: T xi1 xi2 ~ ty -- and we need to apply fun_co to the arg_cos. The problem is -- that using mkAppCo is wrong because that function expects -- its second coercion to be Nominal, and the arg_cos might -- not be. The solution is to use transitivity: -- T <xi1> <xi2> arg_cos ;; fun_co <arg_tys> ; eq_rel <- getEqRel ; let app_xi = mkTyConApp tc (xis ++ arg_xis) app_co = case eq_rel of NomEq -> mkAppCos fun_co arg_cos ReprEq -> mkTcTyConAppCo Representational tc (zipWith mkReflCo tc_roles xis ++ arg_cos) `mkTcTransCo` mkAppCos fun_co (map mkNomReflCo arg_tys) ; return (app_xi, app_co, kind_co) } Nothing -> do { (arg_xis, arg_cos, kind_co) <- flatten_vector (tcTypeKind fun_xi) (repeat Nominal) arg_tys ; let arg_xi = mkAppTys fun_xi arg_xis arg_co = mkAppCos fun_co arg_cos ; return (arg_xi, arg_co, kind_co) } ; role <- getRole ; return (homogenise_result xi co role kind_co) } flatten_ty_con_app :: TyCon -> [TcType] -> FlatM (Xi, Coercion) flatten_ty_con_app tc tys = do { role <- getRole ; (xis, cos, kind_co) <- flatten_args_tc tc (tyConRolesX role tc) tys ; let tyconapp_xi = mkTyConApp tc xis tyconapp_co = mkTyConAppCo role tc cos ; return (homogenise_result tyconapp_xi tyconapp_co role kind_co) } -- Make the result of flattening homogeneous (Note [Flattening] (F2)) homogenise_result :: Xi -- a flattened type -> Coercion -- :: xi ~r original ty -> Role -- r -> CoercionN -- kind_co :: tcTypeKind(xi) ~N tcTypeKind(ty) -> (Xi, Coercion) -- (xi |> kind_co, (xi |> kind_co) -- ~r original ty) homogenise_result xi co r kind_co -- the explicit pattern match here improves the performance of T9872a, b, c by -- ~2% | isGReflCo kind_co = (xi `mkCastTy` kind_co, co) | otherwise = (xi `mkCastTy` kind_co , (mkSymCo $ GRefl r xi (MCo kind_co)) `mkTransCo` co) {-# INLINE homogenise_result #-} -- Flatten a vector (list of arguments). flatten_vector :: Kind -- of the function being applied to these arguments -> [Role] -- If we're flatten w.r.t. ReprEq, what roles do the -- args have? -> [Type] -- the args to flatten -> FlatM ([Xi], [Coercion], CoercionN) flatten_vector ki roles tys = do { eq_rel <- getEqRel ; case eq_rel of NomEq -> flatten_args bndrs any_named_bndrs inner_ki fvs (repeat Nominal) tys ReprEq -> flatten_args bndrs any_named_bndrs inner_ki fvs roles tys } where (bndrs, inner_ki, any_named_bndrs) = split_pi_tys' ki fvs = tyCoVarsOfType ki {-# INLINE flatten_vector #-} {- Note [Flattening synonyms] ~~~~~~~~~~~~~~~~~~~~~~~~~~ Not expanding synonyms aggressively improves error messages, and keeps types smaller. But we need to take care. Suppose type T a = a -> a and we want to flatten the type (T (F a)). Then we can safely flatten the (F a) to a skolem, and return (T fsk). We don't need to expand the synonym. This works because TcTyConAppCo can deal with synonyms (unlike TyConAppCo), see Note [TcCoercions] in TcEvidence. But (Trac #8979) for type T a = (F a, a) where F is a type function we must expand the synonym in (say) T Int, to expose the type function to the flattener. Note [Flattening under a forall] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Under a forall, we (a) MUST apply the inert substitution (b) MUST NOT flatten type family applications Hence FMSubstOnly. For (a) consider c ~ a, a ~ T (forall b. (b, [c])) If we don't apply the c~a substitution to the second constraint we won't see the occurs-check error. For (b) consider (a ~ forall b. F a b), we don't want to flatten to (a ~ forall b.fsk, F a b ~ fsk) because now the 'b' has escaped its scope. We'd have to flatten to (a ~ forall b. fsk b, forall b. F a b ~ fsk b) and we have not begun to think about how to make that work! ************************************************************************ * * Flattening a type-family application * * ************************************************************************ -} flatten_fam_app :: TyCon -> [TcType] -> FlatM (Xi, Coercion) -- flatten_fam_app can be over-saturated -- flatten_exact_fam_app is exactly saturated -- flatten_exact_fam_app_fully lifts out the application to top level -- Postcondition: Coercion :: Xi ~ F tys flatten_fam_app tc tys -- Can be over-saturated = ASSERT2( tys `lengthAtLeast` tyConArity tc , ppr tc $$ ppr (tyConArity tc) $$ ppr tys) do { mode <- getMode ; case mode of { FM_SubstOnly -> flatten_ty_con_app tc tys ; FM_FlattenAll -> -- Type functions are saturated -- The type function might be *over* saturated -- in which case the remaining arguments should -- be dealt with by AppTys do { let (tys1, tys_rest) = splitAt (tyConArity tc) tys ; (xi1, co1) <- flatten_exact_fam_app_fully tc tys1 -- co1 :: xi1 ~ F tys1 ; flatten_app_ty_args xi1 co1 tys_rest } } } -- the [TcType] exactly saturate the TyCon -- See note [flatten_exact_fam_app_fully performance] flatten_exact_fam_app_fully :: TyCon -> [TcType] -> FlatM (Xi, Coercion) flatten_exact_fam_app_fully tc tys -- See Note [Reduce type family applications eagerly] -- the following tcTypeKind should never be evaluated, as it's just used in -- casting, and casts by refl are dropped = do { let reduce_co = mkNomReflCo (tcTypeKind (mkTyConApp tc tys)) ; mOut <- try_to_reduce_nocache tc tys reduce_co id ; case mOut of Just out -> pure out Nothing -> do { -- First, flatten the arguments ; (xis, cos, kind_co) <- setEqRel NomEq $ -- just do this once, instead of for -- each arg flatten_args_tc tc (repeat Nominal) tys -- kind_co :: tcTypeKind(F xis) ~N tcTypeKind(F tys) ; eq_rel <- getEqRel ; cur_flav <- getFlavour ; let role = eqRelRole eq_rel ret_co = mkTyConAppCo role tc cos -- ret_co :: F xis ~ F tys; might be heterogeneous -- Now, look in the cache ; mb_ct <- liftTcS $ lookupFlatCache tc xis ; case mb_ct of Just (co, rhs_ty, flav) -- co :: F xis ~ fsk -- flav is [G] or [WD] -- See Note [Type family equations] in TcSMonad | (NotSwapped, _) <- flav `funEqCanDischargeF` cur_flav -> -- Usable hit in the flat-cache do { traceFlat "flatten/flat-cache hit" $ (ppr tc <+> ppr xis $$ ppr rhs_ty) ; (fsk_xi, fsk_co) <- flatten_one rhs_ty -- The fsk may already have been unified, so -- flatten it -- fsk_co :: fsk_xi ~ fsk ; let xi = fsk_xi `mkCastTy` kind_co co' = mkTcCoherenceLeftCo role fsk_xi kind_co fsk_co `mkTransCo` maybeSubCo eq_rel (mkSymCo co) `mkTransCo` ret_co ; return (xi, co') } -- :: fsk_xi ~ F xis -- Try to reduce the family application right now -- See Note [Reduce type family applications eagerly] _ -> do { mOut <- try_to_reduce tc xis kind_co (`mkTransCo` ret_co) ; case mOut of Just out -> pure out Nothing -> do { loc <- getLoc ; (ev, co, fsk) <- liftTcS $ newFlattenSkolem cur_flav loc tc xis -- The new constraint (F xis ~ fsk) is not -- necessarily inert (e.g. the LHS may be a -- redex) so we must put it in the work list ; let ct = CFunEqCan { cc_ev = ev , cc_fun = tc , cc_tyargs = xis , cc_fsk = fsk } ; emitFlatWork ct ; traceFlat "flatten/flat-cache miss" $ (ppr tc <+> ppr xis $$ ppr fsk $$ ppr ev) -- NB: fsk's kind is already flattened because -- the xis are flattened ; let fsk_ty = mkTyVarTy fsk xi = fsk_ty `mkCastTy` kind_co co' = mkTcCoherenceLeftCo role fsk_ty kind_co (maybeSubCo eq_rel (mkSymCo co)) `mkTransCo` ret_co ; return (xi, co') } } } } where -- try_to_reduce and try_to_reduce_nocache (below) could be unified into -- a more general definition, but it was observed that separating them -- gives better performance (lower allocation numbers in T9872x). try_to_reduce :: TyCon -- F, family tycon -> [Type] -- args, not necessarily flattened -> CoercionN -- kind_co :: tcTypeKind(F args) ~N -- tcTypeKind(F orig_args) -- where -- orig_args is what was passed to the outer -- function -> ( Coercion -- :: (xi |> kind_co) ~ F args -> Coercion ) -- what to return from outer function -> FlatM (Maybe (Xi, Coercion)) try_to_reduce tc tys kind_co update_co = do { checkStackDepth (mkTyConApp tc tys) ; mb_match <- liftTcS $ matchFam tc tys ; case mb_match of -- NB: norm_co will always be homogeneous. All type families -- are homogeneous. Just (norm_co, norm_ty) -> do { traceFlat "Eager T.F. reduction success" $ vcat [ ppr tc, ppr tys, ppr norm_ty , ppr norm_co <+> dcolon <+> ppr (coercionKind norm_co) ] ; (xi, final_co) <- bumpDepth $ flatten_one norm_ty ; eq_rel <- getEqRel ; let co = maybeSubCo eq_rel norm_co `mkTransCo` mkSymCo final_co ; flavour <- getFlavour -- NB: only extend cache with nominal equalities ; when (eq_rel == NomEq) $ liftTcS $ extendFlatCache tc tys ( co, xi, flavour ) ; let role = eqRelRole eq_rel xi' = xi `mkCastTy` kind_co co' = update_co $ mkTcCoherenceLeftCo role xi kind_co (mkSymCo co) ; return $ Just (xi', co') } Nothing -> pure Nothing } try_to_reduce_nocache :: TyCon -- F, family tycon -> [Type] -- args, not necessarily flattened -> CoercionN -- kind_co :: tcTypeKind(F args) -- ~N tcTypeKind(F orig_args) -- where -- orig_args is what was passed to the -- outer function -> ( Coercion -- :: (xi |> kind_co) ~ F args -> Coercion ) -- what to return from outer -- function -> FlatM (Maybe (Xi, Coercion)) try_to_reduce_nocache tc tys kind_co update_co = do { checkStackDepth (mkTyConApp tc tys) ; mb_match <- liftTcS $ matchFam tc tys ; case mb_match of -- NB: norm_co will always be homogeneous. All type families -- are homogeneous. Just (norm_co, norm_ty) -> do { (xi, final_co) <- bumpDepth $ flatten_one norm_ty ; eq_rel <- getEqRel ; let co = maybeSubCo eq_rel norm_co `mkTransCo` mkSymCo final_co role = eqRelRole eq_rel xi' = xi `mkCastTy` kind_co co' = update_co $ mkTcCoherenceLeftCo role xi kind_co (mkSymCo co) ; return $ Just (xi', co') } Nothing -> pure Nothing } {- Note [Reduce type family applications eagerly] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If we come across a type-family application like (Append (Cons x Nil) t), then, rather than flattening to a skolem etc, we may as well just reduce it on the spot to (Cons x t). This saves a lot of intermediate steps. Examples that are helped are tests T9872, and T5321Fun. Performance testing indicates that it's best to try this *twice*, once before flattening arguments and once after flattening arguments. Adding the extra reduction attempt before flattening arguments cut the allocation amounts for the T9872{a,b,c} tests by half. An example of where the early reduction appears helpful: type family Last x where Last '[x] = x Last (h ': t) = Last t workitem: (x ~ Last '[1,2,3,4,5,6]) Flattening the argument never gets us anywhere, but trying to flatten it at every step is quadratic in the length of the list. Reducing more eagerly makes simplifying the right-hand type linear in its length. Testing also indicated that the early reduction should *not* use the flat-cache, but that the later reduction *should*. (Although the effect was not large.) Hence the Bool argument to try_to_reduce. To me (SLPJ) this seems odd; I get that eager reduction usually succeeds; and if don't use the cache for eager reduction, we will miss most of the opportunities for using it at all. More exploration would be good here. At the end, once we've got a flat rhs, we extend the flatten-cache to record the result. Doing so can save lots of work when the same redex shows up more than once. Note that we record the link from the redex all the way to its *final* value, not just the single step reduction. Interestingly, using the flat-cache for the first reduction resulted in an increase in allocations of about 3% for the four T9872x tests. However, using the flat-cache in the later reduction is a similar gain. I (Richard E) don't currently (Dec '14) have any knowledge as to *why* these facts are true. ************************************************************************ * * Flattening a type variable * * ********************************************************************* -} -- | The result of flattening a tyvar "one step". data FlattenTvResult = FTRNotFollowed -- ^ The inert set doesn't make the tyvar equal to anything else | FTRFollowed TcType Coercion -- ^ The tyvar flattens to a not-necessarily flat other type. -- co :: new type ~r old type, where the role is determined by -- the FlattenEnv flattenTyVar :: TyVar -> FlatM (Xi, Coercion) flattenTyVar tv = do { mb_yes <- flatten_tyvar1 tv ; case mb_yes of FTRFollowed ty1 co1 -- Recur -> do { (ty2, co2) <- flatten_one ty1 -- ; traceFlat "flattenTyVar2" (ppr tv $$ ppr ty2) ; return (ty2, co2 `mkTransCo` co1) } FTRNotFollowed -- Done, but make sure the kind is zonked -- Note [Flattening] invariant (F1) -> do { tv' <- liftTcS $ updateTyVarKindM zonkTcType tv ; role <- getRole ; let ty' = mkTyVarTy tv' ; return (ty', mkTcReflCo role ty') } } flatten_tyvar1 :: TcTyVar -> FlatM FlattenTvResult -- "Flattening" a type variable means to apply the substitution to it -- Specifically, look up the tyvar in -- * the internal MetaTyVar box -- * the inerts -- See also the documentation for FlattenTvResult flatten_tyvar1 tv = do { mb_ty <- liftTcS $ isFilledMetaTyVar_maybe tv ; case mb_ty of Just ty -> do { traceFlat "Following filled tyvar" (ppr tv <+> equals <+> ppr ty) ; role <- getRole ; return (FTRFollowed ty (mkReflCo role ty)) } ; Nothing -> do { traceFlat "Unfilled tyvar" (ppr tv) ; fr <- getFlavourRole ; flatten_tyvar2 tv fr } } flatten_tyvar2 :: TcTyVar -> CtFlavourRole -> FlatM FlattenTvResult -- The tyvar is not a filled-in meta-tyvar -- Try in the inert equalities -- See Definition [Applying a generalised substitution] in TcSMonad -- See Note [Stability of flattening] in TcSMonad flatten_tyvar2 tv fr@(_, eq_rel) = do { ieqs <- liftTcS $ getInertEqs ; mode <- getMode ; case lookupDVarEnv ieqs tv of Just (ct:_) -- If the first doesn't work, -- the subsequent ones won't either | CTyEqCan { cc_ev = ctev, cc_tyvar = tv , cc_rhs = rhs_ty, cc_eq_rel = ct_eq_rel } <- ct , let ct_fr = (ctEvFlavour ctev, ct_eq_rel) , ct_fr `eqCanRewriteFR` fr -- This is THE key call of eqCanRewriteFR -> do { traceFlat "Following inert tyvar" (ppr mode <+> ppr tv <+> equals <+> ppr rhs_ty $$ ppr ctev) ; let rewrite_co1 = mkSymCo (ctEvCoercion ctev) rewrite_co = case (ct_eq_rel, eq_rel) of (ReprEq, _rel) -> ASSERT( _rel == ReprEq ) -- if this ASSERT fails, then -- eqCanRewriteFR answered incorrectly rewrite_co1 (NomEq, NomEq) -> rewrite_co1 (NomEq, ReprEq) -> mkSubCo rewrite_co1 ; return (FTRFollowed rhs_ty rewrite_co) } -- NB: ct is Derived then fmode must be also, hence -- we are not going to touch the returned coercion -- so ctEvCoercion is fine. _other -> return FTRNotFollowed } {- Note [An alternative story for the inert substitution] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ (This entire note is just background, left here in case we ever want to return the previous state of affairs) We used (GHC 7.8) to have this story for the inert substitution inert_eqs * 'a' is not in fvs(ty) * They are *inert* in the weaker sense that there is no infinite chain of (i1 `eqCanRewrite` i2), (i2 `eqCanRewrite` i3), etc This means that flattening must be recursive, but it does allow [G] a ~ [b] [G] b ~ Maybe c This avoids "saturating" the Givens, which can save a modest amount of work. It is easy to implement, in TcInteract.kick_out, by only kicking out an inert only if (a) the work item can rewrite the inert AND (b) the inert cannot rewrite the work item This is significantly harder to think about. It can save a LOT of work in occurs-check cases, but we don't care about them much. Trac #5837 is an example; all the constraints here are Givens [G] a ~ TF (a,Int) --> work TF (a,Int) ~ fsk inert fsk ~ a ---> work fsk ~ (TF a, TF Int) inert fsk ~ a ---> work a ~ (TF a, TF Int) inert fsk ~ a ---> (attempting to flatten (TF a) so that it does not mention a work TF a ~ fsk2 inert a ~ (fsk2, TF Int) inert fsk ~ (fsk2, TF Int) ---> (substitute for a) work TF (fsk2, TF Int) ~ fsk2 inert a ~ (fsk2, TF Int) inert fsk ~ (fsk2, TF Int) ---> (top-level reduction, re-orient) work fsk2 ~ (TF fsk2, TF Int) inert a ~ (fsk2, TF Int) inert fsk ~ (fsk2, TF Int) ---> (attempt to flatten (TF fsk2) to get rid of fsk2 work TF fsk2 ~ fsk3 work fsk2 ~ (fsk3, TF Int) inert a ~ (fsk2, TF Int) inert fsk ~ (fsk2, TF Int) ---> work TF fsk2 ~ fsk3 inert fsk2 ~ (fsk3, TF Int) inert a ~ ((fsk3, TF Int), TF Int) inert fsk ~ ((fsk3, TF Int), TF Int) Because the incoming given rewrites all the inert givens, we get more and more duplication in the inert set. But this really only happens in pathalogical casee, so we don't care. ************************************************************************ * * Unflattening * * ************************************************************************ An unflattening example: [W] F a ~ alpha flattens to [W] F a ~ fmv (CFunEqCan) [W] fmv ~ alpha (CTyEqCan) We must solve both! -} unflattenWanteds :: Cts -> Cts -> TcS Cts unflattenWanteds tv_eqs funeqs = do { tclvl <- getTcLevel ; traceTcS "Unflattening" $ braces $ vcat [ text "Funeqs =" <+> pprCts funeqs , text "Tv eqs =" <+> pprCts tv_eqs ] -- Step 1: unflatten the CFunEqCans, except if that causes an occurs check -- Occurs check: consider [W] alpha ~ [F alpha] -- ==> (flatten) [W] F alpha ~ fmv, [W] alpha ~ [fmv] -- ==> (unify) [W] F [fmv] ~ fmv -- See Note [Unflatten using funeqs first] ; funeqs <- foldrBagM unflatten_funeq emptyCts funeqs ; traceTcS "Unflattening 1" $ braces (pprCts funeqs) -- Step 2: unify the tv_eqs, if possible ; tv_eqs <- foldrBagM (unflatten_eq tclvl) emptyCts tv_eqs ; traceTcS "Unflattening 2" $ braces (pprCts tv_eqs) -- Step 3: fill any remaining fmvs with fresh unification variables ; funeqs <- mapBagM finalise_funeq funeqs ; traceTcS "Unflattening 3" $ braces (pprCts funeqs) -- Step 4: remove any tv_eqs that look like ty ~ ty ; tv_eqs <- foldrBagM finalise_eq emptyCts tv_eqs ; let all_flat = tv_eqs `andCts` funeqs ; traceTcS "Unflattening done" $ braces (pprCts all_flat) ; return all_flat } where ---------------- unflatten_funeq :: Ct -> Cts -> TcS Cts unflatten_funeq ct@(CFunEqCan { cc_fun = tc, cc_tyargs = xis , cc_fsk = fmv, cc_ev = ev }) rest = do { -- fmv should be an un-filled flatten meta-tv; -- we now fix its final value by filling it, being careful -- to observe the occurs check. Zonking will eliminate it -- altogether in due course rhs' <- zonkTcType (mkTyConApp tc xis) ; case occCheckExpand [fmv] rhs' of Just rhs'' -- Normal case: fill the tyvar -> do { setReflEvidence ev NomEq rhs'' ; unflattenFmv fmv rhs'' ; return rest } Nothing -> -- Occurs check return (ct `consCts` rest) } unflatten_funeq other_ct _ = pprPanic "unflatten_funeq" (ppr other_ct) ---------------- finalise_funeq :: Ct -> TcS Ct finalise_funeq (CFunEqCan { cc_fsk = fmv, cc_ev = ev }) = do { demoteUnfilledFmv fmv ; return (mkNonCanonical ev) } finalise_funeq ct = pprPanic "finalise_funeq" (ppr ct) ---------------- unflatten_eq :: TcLevel -> Ct -> Cts -> TcS Cts unflatten_eq tclvl ct@(CTyEqCan { cc_ev = ev, cc_tyvar = tv , cc_rhs = rhs, cc_eq_rel = eq_rel }) rest | NomEq <- eq_rel -- See Note [Do not unify representational equalities] -- in TcInteract , isFmvTyVar tv -- Previously these fmvs were untouchable, -- but now they are touchable -- NB: unlike unflattenFmv, filling a fmv here /does/ -- bump the unification count; it is "improvement" -- Note [Unflattening can force the solver to iterate] = ASSERT2( tyVarKind tv `eqType` tcTypeKind rhs, ppr ct ) -- CTyEqCan invariant should ensure this is true do { is_filled <- isFilledMetaTyVar tv ; elim <- case is_filled of False -> do { traceTcS "unflatten_eq 2" (ppr ct) ; tryFill ev tv rhs } True -> do { traceTcS "unflatten_eq 3" (ppr ct) ; try_fill_rhs ev tclvl tv rhs } ; if elim then do { setReflEvidence ev eq_rel (mkTyVarTy tv) ; return rest } else return (ct `consCts` rest) } | otherwise = return (ct `consCts` rest) unflatten_eq _ ct _ = pprPanic "unflatten_irred" (ppr ct) ---------------- try_fill_rhs ev tclvl lhs_tv rhs -- Constraint is lhs_tv ~ rhs_tv, -- and lhs_tv is filled, so try RHS | Just (rhs_tv, co) <- getCastedTyVar_maybe rhs -- co :: kind(rhs_tv) ~ kind(lhs_tv) , isFmvTyVar rhs_tv || (isTouchableMetaTyVar tclvl rhs_tv && not (isTyVarTyVar rhs_tv)) -- LHS is a filled fmv, and so is a type -- family application, which a TyVarTv should -- not unify with = do { is_filled <- isFilledMetaTyVar rhs_tv ; if is_filled then return False else tryFill ev rhs_tv (mkTyVarTy lhs_tv `mkCastTy` mkSymCo co) } | otherwise = return False ---------------- finalise_eq :: Ct -> Cts -> TcS Cts finalise_eq (CTyEqCan { cc_ev = ev, cc_tyvar = tv , cc_rhs = rhs, cc_eq_rel = eq_rel }) rest | isFmvTyVar tv = do { ty1 <- zonkTcTyVar tv ; rhs' <- zonkTcType rhs ; if ty1 `tcEqType` rhs' then do { setReflEvidence ev eq_rel rhs' ; return rest } else return (mkNonCanonical ev `consCts` rest) } | otherwise = return (mkNonCanonical ev `consCts` rest) finalise_eq ct _ = pprPanic "finalise_irred" (ppr ct) tryFill :: CtEvidence -> TcTyVar -> TcType -> TcS Bool -- (tryFill tv rhs ev) assumes 'tv' is an /un-filled/ MetaTv -- If tv does not appear in 'rhs', it set tv := rhs, -- binds the evidence (which should be a CtWanted) to Refl<rhs> -- and return True. Otherwise returns False tryFill ev tv rhs = ASSERT2( not (isGiven ev), ppr ev ) do { rhs' <- zonkTcType rhs ; case () of _ | Just tv' <- tcGetTyVar_maybe rhs' , tv == tv' -- tv == rhs -> return True _ | Just rhs'' <- occCheckExpand [tv] rhs' -> do { -- Fill the tyvar unifyTyVar tv rhs'' ; return True } _ | otherwise -- Occurs check -> return False } setReflEvidence :: CtEvidence -> EqRel -> TcType -> TcS () setReflEvidence ev eq_rel rhs = setEvBindIfWanted ev (evCoercion refl_co) where refl_co = mkTcReflCo (eqRelRole eq_rel) rhs {- Note [Unflatten using funeqs first] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [W] G a ~ Int [W] F (G a) ~ G a do not want to end up with [W] F Int ~ Int because that might actually hold! Better to end up with the two above unsolved constraints. The flat form will be G a ~ fmv1 (CFunEqCan) F fmv1 ~ fmv2 (CFunEqCan) fmv1 ~ Int (CTyEqCan) fmv1 ~ fmv2 (CTyEqCan) Flatten using the fun-eqs first. -} -- | Like 'splitPiTys'' but comes with a 'Bool' which is 'True' iff there is at -- least one named binder. split_pi_tys' :: Type -> ([TyCoBinder], Type, Bool) split_pi_tys' ty = split ty ty where split orig_ty ty | Just ty' <- coreView ty = split orig_ty ty' split _ (ForAllTy b res) = let (bs, ty, _) = split res res in (Named b : bs, ty, True) split _ (FunTy arg res) = let (bs, ty, named) = split res res in (Anon arg : bs, ty, named) split orig_ty _ = ([], orig_ty, False) {-# INLINE split_pi_tys' #-} -- | Like 'tyConBindersTyCoBinders' but you also get a 'Bool' which is true iff -- there is at least one named binder. ty_con_binders_ty_binders' :: [TyConBinder] -> ([TyCoBinder], Bool) ty_con_binders_ty_binders' = foldr go ([], False) where go (Bndr tv (NamedTCB vis)) (bndrs, _) = (Named (Bndr tv vis) : bndrs, True) go (Bndr tv AnonTCB) (bndrs, n) = (Anon (tyVarKind tv) : bndrs, n) {-# INLINE go #-} {-# INLINE ty_con_binders_ty_binders' #-}