Safe HaskellSafe-Infered



Abstract syntax definitions for Template Haskell.



class (Monad m, Applicative m) => Quasi m whereSource




:: String 
-> m Name

Fresh names



:: Bool 
-> String 
-> m ()

Report an error (True) or warning (False) ...but carry on; use fail to stop



:: m a

the error handler

-> m a

action which may fail

-> m a

Recover from the monadic fail

qLookupName :: Bool -> String -> m (Maybe Name)Source

qReify :: Name -> m InfoSource

qReifyInstances :: Name -> [Type] -> m [Dec]Source

qLocation :: m LocSource

qRunIO :: IO a -> m aSource

Input/output (dangerous)

qAddDependentFile :: FilePath -> m ()Source


newtype Q a Source




unQ :: forall m. Quasi m => m a

runQ :: Quasi m => Q a -> m aSource

newName :: String -> Q NameSource

Generate a fresh name, which cannot be captured.

For example, this:

f = $(do
  nm1 <- newName "x"
  let nm2 = mkName "x"
  return (LamE [VarP nm1] (LamE [VarP nm2] (VarE nm1)))

will produce the splice

f = \x0 -> \x -> x0

In particular, the occurrence VarE nm1 refers to the binding VarP nm1, and is not captured by the binding VarP nm2.

Although names generated by newName cannot be captured, they can capture other names. For example, this:

g = $(do
  nm1 <- newName "x"
  let nm2 = mkName "x"
  return (LamE [VarP nm2] (LamE [VarP nm1] (VarE nm2)))

will produce the splice

g = \x -> \x0 -> x0

since the occurrence VarE nm2 is captured by the innermost binding of x, namely VarP nm1.

report :: Bool -> String -> Q ()Source

Report an error (True) or warning (False), but carry on; use fail to stop.

reportError :: String -> Q ()Source

Report an error to the user, but allow the current splice's computation to carry on. To abort the computation, use fail.

reportWarning :: String -> Q ()Source

Report a warning to the user, and carry on.



:: Q a

handler to invoke on failure

-> Q a

computation to run

-> Q a 

Recover from errors raised by reportError or fail.

lookupTypeName :: String -> Q (Maybe Name)Source

Look up the given name in the (type namespace of the) current splice's scope. See Language.Haskell.TH.Syntax for more details.

lookupValueName :: String -> Q (Maybe Name)Source

Look up the given name in the (value namespace of the) current splice's scope. See Language.Haskell.TH.Syntax for more details.

The functions lookupTypeName and lookupValueName provide a way to query the current splice's context for what names are in scope. The function lookupTypeName queries the type namespace, whereas lookupValueName queries the value namespace, but the functions are otherwise identical.

A call lookupValueName s will check if there is a value with name s in scope at the current splice's location. If there is, the Name of this value is returned; if not, then Nothing is returned.

The returned name cannot be "captured". For example:

 f = "global"
 g = $( do
          Just nm <- lookupValueName "f"
          [| let f = "local" in $( varE nm ) |]

In this case, g = "global"; the call to lookupValueName returned the global f, and this name was not captured by the local definition of f.

The lookup is performed in the context of the top-level splice being run. For example:

 f = "global"
 g = $( [| let f = "local" in 
                Just nm <- lookupValueName "f"
                varE nm
             ) |] )

Again in this example, g = "global", because the call to lookupValueName queries the context of the outer-most $(...).

Operators should be queried without any surrounding parentheses, like so:

 lookupValueName "+"

Qualified names are also supported, like so:

 lookupValueName "Prelude.+"
 lookupValueName "Prelude.map"

reify :: Name -> Q InfoSource

reify looks up information about the Name.

It is sometimes useful to construct the argument name using lookupTypeName or lookupValueName to ensure that we are reifying from the right namespace. For instance, in this context:

 data D = D

which D does reify (mkName "D") return information about? (Answer: D-the-type, but don't rely on it.) To ensure we get information about D-the-value, use lookupValueName:

   Just nm <- lookupValueName "D"
   reify nm

and to get information about D-the-type, use lookupTypeName.

reifyInstances :: Name -> [Type] -> Q [InstanceDec]Source

reifyInstances nm tys returns a list of visible instances of nm tys. That is, if nm is the name of a type class, then all instances of this class at the types tys are returned. Alternatively, if nm is the name of a data family or type family, all instances of this family at the types tys are returned.

isInstance :: Name -> [Type] -> Q BoolSource

Is the list of instances returned by reifyInstances nonempty?

location :: Q LocSource

The location at which this computation is spliced.

runIO :: IO a -> Q aSource

The runIO function lets you run an I/O computation in the Q monad. Take care: you are guaranteed the ordering of calls to runIO within a single Q computation, but not about the order in which splices are run.

Note: for various murky reasons, stdout and stderr handles are not necesarily flushed when the compiler finishes running, so you should flush them yourself.

addDependentFile :: FilePath -> Q ()Source

Record external files that runIO is using (dependent upon). The compiler can then recognize that it should re-compile the file using this TH when the external file changes. Note that ghc -M will still not know about these dependencies - it does not execute TH. Expects an absolute file path.

returnQ :: a -> Q aSource

bindQ :: Q a -> (a -> Q b) -> Q bSource

sequenceQ :: [Q a] -> Q [a]Source

class Lift t whereSource


lift :: t -> Q ExpSource


Lift Bool 
Lift Char 
Lift Int 
Lift Integer 
Lift a => Lift [a] 
Lift a => Lift (Maybe a) 
(Lift a, Lift b) => Lift (Either a b) 
(Lift a, Lift b) => Lift (a, b) 
(Lift a, Lift b, Lift c) => Lift (a, b, c) 
(Lift a, Lift b, Lift c, Lift d) => Lift (a, b, c, d) 
(Lift a, Lift b, Lift c, Lift d, Lift e) => Lift (a, b, c, d, e) 
(Lift a, Lift b, Lift c, Lift d, Lift e, Lift f) => Lift (a, b, c, d, e, f) 
(Lift a, Lift b, Lift c, Lift d, Lift e, Lift f, Lift g) => Lift (a, b, c, d, e, f, g) 

Much of Name API is concerned with the problem of name capture, which can be seen in the following example.

 f expr = [| let x = 0 in $expr |]
 g x = $( f [| x |] )
 h y = $( f [| y |] )

A naive desugaring of this would yield:

 g x = let x = 0 in x
 h y = let x = 0 in y

All of a sudden, g and h have different meanings! In this case, we say that the x in the RHS of g has been captured by the binding of x in f.

What we actually want is for the x in f to be distinct from the x in g, so we get the following desugaring:

 g x = let x' = 0 in x
 h y = let x' = 0 in y

which avoids name capture as desired.

In the general case, we say that a Name can be captured if the thing it refers to can be changed by adding new declarations.

data Name Source

An abstract type representing names in the syntax tree.

Names can be constructed in several ways, which come with different name-capture guarantees (see Language.Haskell.TH.Syntax for an explanation of name capture):

  • the built-in syntax 'f and ''T can be used to construct names, The expression 'f gives a Name which refers to the value f currently in scope, and ''T gives a Name which refers to the type T currently in scope. These names can never be captured.
  • lookupValueName and lookupTypeName are similar to 'f and ''T respectively, but the Names are looked up at the point where the current splice is being run. These names can never be captured.
  • newName monadically generates a new name, which can never be captured.
  • mkName generates a capturable name.

Names constructed using newName and mkName may be used in bindings (such as let x = ... or x -> ...), but names constructed using lookupValueName, lookupTypeName, 'f, ''T may not.


Name OccName NameFlavour 

data NameFlavour Source



An unqualified name; dynamically bound

NameQ ModName

A qualified name; dynamically bound

NameU Int#

A unique local name

NameL Int#

Local name bound outside of the TH AST

NameG NameSpace PkgName ModName

Global name bound outside of the TH AST: An original name (occurrences only, not binders) Need the namespace too to be sure which thing we are naming


Eq NameFlavour 
Data NameFlavour

Although the NameFlavour type is abstract, the Data instance is not. The reason for this is that currently we use Data to serialize values in annotations, and in order for that to work for Template Haskell names introduced via the 'x syntax we need gunfold on NameFlavour to work. Bleh!

The long term solution to this is to use the binary package for annotation serialization and then remove this instance. However, to do _that_ we need to wait on binary to become stable, since boot libraries cannot be upgraded seperately from GHC itself.

This instance cannot be derived automatically due to bug #2701

Ord NameFlavour 
Typeable NameFlavour 

data NameSpace Source





Data constructors


Type constructors and classes; Haskell has them in the same name space for now.

nameBase :: Name -> StringSource

The name without its module prefix

nameModule :: Name -> Maybe StringSource

Module prefix of a name, if it exists

mkName :: String -> NameSource

Generate a capturable name. Occurrences of such names will be resolved according to the Haskell scoping rules at the occurrence site.

For example:

 f = [| pi + $(varE (mkName "pi")) |]
 g = let pi = 3 in $f

In this case, g is desugared to

 g = Prelude.pi + 3

Note that mkName may be used with qualified names:

 mkName "Prelude.pi"

See also dyn for a useful combinator. The above example could be rewritten using dyn as

 f = [| pi + $(dyn "pi") |]

mkNameU :: String -> Uniq -> NameSource

Only used internally

mkNameL :: String -> Uniq -> NameSource

Only used internally

mkNameG :: NameSpace -> String -> String -> String -> NameSource

Used for 'x etc, but not available to the programmer

data NameIs Source



tupleDataName :: Int -> NameSource

Tuple data constructor

tupleTypeName :: Int -> NameSource

Tuple type constructor

unboxedTupleDataName :: Int -> NameSource

Unboxed tuple data constructor

unboxedTupleTypeName :: Int -> NameSource

Unboxed tuple type constructor

type CharPosSource


 = (Int, Int)

Line and character position

data Info Source

Obtained from reify in the Q Monad.


ClassI Dec [InstanceDec]

A class, with a list of its visible instances

ClassOpI Name Type ParentName Fixity

A class method

TyConI Dec

A "plain" type constructor. "Fancier" type constructors are returned using PrimTyConI or FamilyI as appropriate

FamilyI Dec [InstanceDec]

A type or data family, with a list of its visible instances

PrimTyConI Name Arity Unlifted

A "primitive" type constructor, which can't be expressed with a Dec. Examples: (->), Int#.

DataConI Name Type ParentName Fixity

A data constructor

VarI Name Type (Maybe Dec) Fixity

A "value" variable (as opposed to a type variable, see TyVarI).

The Maybe Dec field contains Just the declaration which defined the variable -- including the RHS of the declaration -- or else Nothing, in the case where the RHS is unavailable to the compiler. At present, this value is _always_ Nothing: returning the RHS has not yet been implemented because of lack of interest.

TyVarI Name Type

A type variable.

The Type field contains the type which underlies the variable. At present, this is always VarT theName, but future changes may permit refinement of this.

type ParentName = NameSource

In ClassOpI and DataConI, name of the parent class or type

type Arity = IntSource

In PrimTyConI, arity of the type constructor

type Unlifted = BoolSource

In PrimTyConI, is the type constructor unlifted?

type InstanceDec = DecSource

InstanceDec desribes a single instance of a class or type function. It is just a Dec, but guaranteed to be one of the following:

maxPrecedence :: IntSource

Highest allowed operator precedence for Fixity constructor (answer: 9)

defaultFixity :: FixitySource

Default fixity: infixl 9

When implementing antiquotation for quasiquoters, one often wants to parse strings into expressions:

 parse :: String -> Maybe Exp

But how should we parse a + b * c? If we don't know the fixities of + and *, we don't know whether to parse it as a + (b * c) or (a + b) * c.

In cases like this, use UInfixE or UInfixP, which stand for "unresolved infix expression" and "unresolved infix pattern". When the compiler is given a splice containing a tree of UInfixE applications such as

   (UInfixE e1 op1 e2)
   (UInfixE e3 op3 e4)

it will look up and the fixities of the relevant operators and reassociate the tree as necessary.

  • trees will not be reassociated across ParensE or ParensP, which are of use for parsing expressions like
 (a + b * c) + d * e
  • InfixE and InfixP expressions are never reassociated.
  • The UInfixE constructor doesn't support sections. Sections such as (a *) have no ambiguity, so InfixE suffices. For longer sections such as (a + b * c -), use an InfixE constructor for the outer-most section, and use UInfixE constructors for all other operators:
   Just (UInfixE ...a + b * c...)

Sections such as (a + b +) and ((a + b) +) should be rendered into Exps differently:

 (+ a + b)   ---> InfixE Nothing + (Just $ UInfixE a + b)
                    -- will result in a fixity error if (+) is left-infix
 (+ (a + b)) ---> InfixE Nothing + (Just $ ParensE $ UInfixE a + b)
                    -- no fixity errors
  • Quoted expressions such as
 [| a * b + c |] :: Q Exp
 [p| a : b : c |] :: Q Pat

will never contain UInfixE, UInfixP, ParensE, or ParensP constructors.

data Lit Source


CharL Char 
StringL String 
IntegerL Integer

Used for overloaded and non-overloaded literals. We don't have a good way to represent non-overloaded literals at the moment. Maybe that doesn't matter?

RationalL Rational 
IntPrimL Integer 
WordPrimL Integer 
FloatPrimL Rational 
DoublePrimL Rational 
StringPrimL [Word8]

A primitive C-style string, type Addr#


data Pat Source

Pattern in Haskell given in {}


LitP Lit
{ 5 or c }
VarP Name
{ x }
TupP [Pat]
{ (p1,p2) }
UnboxedTupP [Pat]
{ () }
ConP Name [Pat]
data T1 = C1 t1 t2; {C1 p1 p1} = e
InfixP Pat Name Pat
foo ({x :+ y}) = e
UInfixP Pat Name Pat
foo ({x :+ y}) = e

See Language.Haskell.TH.Syntax

ParensP Pat

See Language.Haskell.TH.Syntax

TildeP Pat
{ ~p }
BangP Pat
{ !p }
AsP Name Pat
{ x @ p }
{ _ }
RecP Name [FieldPat]
f (Pt { pointx = x }) = g x
ListP [Pat]
{ [1,2,3] }
SigP Pat Type
{ p :: t }
ViewP Exp Pat
{ e -> p }


data Match Source


Match Pat Body [Dec]
case e of { pat -> body where decs }

data Clause Source


Clause [Pat] Body [Dec]
f { p1 p2 = body where decs }

data Exp Source


VarE Name
{ x }
ConE Name
data T1 = C1 t1 t2; p = {C1} e1 e2
LitE Lit
{ 5 or c}
AppE Exp Exp
{ f x }
InfixE (Maybe Exp) Exp (Maybe Exp)
{x + y} or {(x+)} or {(+ x)} or {(+)}
UInfixE Exp Exp Exp
{x + y}

See Language.Haskell.TH.Syntax

ParensE Exp
{ (e) }

See Language.Haskell.TH.Syntax

LamE [Pat] Exp
{  p1 p2 -> e }
LamCaseE [Match]
{ case m1; m2 }
TupE [Exp]
{ (e1,e2) }
UnboxedTupE [Exp]
{ () }
CondE Exp Exp Exp
{ if e1 then e2 else e3 }
MultiIfE [(Guard, Exp)]
{ if | g1 -> e1 | g2 -> e2 }
LetE [Dec] Exp
{ let x=e1;   y=e2 in e3 }
CaseE Exp [Match]
{ case e of m1; m2 }
DoE [Stmt]
{ do { p <- e1; e2 }  }
CompE [Stmt]
{ [ (x,y) | x <- xs, y <- ys ] }

The result expression of the comprehension is the last of the Stmts, and should be a NoBindS.

E.g. translation:

 [ f x | x <- xs ]
 CompE [BindS (VarP x) (VarE xs), NoBindS (AppE (VarE f) (VarE x))]
ArithSeqE Range
{ [ 1 ,2 .. 10 ] }
ListE [Exp]
{ [1,2,3] }
SigE Exp Type
{ e :: t }
RecConE Name [FieldExp]
{ T { x = y, z = w } }
RecUpdE Exp [FieldExp]
{ (f x) { z = w } }


data Body Source


GuardedB [(Guard, Exp)]
f p { | e1 = e2 
      | e3 = e4 } 
 where ds
NormalB Exp
f p { = e } where ds

data Guard Source


NormalG Exp
f x { | odd x } = x
PatG [Stmt]
f x { | Just y <- x, Just z <- y } = z

data Stmt Source


BindS Pat Exp 
LetS [Dec] 
NoBindS Exp 
ParS [[Stmt]] 

data Dec Source


FunD Name [Clause]
{ f p1 p2 = b where decs }
ValD Pat Body [Dec]
{ p = b where decs }
DataD Cxt Name [TyVarBndr] [Con] [Name]
{ data Cxt x => T x = A x | B (T x)
       deriving (Z,W)}
NewtypeD Cxt Name [TyVarBndr] Con [Name]
{ newtype Cxt x => T x = A (B x)
       deriving (Z,W)}
TySynD Name [TyVarBndr] Type
{ type T x = (x,x) }
ClassD Cxt Name [TyVarBndr] [FunDep] [Dec]
{ class Eq a => Ord a where ds }
InstanceD Cxt Type [Dec]
{ instance Show w => Show [w]
       where ds }
SigD Name Type
{ length :: [a] -> Int }
ForeignD Foreign
{ foreign import ... }
{ foreign export ... }
InfixD Fixity Name
{ infix 3 foo }
PragmaD Pragma
{ {--} }
FamilyD FamFlavour Name [TyVarBndr] (Maybe Kind)
{ type family T a b c :: * }
DataInstD Cxt Name [Type] [Con] [Name]
{ data instance Cxt x => T [x] = A x 
                                | B (T x)
       deriving (Z,W)}
NewtypeInstD Cxt Name [Type] Con [Name]
{ newtype instance Cxt x => T [x] = A (B x)
       deriving (Z,W)}
TySynInstD Name [Type] Type
{ type instance T (Maybe x) = (x,x) }


type CxtSource


 = [Pred]
(Eq a, Ord b)

data Pred Source


ClassP Name [Type]
Eq (Int, a)
EqualP Type Type
F a ~ Bool

data Con Source


NormalC Name [StrictType]
C Int a
RecC Name [VarStrictType]
C { v :: Int, w :: a }
InfixC StrictType Name StrictType
Int :+ a
ForallC [TyVarBndr] Cxt Con
forall a. Eq a => C [a]


data Type Source


ForallT [TyVarBndr] Cxt Type
forall <vars>. <ctxt> -> <type>
AppT Type Type
T a b
SigT Type Kind
t :: k
VarT Name
ConT Name
PromotedT Name
TupleT Int
(,), (,,), etc.
UnboxedTupleT Int
(), (), etc.
PromotedTupleT Int
'(), '(,), '(,,), etc.
LitT TyLit
0,1,2, etc.

type Kind = TypeSource

To avoid duplication between kinds and types, they are defined to be the same. Naturally, you would never have a type be StarT and you would never have a kind be SigT, but many of the other constructors are shared. Note that the kind Bool is denoted with ConT, not PromotedT. Similarly, tuple kinds are made with TupleT, not PromotedTupleT.