th-abstraction

Nicer interface for reified information about data types

https://github.com/glguy/th-abstraction

Version on this page:0.7.0.0
LTS Haskell 23.1:0.7.1.0
Stackage Nightly 2024-12-21:0.7.1.0
Latest on Hackage:0.7.1.0

See all snapshots th-abstraction appears in

ISC licensed by Eric Mertens
Maintained by [email protected]
This version can be pinned in stack with:th-abstraction-0.7.0.0@sha256:4d80fe53ed140866018278c81e1c0c9bc8fc73d8f2a530321a588f24be491fc4,2248

Module documentation for 0.7.0.0

th-abstraction

Hackage Build Status

This package provides a consistent interface to a subset of Template Haskell.

Currently the package provides a consistent view of the reified declaration information about datatypes, newtypes, and data family instances. These interfaces abstract away the differences in the normal and GADT syntax used to define these types.

Contact Information

Please contact me via GitHub or on the #haskell IRC channel on irc.libera.chat

Changes

Revision history for th-abstraction

0.7.0.0 – 2024.03.17

  • DatatypeInfo now has an additional datatypeReturnKind field. Most of the time, this will be StarT, but this can also be more exotic kinds such as ConT ''UnliftedType if dealing with primitive types, UnliftedDatatypes, or UnliftedNewtypes.
  • reifyDatatype and related functions now support primitive types such as Int#. These will be reified as DatatypeInfos with no ConstructorInfos and with Datatype as the datatypeVariant.
  • normalizeCon now takes a Kind argument representing the return kind of the parent data type. (This is sometimes necessary to determine which type variables in the data constructor are universal or existential, depending on if the variables appear in the return kind.)
  • Fix a couple of bugs in which normalizeDec would return incorrect results for GADTs that use foralls in their return kind.

0.6.0.0 – 2023.07.31

  • Support building with template-haskell-2.21.0.0 (GHC 9.8).

  • Adapt to TyVarBndrs for type-level declarations changing their type from TyVarBndr () to TyVarBndr BndrVis in template-haskell:

    • Language.Haskell.TH.Datatype.TyVarBndr now backports type BndrVis = (), as well as BndrReq and BndrInvis pattern synonyms. These make it possible to write code involving BndrVis that is somewhat backwards compatible (but do see the caveats in the Haddocks for BndrInvis).
    • Language.Haskell.TH.Datatype.TyVarBndr also backports the following definitions:
      • The type TyVarBndrVis = TyVarBndr BndrVis type synonym.
      • The DefaultBndrFlag class, which can be used to write code that is polymorphic over TyVarBndr flags while still allowing the code to return a reasonable default value for the flag.
      • The bndrReq and bndrInvis definitions, which behave identically to BndrReq and BndrInvis.
    • Language.Haskell.TH.Datatype.TyVarBndr now defines the following utility functions, which are not present in template-haskell:
      • plainTVReq, plainTVInvis, kindedTVReq, and kindedTVInvis functions, which construct PlainTVs and KindedTVs with particular BndrVis flags.
      • An elimTVFlag, which behaves like elimTV, but where the continuation arguments also take a flag argument. (Note that the type of this function is slightly different on old versions of template-haskell. See the Haddocks for more.)
      • A tvFlag function, which extracts the flag from a TyVarBndr. (Note that the type of this function is slightly different on old versions of template-haskell. See the Haddocks for more.)
    • The types of the dataDCompat and newtypeDCompat functions have had their [TyVarBndrUnit] arguments changed to [TyVarBndrVis], matching similar changes to DataD and NewtypeD in template-haskell.

    Because BndrVis is a synonym for () on pre-9.8 versions of GHC, this change is unlikely to break any existing code, provided that you build it with GHC 9.6 or earlier. If you build with GHC 9.8 or later, on the other hand, it is likely that you will need to update your existing code. Here are some possible ways that your code might fail to compile with GHC 9.8, along with some migration strategies:

    • Your code passes a TyVarBndrUnit in a place where a TyVarBndrVis is now expected in GHC 9.8, such as in the arguments to dataDCompat:

      import "template-haskell" Language.Haskell.TH
      import "th-abstraction"   Language.Haskell.TH.Datatype (dataDCompat)
      
      dec :: DecQ
      dec = dataDCompat (pure []) d [PlainTV a ()] [] []
        where
          d = mkName "d"
          a = mkName "a"
      

      With GHC 9.8, this will fail to compile with:

      error: [GHC-83865]
          • Couldn't match expected type ‘BndrVis’ with actual type ‘()’
          • In the second argument of ‘PlainTV’, namely ‘()’
            In the expression: PlainTV a ()
            In the third argument of ‘dataDCompat’, namely ‘[PlainTV a ()]’
        |
        | dec = dataDCompat (pure []) d [PlainTV a ()] [] []
        |                                          ^^
      

      Some possible ways to migrate this code include:

      • Use the bndrReq function or BndrReq pattern synonym in place of (), making sure to import them from Language.Haskell.TH.Datatype.TyVarBndr:

        ...
        import "th-abstraction" Language.Haskell.TH.Datatype.TyVarBndr
        
        dec :: DecQ
        dec = dataDCompat (pure []) d [PlainTV a bndrReq] [] []
        -- Or, alternatively:
        {-
        dec = dataDCompat (pure []) d [PlainTV a BndrReq] [] []
        -}
          where
            ...
        
      • Use the plainTV function from Language.Haskell.TH.Datatype.TyVarBndr, which is now sufficiently polymorphic to work as both a TyVarBndrUnit and a TyVarBndrVis:

        ...
        import Language.Haskell.TH.Datatype.TyVarBndr
        
        dec :: DecQ
        dec = dataDCompat (pure []) d [plainTV a] [] []
          where
            ...
        
    • You may have to replace some uses of TyVarBndrUnit with TyVarBndrVis in your code. For instance, this will no longer typecheck in GHC 9.8 for similar reasons to the previous example:

      import "template-haskell" Language.Haskell.TH
      import "th-abstraction"   Language.Haskell.TH.Datatype (dataDCompat)
      
      dec :: DecQ
      dec = dataDCompat (pure []) d tvbs [] []
        where
          tvbs :: [TyVarBndrUnit]
          tvbs = [plainTV a]
      
          d = mkName "d"
          a = mkName "a"
      

      Here is a version that will typecheck with GHC 9.8 and earlier:

      ...
      import "th-abstraction" Language.Haskell.TH.Datatype.TyVarBndr
      
      dec :: DecQ
      dec = dataDCompat (pure []) d tvbs [] []
        where
          tvbs :: [TyVarBndrVis]
          tvbs = [plainTV a]
      
          ...
      
    • In some cases, the TyVarBndrUnits might come from another place in the code, e.g.,

      import "template-haskell" Language.Haskell.TH
      import "th-abstraction"   Language.Haskell.TH.Datatype (dataDCompat)
      
      dec :: [TyVarBndrUnit] -> DecQ
      dec tvbs = dataDCompat (pure []) d tvbs [] []
        where
          d = mkName "d"
      

      If it is not straightforward to change dec’s type to accept [TyVarBndrVis] as an argument, another viable option is to use the changeTVFlags function:

      ...
      import "th-abstraction" Language.Haskell.TH.Datatype.TyVarBndr
      
      dec :: [TyVarBndrUnit] -> DecQ
      dec tvbs = dataDCompat (pure []) d tvbs' [] []
        where
          tvbs' :: [TyVarBndrVis]
          tvbs' = changeTVFlags bndrReq tvbs
      
          ...
      

    This guide, while not comprehensive, should cover most of the common cases one will encounter when migrating their th-abstraction code to support GHC 9.8.

0.5.0.0 – 2023.02.27

  • Support the TypeData language extension added in GHC 9.6. The DatatypeVariant data type now has a separate TypeData constructor to represent type data declarations.
  • Add a Lift instance for th-abstraction’s compatibility shim for Specificity when building with pre-9.0 versions of GHC.

0.4.5.0 – 2022.09.12

  • Fix a bug in which data family declarations with interesting return kinds (e.g., data family F :: Type -> Type) would be reified incorrectly when using reifyDatatype.

0.4.4.0 – 2022.07.23

  • Support free variable substitution and infix resolution for PromotedInfixT and PromotedUInfixT on template-haskell-2.19.0.0 or later.

0.4.3.0 – 2021.08.30

  • Make applySubstitution avoid capturing type variable binders when substituting into foralls.
  • Fix a bug in which resolveTypeSynonyms would incorrectly expand type synonyms that are not applied to enough arguments.
  • Allow the test suite to build with GHC 9.2.

0.4.2.0 – 2020-12-30

  • Explicitly mark modules as Safe (or Trustworthy for GHC versions prior to 8.4).

0.4.1.0 – 2020-12-09

  • Fix a bug in which normalizeDec would give incorrect kind annotations to type variables in quoted Decs. normalizeDec now leaves the kinds of type variable binders alone.

0.4.0.0 – 2020-09-29

  • Adapt to the TyVarBndr data type gaining a new flag type parameter (in template-haskell-2.17.0.0) to represent its specificity:
    • Introduce a new Language.Haskell.TH.Datatype.TyVarBndr module that defines TyVarBndr_, a backwards-compatible type synonym for TyVarBndr, as well as backporting TyVarBndrSpec, TyVarBndrUnit, and Specificity. This module also defines other useful functions for constructing and manipulating TyVarBndrs.
    • The types in Language.Haskell.TH.Datatype now use TyVarBndr_, TyVarBndrUnit, and TyVarBndrSpec where appropriate. Technically, this is not a breaking change, since all three are simple type synonyms around TyVarBndr, but it is likely that you will need to update your th-abstraction-using code anyway if it involves a TyVarBndr-consuming function.

0.3.2.0 – 2020-02-06

  • Support substituting into and extracting free variables from ForallVisTs on template-haskell-2.16.0.0 (GHC 8.10) or later.
  • Fix a bug in which freeVariables could report duplicate kind variables when they occur in the kinds of the type variable binders in a ForallT.
  • Fix a bug in which resolveInfixT would not resolve UInfixTs occurring in the kinds of type variable binders in a ForallT.
  • Fix a bug in which the TypeSubstitution ConstructorInfo instance would not detect free kind variables in the constructorVars.

0.3.1.0 – 2019-04-28

  • Fix a bug which would cause data family information to be reified incorrectly with GHC 8.8+ in some situations.

0.3.0.0 – 2019-04-26

  • Breaking change: the datatypeVars field of DatatypeInfo is now of type [TyVarBndr] instead of [Type], as it now refers to all of the bound type variables in the data type. The old datatypeVars field has been renamed to datatypeInstTypes to better reflect its purpose.

    In addition, the type of normalizeCon now has an additional [TyVarBndr] argument, since DatatypeInfo now requires it.

  • Support template-haskell-2.15.

  • Fix a bug in which normalizeDec would not detect existential type variables in a GADT constructor if they were implicitly quantified.

  • Fix a bug in which normalizeDec would report an incorrect number of datatypeVars for GADT declarations with explicit return kinds (such as data Foo :: * -> * where).

0.2.11.0 – 2019-02-26

  • Fix a bug in which freeVariablesWellScoped would sometimes not preserve the left-to-right ordering of Names generated with newName.

0.2.10.0 – 2018-12-20

  • Optimization: quantifyType now collapses consecutive foralls. For instance, calling quantifyType on forall b. a -> b -> T a now produces forall a b. a -> b -> T a instead of forall a. forall b. a -> b -> T a.

0.2.9.0 – 2018-12-20

  • Fix a bug in which resolveTypeSynonyms would not look into ForallTs, SigTs, InfixTs, or ParensTs.
  • Fix a bug in which quantifyType would not respect the dependency order of type variables (e.g., Proxy (a :: k) would have erroneously been quantified as forall a k. Proxy (a :: k)).
  • Fix a bug in which asEqualPred would return incorrect results with GHC 8.7.
  • Add a freeVariablesWellScoped function which computes the free variables of a list of types and sorts them according to dependency order.
  • Add a resolveKindSynonyms function which expands all type synonyms in a Kind. This is mostly useful for supporting old GHCs where Type and Kind were not the same.

0.2.8.0 – 2018-06-29

  • GADT reification is now much more robust with respect to PolyKinds:
    • A bug in which universally quantified kind variables were mistakenly flagged as existential has been fixed.

    • A bug in which the kinds of existentially quantified type variables were not substituted properly has been fixed.

    • More kind equalities are detected than before. For example, in the following data type:

      data T (a :: k) where
        MkT :: forall (a :: Bool). T a
      

      We now catch the k ~ Bool equality.

  • Tweak resolveTypeSynonyms so that failing to reify a type constructor name so longer results in an error. Among other benefits, this makes it possible to pass data types with GADT syntax to normalizeDec.

0.2.7.0 – 2018-06-17

  • Fix bug in which data family instances with duplicate occurrences of type variables in the left-hand side would have redundant equality constraints in their contexts.

0.2.6.0 – 2017-09-04

  • Fix bug in which applySubstitution and freeVariables would ignore type variables in the kinds of type variable binders.

0.2.5.0

  • Added pragLineDCompat, newtypeDCompat and tySynInstDCompat

0.2.4.0 – 2017-07-31

  • Fix bug that caused GADT equality constraints to be incorrect in some cases.
  • Expose Unpackedness and Strictness (which were unexported by accident).

0.2.3.0 – 2017-06-26

  • Add resolvePredSynonyms
  • Add reifyConstructor, which allows reification of ConstructorInfo from a constructor name, and lookupByConstructorName, which allows directly looking up a ConstructorInfo from a DatatypeInfo value for a given constructor Name.
  • Augment reifyDatatype to be able to look up DatatypeInfo from the Name of a record selector for one of its constructors. Also add reifyRecord for reification of of ConstructorInfo from a record name, and lookupByRecordName, which allows directly looking up a ConstructorInfo from a DatatypeInfo value for a given record Name.
  • Fix bug that caused th-abstraction to fail on GHC 7.0 and 7.2 when passing a vanilla constructor name to reifyDatatype
  • Make normalizeDec and normalizeCon more robust with respect to data family instances on GHC 7.6 and 7.8

0.2.2.0 – 2017-06-10

  • Fix freeVariables on lists not not produce duplicates.

0.2.1.0 – 2017-06-09

  • Add sensible reify defaults and error messages when we can’t backport fixes to old GHC Template Haskell output due to hand-written Decs being processed.

0.2.0.0 – 2017-06-03

  • Added reifyFixityCompat
  • Added constructorStrictness field to ConstructorInfo
  • Infer more kind signatures when missing on old GHCs
  • Added parameter to normalizeCon
  • Support GHC back to 7.0.4

0.1.3.0 – 2017-05-27

  • Added resolveInfixT which uses reified fixity information to resolve UInfixT
  • Added asEqualPred and asClassPred
  • Fixed data-instance GADTs

0.1.2.1 – 2017-05-21

  • Add eta reduction fixes to GHC 7.6

0.1.2.0 – 2017-05-21

  • Added arrowKCompat
  • Added workaround for GHC 7.8 data instance eta reduction bug
  • Added kind signatures to datatypeVars

0.1.1.0 – 2017-05-20

  • Better matching of constraints generated for GADTs across GHC versions
  • Added dataDCompat
  • Support for giving value constructors to reifyDatatype. This enables data families to be reified easily.

0.1.0.0 – 2017-04-26

  • First version.