aboutsummaryrefslogtreecommitdiff
path: root/haddock-api
diff options
context:
space:
mode:
Diffstat (limited to 'haddock-api')
-rw-r--r--haddock-api/src/Haddock/Backends/Xhtml/Specialize.hs95
1 files changed, 74 insertions, 21 deletions
diff --git a/haddock-api/src/Haddock/Backends/Xhtml/Specialize.hs b/haddock-api/src/Haddock/Backends/Xhtml/Specialize.hs
index fccdaa95..1da089d9 100644
--- a/haddock-api/src/Haddock/Backends/Xhtml/Specialize.hs
+++ b/haddock-api/src/Haddock/Backends/Xhtml/Specialize.hs
@@ -30,23 +30,31 @@ import Data.Set (Set)
import qualified Data.Set as Set
+-- | Instantiate all occurrences of given name with particular type.
specialize :: (Eq name, Typeable name)
=> Data a
=> name -> HsType name -> a -> a
-specialize name details = everywhere (mkT $ specializeStep name details)
+specialize name details =
+ everywhere $ mkT step
+ where
+ step (HsTyVar name') | name == name' = details
+ step typ = typ
+-- | Instantiate all occurrences of given names with corresponding types.
+--
+-- It is just a convenience function wrapping 'specialize' that supports more
+-- that one specialization.
specialize' :: (Eq name, Typeable name)
=> Data a
=> [(name, HsType name)] -> a -> a
specialize' = flip $ foldr (uncurry specialize)
-specializeStep :: Eq name => name -> HsType name -> HsType name -> HsType name
-specializeStep name details (HsTyVar name') | name == name' = details
-specializeStep _ _ typ = typ
-
-
+-- | Instantiate given binders with corresponding types.
+--
+-- Again, it is just a convenience function around 'specialize'. Note that
+-- length of type list should be the same as the number of binders.
specializeTyVarBndrs :: (Eq name, Typeable name, DataId name)
=> LHsTyVarBndrs name -> [HsType name]
-> HsType name -> HsType name
@@ -58,6 +66,12 @@ specializeTyVarBndrs bndrs typs =
bname (KindedTyVar (L _ name) _) = name
+-- | Make given type use tuple and list literals where appropriate.
+--
+-- After applying 'specialize' function some terms may not use idiomatic list
+-- and tuple literals resulting in types like @[] a@ or @(,,) a b c@. This
+-- can be fixed using 'sugar' function, that will turn such types into @[a]@
+-- and @(a, b, c)@.
sugar :: forall name. (NamedThing name, DataId name)
=> HsType name -> HsType name
sugar =
@@ -93,6 +107,19 @@ sugarTuples typ =
aux _ _ = typ
+-- | Compute arity of given tuple operator.
+--
+-- >>> parseTupleArity "(,,)"
+-- Just 3
+--
+-- >>> parseTupleArity "(,,,,)"
+-- Just 5
+--
+-- >>> parseTupleArity "abc"
+-- Nothing
+--
+-- >>> parseTupleArity "()"
+-- Nothing
parseTupleArity :: String -> Maybe Int
parseTupleArity ('(':commas) = do
n <- parseCommas commas
@@ -105,6 +132,17 @@ parseTupleArity ('(':commas) = do
parseTupleArity _ = Nothing
+-- | Haskell AST type representation.
+--
+-- This type is used for renaming (more below), essentially the ambiguous (!)
+-- version of 'Name'. So, why is this 'FastString' instead of 'OccName'? Well,
+-- it was 'OccName' before, but turned out that 'OccName' sometimes also
+-- contains namespace information, differentiating visually same types.
+--
+-- And 'FastString' is used because it is /visual/ part of 'OccName' - it is
+-- not converted to 'String' or alike to avoid new allocations. Additionally,
+-- since it is stored mostly in 'Set', fast comparison of 'FastString' is also
+-- quite nice.
type NameRep = FastString
getNameRep :: NamedThing name => name -> NameRep
@@ -127,10 +165,39 @@ setInternalOccName occ name =
nname' = mkInternalName (nameUnique nname) occ (nameSrcSpan nname)
+-- | Compute set of free variables of given type.
+freeVariables :: forall name. (NamedThing name, DataId name)
+ => HsType name -> Set NameRep
+freeVariables =
+ everythingWithState Set.empty Set.union query
+ where
+ query term ctx = case cast term :: Maybe (HsType name) of
+ Just (HsForAllTy _ _ bndrs _ _) ->
+ (Set.empty, Set.union ctx (bndrsNames bndrs))
+ Just (HsTyVar name)
+ | getName name `Set.member` ctx -> (Set.empty, ctx)
+ | otherwise -> (Set.singleton $ getNameRep name, ctx)
+ _ -> (Set.empty, ctx)
+ bndrsNames = Set.fromList . map (getName . tyVarName . unLoc) . hsq_tvs
+
+
+-- | Make given type visually unambiguous.
+--
+-- After applying 'specialize' method, some free type variables may become
+-- visually ambiguous - for example, having @a -> b@ and specializing @a@ to
+-- @(a -> b)@ we get @(a -> b) -> b@ where first occurrence of @b@ refers to
+-- different type variable than latter one. Applying 'rename' function
+-- will fix that type to be visually unambiguous again (making it something
+-- like @(a -> c) -> b@).
rename :: SetName name => Set NameRep -> HsType name -> HsType name
rename fv typ = fst $ evalRWS (renameType typ) fv Map.empty
+-- | Renaming monad.
+--
+-- This is just a simple RWS instance, where /reader/ part consists of names
+-- that are initially taken and cannot change, /state/ part is just context
+-- with name bindings and /writer/ part is not used.
type Rename name a = RWS (Set NameRep) () (Map Name name) a
@@ -171,21 +238,6 @@ renameType HsWildcardTy = pure HsWildcardTy
renameType (HsNamedWildcardTy name) = HsNamedWildcardTy <$> renameName name
-freeVariables :: forall name. (NamedThing name, DataId name)
- => HsType name -> Set NameRep
-freeVariables =
- everythingWithState Set.empty Set.union query
- where
- query term ctx = case cast term :: Maybe (HsType name) of
- Just (HsForAllTy _ _ bndrs _ _) ->
- (Set.empty, Set.union ctx (bndrsNames bndrs))
- Just (HsTyVar name)
- | getName name `Set.member` ctx -> (Set.empty, ctx)
- | otherwise -> (Set.singleton $ getNameRep name, ctx)
- _ -> (Set.empty, ctx)
- bndrsNames = Set.fromList . map (getName . tyVarName . unLoc) . hsq_tvs
-
-
renameLType :: SetName name => LHsType name -> Rename name (LHsType name)
renameLType = located renameType
@@ -235,6 +287,7 @@ renameName name = do
Nothing -> name
+-- | Generate fresh occurrence name, put it into context and return.
freshName :: SetName name => name -> Rename name name
freshName name = do
fv <- ask