wiki:TypeFunctionsStatus

Version 282 (modified by chak, 6 years ago) (diff)

--

TypeFunctions/Status

Type Functions: Implementation Status

Open Trac bugs related to type families

  • #1754 & #1808 (high priority - people stumble over it often)
  • #1948
  • #1897 & #1900
  • #1834
  • #1809
  • #1775
  • #1716 (maybe the same problem as #1754)
  • #1772
  • #1815 (type families & GADTs)
  • #1722 (type families & GADTs) [look at when GADTs are implemented by equalities]
  • #1723 (type families & GADTs) [will be fixed when GADTs are implemented by equalities; we'll want to add the test case to the testsuite]
  • #1716 (bogus evidence generation with type equalities)
  • #1769 (deriving typeable for data families)

Failing testsuite tests

All these tests are in testsuite/tests/ghc-regress/indexed-types:

  • should_run/GMapAssoc(profc,profasm) (data type families)
  • should_run/GMapTop(profc,profasm) (data type families)
  • should_run/ind2(profc,profasm) (data type families)
  • should_run/Simple12(normal,optc,profc,profasm) (type synonym families)

Check whether these still fail.

Debugging of type families:

  1. Replacing GADT refinements by explicit equality constraints:
    • Impact on testsuite:
      == indexed-types/ ==
      Unexpected passes:
         GADT4(normal)
         GADT5(normal)
         InstEqContext2(normal)
      Unexpected failures:
         GADT3(normal) -- ok, just tickles a known bug
      
      == typecheck/ ==
      Unexpected failures:
         tcfail167(normal)    -- Doesn't produce inaccessible case alternative message anymore.
      
      == gadt/ ==
      Unexpected failures:
         gadt21(normal)       -- OK!  (Just be a different error message.)
         lazypatok(normal)  -- Need to fix this, but low priority.
         nbe(normal)
         while(normal)
      
    • Handling of cases expression scrutinising GADTs:
      • Remove the dodgy rigidity test that is in tcConPat right now.
      • implement proposal where we infer a rigidity flag for case scutinees and pass that down when type checking the patterns,
      • We infer the rigidity flag for the case scrutinee by generalising its type and checking whether that has an foralls at the top. It's rigid if it has no foralls.
      • if a pattern has a GADT constructor (ie, any constraints in the data constructor signature), the scutinee must be rigid,
      • we need to know of types whether they are rigid (not only whether they contain unification variables, but by a flag in the environment that indicates whether the computation of that type involved non-rigid type variables)
    • In TcUnify, make all occurs checks more elaborate. They should only defer if the checked variable occurs as part of an argument to a type family application; in other cases, still fail right away. DONE?
    • TcGadt.tcUnifyTys can now probably be replaced again by the non-side-effecting unifier that was in types/Unify.hs (recover from previous repo states).
    • CLEANUP:
      • TcPat.refineAlt: This function is now dead code, so is all its support code.
      • pat_reft field of TcPat.PatState: Not needed anymore and code maintaining can go, too.
      • We can remove the CoVars and Refinement argument of TcSimplify.tcSimplifyCheckPat.
    • Re tcfail167, SPJ proposes that could generate a better error message, at least most of the time. If the "expected type" of a pattern is 's', and we meet a constructor with result type (T t1 ..tn), then one could imagine a 2-step process:
      1. check that 's' is (or can be made to be) of form (T ....)
      2. check that the ... can be unified with t1..tn
      If (1) succeeds but (2) fails, the alternative is in accessible. Of course, (2) might fail "later" by generating a constraint that later can't be satisfied, and we won't report that well, but we'd get a good message in the common fails-fast case. We could even improve the message from (1) to say: "Constructor C is from data type T, but a pattern of type s is expected.
  2. substEqInDict needs to be symmetric (i.e., also apply right-to-left rules); try to re-use existing infrastructure. It would be neater, easier to understand, and more efficient to have one loop that goes for a fixed point of simultaneously rewriting with given_eqs, wanted_eqs, and type instances.
  3. skolemOccurs for wanteds? At least F a ~ [G (F a)] and similar currently result in an occurs check error. Without skolemOccurs in wanted, the occurs check for wanted would need to be smarter (and just prevent cyclic substitutions of the outlined form silently). However, when inferring a type, having the rewrites enabled by skolemOccurs available will leads to potentially simpler contexts.
  4. Comments:
    • When we raise a mismatch error in TcSimplify for unresolvable equalities, we effectively tidy the two non-matching types twice. Add a comment to highlight this and say way it is ok (i.e., they are never grouped together with groupErrs or similar).
  5. :t in ghci doesn't print equalities in contexts properly.
  6. ghci command to print normalised type and add http://article.gmane.org/gmane.comp.lang.haskell.cafe/28799 as a test to the testsuite.
  7. Check that the restrictions on equality constraints in instance and class contexts are enforced. We should have tests for that in the testsuite. Document the exact restrictions on the Haskell wiki tutorial page.
  8. When can foralls appear in equalities? What constraints does that place on GADTs? Also, the code in TcTyFuns doesn't really deal with rank-n types properly, esp decompRule.
  9. To fix Simple8:
    • Fix tcLookupFamInst to gracefully handle this case. (This requires some care to not violate assumptions made by other clients of this function, as it is also used for data families, but I see no fundamental problem.)
    • Issue a warning if there are two identical instances (as per Roman's suggestion).
  10. CONCEPTUAL issue: At least with skolemOccurs, the policy of not zonking the types embedded in the kinds of coercion type variables does no longer work. This becomes, for example in the test Simple13, apparent. The skolem introduced in skolemOccurs finds its way into variable kinds (which is visible when inspecting them during TcMType.zonk_tc_tyvar).
  11. When Simple13 is compiled with a compiler that was built with -DDEBUG, it prints a warning about not matching types being used during constructing a trans coercion.
  12. To fix superclass equalities (specifically getting the coercion evidence), we could introduce a kind of typelet just for evidence. In fact, re-use HsBind.VarBind and make its right-hand side a specially data structure describing evidence construction, instead of being a general HsExpr. That evidence construction generation can have a case for extracting superclass constraints. The desugarer than has to generate the case expression bringing the equality in scope from that.
  13. In TcTyFuns.genericNormaliseInst, we need to figure out what to do with ImplicInst, Method, and LitInst dictionaries.
  14. ghc falls over if a bang pattern is put at an argument of type F a.
  15. Fix export list problem (ie, export of data constructors introduced by orphan data instances):
    • Change HscTypes.IfaceExport to use Name instead of OccName.
    • Then, there is also no need for the grouping of the identifiers by module anymore (but sort it to avoid spurious iface changes dur to re-ordering when re-compiling).
    • We still need to have the name parent map, though.
    • See email for example.
  16. Allow data family GADT instances:
    • Need to check the result types of the data constructors, probably in checkValidDataCon.
    • tcFamInstDecl1 needs to allow family GADT instances.
  17. Eliminate code duplication between tcTyClDecl1 and tcFamInstDecl1. The code for vanilla data/newtype declarations and the code for data/newtype instances has many commonalities.
  18. Fix everything in the testsuite.
  19. Can't we now allow non-left-linear declarations; e.g., instance type F a a = ..?
  20. The tests tcfail068 and rw used to raise more type errors right away. Now, we see less recovery.
  21. What about filtering the EqInsts in TcSimplify.addSCs. We need them, don't we? But they give rise to Vars, not Ids, and we haven't got selectors.
  22. Consider
    type family F a
    data T a b = MkT1 { fa :: F a, fb :: b }
    upd t x = t { fb = x }
    
    What is the most general type of upd? It's
    upd :: (F a ~ F d) => T a b -> c -> T d c
    
    However, we currently insist on the less general
    upd :: T a b -> c -> T a c
    
    It seems a bit complicated to come up with the most general type. THe relevant code is in TcExpr.tcExpr in STEP 4 of the RecordUpd case.
  23. Can we support
    {-# LANGUAGE TypeFamilies, TypeOperators, GADTs,  RankNTypes, FlexibleContexts #-}
    module Equality( (:=:), eq_elim, eq_refl ) where
    
    data a:=: b where
      EQUAL :: a :=: a
    
    eq_refl :: a :=: a
    eq_refl = EQUAL
    
    eq_elim :: (a~b) => a :=: b -> (a~b => p) -> p
    eq_elim EQUAL p = p 
    

Current:

  • Add some trac wiki documentation of how inference with type families works.

Parsing and Renaming

Todo (low-level): None.

Todo (high-level):

  1. Defaults for associated type synonyms. (Having both a kind signature and vanilla synonym is problematic as in RnNames.getLocalDeclBinders its hard to see that not both of them are defining declarations, which leads to a multiple declarations error. Defaults are quite different from vanilla synonyms anyway, as they usually have tyvars on their rhs that do not occur on the lhs.)

Done:

  • Parsing and renaming of kind signatures (toplevel and in classes).
  • Parsing and renaming of indexed type declarations (toplevel and in classes).
  • Using new syntax with family and instance on top level.
  • Added -findexed-types switch.
  • Allowing type tag in export lists to list associated types in the sub-binder list of an import/export item for a class.
  • Import/export lists: ATs can be listed as subnames of classes and the data constructors of instances of a data family are subnames of that family.
  • Parsing and renaming of equational constraints in contexts.

Type Checking

Todo (low-level):

  • Allow data family GADT instances.
  • Deriving Typeable for data families (#1769)
  • If an associated synonym has a default definition, use that in the instances. In contrast to methods, this cannot be overridden by a specialised definition. (Confluence requires that any specialised version is extensionally the same as the default.)

Todo (high-level):

  1. Type checking of type families; routines in TcUnify that still need to be extended:
    • boxySplitTyConApp: The second argument (BoxyRhoType) can be a synonym family application. Then, we must produce a wanted coercion and return a HsWrapper value that applies that coercion.
    • boxySplitAppTy: Basically, the same deal as the previous.
  2. Type checking in the presence of associated synonym defaults. (Default AT synonyms are only allowed for ATs defined in the same class.)
  3. Type check functional dependencies as type functions.

Done:

  • Kind and type checking of kind signatures.
  • Kind and type checking of instance declarations of indexed types, including the generation of representation tycons.
  • Wrapper generation and type checking of pattern matching for indexed data and newtypes.
  • Consistency checking for family instances.
  • Enforce syntactic constraints on type instances needed to ensure the termination of constraint entailment checking.

Desugaring

Todo (low-level): None.

Todo (high-level): None.

Done:

  • Representation of family kind signatures as TyCon.TyCons.
  • Extension of Class.Class by associated TyCons.
  • Extension of TyCon.TyCon with a reference to the parent TyCon for data instances.
  • Extension of DataCon.DataCon with instance types for constructors belonging to data instances.
  • Extension of TyCon.TyCon such that the parent of a data instance is paired with a coercion identifying family instance and representation type.
  • For indexed data types, the datacon wrapper uses data instance coercion and pattern matching casts the scrutinee via an ExprCoFn in a CoPat.
  • Import and exporting.
  • Generation and plumbing through of rough matches.
  • Equational constraints in contexts.