th-desugar
Functions to desugar Template Haskell
https://github.com/goldfirere/th-desugar
Version on this page: | 1.11 |
LTS Haskell 23.0: | 1.16@rev:1 |
Stackage Nightly 2024-12-12: | 1.17 |
Latest on Hackage: | 1.18 |
th-desugar-1.11@sha256:573a176b4642c1ad1f40eef76ad2bfc6bdafe3e5d162f7a6bb08132488cc5496,3577
Module documentation for 1.11
- Language
- Language.Haskell
th-desugar
Package
This package provides the Language.Haskell.TH.Desugar
module, which desugars
Template Haskell’s rich encoding of Haskell syntax into a simpler encoding.
This desugaring discards surface syntax information (such as the use of infix
operators) but retains the original meaning of the TH code. The intended use
of this package is as a preprocessor for more advanced code manipulation
tools. Note that the input to any of the ds...
functions should be produced
from a TH quote, using the syntax [| ... |]
. If the input to these functions
is a hand-coded TH syntax tree, the results may be unpredictable. In
particular, it is likely that promoted datatypes will not work as expected.
One explicit goal of this package is to reduce the burden of supporting multiple GHC / TH versions. Thus, the desugared language is the same across all GHC versions, and any inconsistencies are handled internally.
The package was designed for use with the singletons
package, so some design
decisions are based on that use case, when more than one design choice was
possible.
I will try to keep this package up-to-date with respect to changes in GHC.
Known limitations
th-desugar
sometimes has to construct types for certain Haskell entities.
For instance, th-desugar
desugars all Haskell98-style constructors to use
GADT syntax, so the following:
data T (a :: k) = MkT (Proxy a)
Will be desugared to something like this:
data T (a :: k) where
MkT :: forall k (a :: k). Proxy a -> T (a :: k)
Notice that k
is explicitly quantified in the type of MkT
. This is due to
an additional pass that th-desugar
performs over the type variable binders
of T
to extract all implicitly quantified variables and make them explicit.
This makes the desugared types forwards-compatible with a
future version of GHC
that requires all kind variables in a top-level forall
to be explicitly
quantified.
This process of extracting all implicitly quantified kind variables is not
perfect, however. There are some obscure programs that will cause th-desugar
to produce type variable binders that are ill scoped. Here is one example:
data P k (a :: k)
data Foo (a :: Proxy j) (b :: k) c = MkFoo c (P k j)
If you squint hard at MkFoo
, you’ll notice that j :: k
. However, this
relationship is not expressed syntactically, which means that th-desugar
will not be aware of it. Therefore, th-desugar
will desugar Foo
to:
data Foo (a :: Proxy j) (b :: k) c where
MkFoo :: forall j k (a :: Proxy j) (b :: k) c.
c -> P k j -> Foo (a :: Proxy j) (b :: k) c
This is incorrect since k
must come before j
in order to be well scoped.
There is a workaround to this issue, however: add more explicit kind
information. If you had instead written this:
data Foo (a :: Proxy (j :: k)) (b :: k) c = MkFoo c (P k j)
Then the fact that j :: k
is expressed directly in the AST, so th-desugar
is able to pick up on it and pick forall k j (a :: Proxy j) (b :: k) c. <...>
as the telescope for the type of MkFoo
.
The following constructs are known to be susceptible to this issue:
- Desugared Haskell98-style constructors
- Locally reified class methods
- Locally reified record selectors
- Locally reified data constructors
- Locally reified type family instances (on GHC 8.8 and later, in which the
Template Haskell AST supports explicit
foralls
in type family equations)
Changes
th-desugar
release notes
Version 1.11 [????.??.??]
- Support GHC 8.10.
- Add support for visible dependent quantification. As part of this change,
the way
th-desugar
representsforall
and constraint types has been overhauled:-
The existing
DForallT
constructor has been split into two smaller constructors:data DType = ... - | DForallT [DTyVarBndr] DCxt DType + | DForallT ForallVisFlag [DTyVarBndr] DType + | DConstrainedT DCxt DType | ... +data ForallVisFlag + = ForallVis + | ForallInvis
The previous design combined
forall
s and constraints into a single constructor, while the new design puts them in distinct constructorsDForallT
andDConstrainedT
, respectively. The newDForallT
constructor also has aForallVisFlag
field to distinguish invisibleforall
s (e.g.,forall a. a
) from visibleforall
s (e.g.,forall a -> a
). -
The
unravel
function has been renamed tounravelDType
and now returns(DFunArgs, DType)
, whereDFunArgs
is a data type that represents the possible arguments in a function type (see the Haddocks forDFunArgs
for more details). There is also anunravelDType
counterpart forType
s namedunravelType
, complete with its ownFunArgs
data type.{D}FunArgs
also have some supporting operations, includingfilter{D}VisFunArgs
(to obtain only the visible arguments) andravel{D}Type
(to construct a function type using{D}FunArgs
and a return{D}Type
).
-
- Support standalone kind signatures by adding a
DKiSigD
constructor toDDec
. - Add
dsReifyType
,reifyTypeWithLocals_maybe
, andreifyTypeWithLocals
, which allow looking up the types or kinds of locally declared entities. - Fix a bug in which
reifyFixityWithLocals
would not look into local fixity declarations inside of type classes. - Fix a bug in which
reifyFixityWithLocals
would return incorrect results for classes with associated type family defaults.
Version 1.10
-
Support GHC 8.8. Drop support for GHC 7.6.
-
Add support for visible kind application, type variable
foralls
inRULES
, and explicitforall
s in type family instances. Correspondingly,-
There is now a
DAppKindT
constructor inDType
. -
Previously, the
DDataInstD
constructor had fields of typeName
and[DType]
. Those have been scrapped in favor of a single field of typeDType
, representing the application of the data family name (which was previously theName
) to its arguments (which was previously the[DType]
).DDataInstD
also has a new field of typeMaybe [DTyVarBndr]
to represent its explicitly quantified type variables (if present). -
Previously, the
DTySynEqn
constructor had a field of type[DType]
. That has been scrapped in favor of a field of typeDType
, representing the application of the type family name (whichDTySynEqn
did not used to contain!) to its arguments (which was previously the[DType]
).DTySynEqn
also has a new field of typeMaybe [DTyVarBndr]
to represent its explicitly quantified type variables (if present). -
DTySynInstD
no longer has a field of typeName
, as that is redundant now that eachDTySynEqn
contains the sameName
. -
There is now a field of type
Maybe [DTyVarBndr]
in theDRuleP
constructor to represent bound type variables inRULES
(if present).
-
-
Add a field of type
Maybe [DTyVarBndr]
toDInstanceD
andDStandaloneDerivD
for optionally quantifying type variables explicitly. If supplied with aJust
, this sweetens the instance type to use aForallT
to represent the explicit quantification. This trick is not supported forInstanceD
on GHC 8.0 and forStandaloneDerivD
on GHC 7.10 or 8.0, so be aware of this limitation if you supplyJust
for this field. -
Add support for desugaring implicit params. This does not involve any changes to the
th-desugar
AST, as:(?x :: a) => ...
is desugared toIP "x" a => ...
.id ?x
is desugared toid (ip @"x")
.let ?x = 42 in ...
is desugared tolet new_x_val = 42 in bindIP @"x" new_x_val ...
(wherebindIP
is a new utility function exported byLanguage.Haskell.TH.Desugar
on GHC 8.0 or later).
In order to support this desugaring, the type signatures of
dsLetDec
anddsLetDecs
now return([DLetDec], DExp -> DExp)
instead of just[DLetDec]
, whereDExp -> DExp
is the expression which binds the values of implicit params (e.g.,\z -> bindIP @"x" new_x_val z
) if any are bound. (If none are bound, this is simply theid
function.) -
Fix a bug in which
toposortTyVarsOf
would error at runtime if given types containingforall
s as arguments. -
Fix a bug in which
fvDType
would return incorrect results if given a type containing quantified constraints. -
Fix a bug in which
expandType
would not expand type synonyms in the kinds of type variable binders inforall
s. -
Fix a bug in which
getRecordSelectors
would omit record selectors from GADT constructors. -
Fix a bug in which
toposortTyVarsOf
would sometimes not preserve the left-to-right ordering ofName
s generated withqNewName
. -
Locally reified class methods, data constructors, and record selectors now quantify kind variables properly.
-
Desugared ADT constructors now quantify kind variables properly.
-
Remove
DPred
, as it has become too similar toDType
. This also means that theDPat
constructors, which previously ended with the suffixPa
, can now use the suffixP
, mirroring TH. -
The type of
applyDType
has changed fromDType -> [DType] -> DType
toDType -> [DTypeArg] -> DType
, whereDTypeArg
is a new data type that encodes whether an argument is a normal type argument (e.g., theInt
inMaybe Int
) or a visible kind argument (e.g., the@Type
inProxy @Type Char
). ATypeArg
data type (which is likeDTypeArg
, but withType
s/Kind
s instead ofDType
s/DKind
s) is also provided.A handful of utility functions for manipulating
TypeArg
s andDTypeArg
s are also exported. -
th-desugar
functions that compute free variables (e.g.,fvDType
) now return anOSet
, a variant ofSet
that remembers the order in which elements were inserted. A consequence of this change is that it fixes a bug that causes free variables to be computed in different orders depending on which unique numbers GHC happened to generate internally. -
Substition and type synonym expansion are now more efficient by avoiding the use of
syb
in inner loops.
Version 1.9
-
Suppose GHC 8.6.
-
Add support for
DerivingVia
. Correspondingly, there is now aDDerivStrategy
data type. -
Add support for
QuantifiedConstraints
. Correspondingly, there is now aDForallPr
constructor inDPred
to represent quantified constraint types. -
Remove the
DStarT
constructor ofDType
in favor ofDConT ''Type
. Two utility functions have been added toLanguage.Haskell.TH.Desugar
to ease this transition:isTypeKindName
: returnsTrue
if the argumentName
is that ofType
or★
(or*
, to support older GHCs).typeKindName
: the name ofType
(on GHC 8.0 or later) or*
(on older GHCs).
-
th-desugar
now desugars all data types to GADT syntax. The most significant API-facing changes resulting from this new design are:-
The
DDataD
,DDataFamilyD
, andDDataFamInstD
constructors ofDDec
now haveMaybe DKind
fields that either haveJust
an explicit return kind (e.g., thek -> Type -> Type
indata Foo :: k -> Type -> Type
) orNothing
(if lacking an explicit return kind). -
The
DCon
constructor previously had a field of typeMaybe DType
, since there was a possibility it could be a GADT (with an explicit return type) or non-GADT (without an explicit return type) constructor. Since all data types are desugared to GADTs now, this field has been changed to be simply aDType
. -
The type signature of
dsCon
was previously:dsCon :: DsMonad q => Con -> q [DCon]
However, desugaring constructors now needs more information than before, since GADT constructors have richer type signatures. Accordingly, the type of
dsCon
is now:dsCon :: DsMonad q => [DTyVarBndr] -- ^ The universally quantified type variables -- (used if desugaring a non-GADT constructor) -> DType -- ^ The original data declaration's type -- (used if desugaring a non-GADT constructor). -> Con -> q [DCon]
The
instance Desugar [Con] [DCon]
has also been removed, as the previous implementation ofdesugar
(concatMapM dsCon
) no longer has enough information to work.
Some other utility functions have also been added as part of this change:
-
A
conExistentialTvbs
function has been introduced to determine the existentially quantified type variables of aDCon
. Note that this function is not 100% accurate—refer to the documentation forconExistentialTvbs
for more information. -
A
mkExtraDKindBinders
function has been introduced to turn a data type’s return kind into explicit, fresh type variable binders. -
A
toposortTyVarsOf
function, which finds the free variables of a list ofDType
s and returns them in a well scoped list that has been sorted in reverse topological order.
-
-
th-desugar
now desugars partial pattern matches indo
-notation and list/monad comprehensions to the appropriate invocation offail
. (Previously, these were incorrectly desugared intocase
expressions with incomplete patterns.) -
Add a
mkDLamEFromDPats
function for constructing aDLamE
expression using a list ofDPat
arguments and aDExp
body. -
Add an
unravel
function for decomposing a function type into itsforall
’d type variables, its context, its argument types, and its result type. -
Export a
substTyVarBndrs
function fromLanguage.Haskell.TH.Desugar.Subst
, which substitutes over type variable binders in a capture-avoiding fashion. -
getDataD
,dataConNameToDataName
, anddataConNameToCon
fromLanguage.Haskell.TH.Desugar.Reify
now look up local declarations. As a result, the contexts in their type signatures have been strengthened fromQuasi
toDsMonad
. -
Export a
dTyVarBndrToDType
function which converts aDTyVarBndr
to aDType
, which preserves its kind. -
Previously,
th-desugar
would silently accept illegal uses of record construction with fields that did not belong to the constructor, such asIdentity { notAField = "wat" }
. This is now an error.
Version 1.8
-
Support GHC 8.4.
-
substTy
now properly substitutes into kind signatures. -
Expose
fvDType
, which computes the free variables of aDType
. -
Incorporate a
DDeclaredInfix
field intoDNormalC
to indicate if it is a constructor that was declared infix. -
Implement
lookupValueNameWithLocals
,lookupTypeNameWithLocals
,mkDataNameWithLocals
, andmkTypeNameWithLocals
, counterparts tolookupValueName
,lookupTypeName
,mkDataName
, andmkTypeName
which have access to local Template Haskell declarations. -
Implement
reifyNameSpace
to determine aName
’sNameSpace
. -
Export
reifyFixityWithLocals
fromLanguage.Haskell.TH.Desugar
. -
Export
matchTy
(among other goodies) from new moduleLanguage.Haskell.TH.Subst
. This function matches a type template against a target.
Version 1.7
-
Support for TH’s support for
TypeApplications
, thanks to @RyanGlScott. -
Support for unboxed sums, thanks to @RyanGlScott.
-
Support for
COMPLETE
pragmas. -
getRecordSelectors
now requires a list ofDCon
s as an argument. This makes it easier to return correct record selector bindings in the event that a record selector appears in multiple constructors. (See goldfirere/singletons#180 for an example of where the old behavior ofgetRecordSelectors
went wrong.) -
Better type family expansion (expanding an open type family with variables works now).
Version 1.6
-
Work with GHC 8, with thanks to @christiaanb for getting this change going. This means that several core datatypes have changed: partcularly, we now have
DTypeFamilyHead
and fixities are now reified separately from other things. -
DKind
is merged withDType
. -
Generic
instances for everything.
Version 1.5.5
- Fix issue #34. This means that desugaring (twice) is idempotent over expressions, after the second time. That is, if you desugar an expression, sweeten it, desugar again, sweeten again, and then desugar a third time, you get the same result as when you desugared the second time. (The extra round-trip is necessary there to make the output smaller in certain common cases.)
Version 1.5.4.1
- Fix issue #32, concerning reification of classes with default methods.
Version 1.5.4
- Added
expandUnsoundly
Version 1.5.3
- More
DsMonad
instances, thanks to David Fox.
Version 1.5.2
- Sweeten kinds more, too.
Version 1.5.1
-
Thanks to David Fox (@ddssff), sweetening now tries to use more of TH’s
Type
constructors. -
Also thanks to David Fox, depend usefully on the th-orphans package.
Version 1.5
-
There is now a facility to register a list of
Dec
that internal reification should use when necessary. This avoids the user needing to break up their definition across different top-level splices. SeewithLocalDeclarations
. This has a side effect of changing theQuasi
typeclass constraint on many functions to be the newDsMonad
constraint. Happily, there areDsMonad
instances forQ
andIO
, the two normal inhabitants ofQuasi
. -
“Match flattening” is implemented! The functions
scExp
andscLetDec
remove any nested pattern matches. -
More is now exported from
Language.Haskell.TH.Desugar
for ease of use. -
expand
can now expand closed type families! It still requires that the type to expand contain no type variables. -
Support for standalone-deriving and default signatures in GHC 7.10. This means that there are now two new constructors for
DDec
. -
Support for
static
expressions, which are new in GHC 7.10.
Version 1.4.2
expand
functions now consider open type families, as long as the type to be expanded has no free variables.
Version 1.4.1
-
Added
Language.Haskell.TH.Desugar.Lift
, which providesLift
instances for all of the th-desugar types, as well as several Template Haskell types. -
Added
applyDExp
andapplyDType
as convenience functions.
Version 1.4.0
-
All
Dec
s can now be desugared, to the newDDec
type. -
Sweetening
Dec
s that do not exist in GHC 7.6.3- works on a “best effort” basis: closed type families are sweetened to open ones, and role annotations are dropped. -
Info
s can now be desugared. Desugaring takes into account GHC bug #8884, which meant that reifying poly-kinded type families in GHC 7.6.3- was subtly wrong. -
There is a new function
flattenDValD
which takes a binding likelet (a,b) = foo
and breaks it apart into separate assignments fora
andb
. -
There is a new
Desugar
class with methodsdesugar
andsweeten
. See the documentation inLanguage.Haskell.TH.Desugar
. -
Variable names that are distinct in desugared code are now guaranteed to have distinct answers to
nameBase
. -
Added a new function
getRecordSelectors
that extracts types and definitions of record selectors from a datatype definition.
Version 1.3.1
- Update cabal file to include testing files in sdist.
Version 1.3.0
- Update to work with
type Pred = Type
in GHC 7.9. This changed theDPred
type for all GHC versions, though.
Version 1.2.0
- Generalized interface to allow any member of the
Qausi
class, instead of justQ
.
Version 1.1.1
- Made compatible with HEAD after change in role annotation syntax.
Version 1.1
- Added module
Language.Haskell.TH.Desugar.Expand
, which allows for expansion of type synonyms in desugared types. - Added
Show
,Typeable
, andData
instances to desugared types. - Fixed bug where an as-pattern in a
let
statement was scoped incorrectly. - Changed signature of
dsPat
to be more specific to as-patterns; this allowed for fixing thelet
scoping bug. - Created new functions
dsPatOverExp
anddsPatsOverExp
to allow for easy desugaring of patterns. - Changed signature of
dsLetDec
to return a list ofDLetDec
s. - Added
dsLetDecs
for convenience. Now, instead of usingmapM dsLetDec
, you should usedsLetDecs
.
Version 1.0
- Initial release