Smart unfolding support #

@[extern lean_get_structural_rec_arg_pos]

Forward declaration. It is defined in the module src/Lean/Elab/PreDefinition/Structural/Eqns.lean. It is possible to avoid this hack if we move Structural.EqnInfo and Structural.eqnInfoExt to this module.

Instances For

    Add auxiliary annotation to indicate the match-expression e must be reduced when performing smart unfolding.

    Instances For

      Add auxiliary annotation to indicate expression e (a match alternative rhs) was successfully reduced by smart unfolding.

      Instances For

        Helper methods #

        Instances For

          Helper functions for reducing recursors #

          Create the ith projection major. It tries to use the auto-generated projection functions if available. Otherwise falls back to Expr.proj.

          Instances For

            Helper functions for reducing Quot.lift and Quot.ind #

            Helper function for extracting "stuck term" #

            Return some (Expr.mvar mvarId) if metavariable mvarId is blocking reduction.

            Weak Head Normal Form auxiliary combinators #

            @[specialize #[]]

            Auxiliary combinator for handling easy WHNF cases. It takes a function for handling the "hard" cases as an argument

            The "match" compiler uses if-then-else expressions and other auxiliary declarations to compile match-expressions such as

            match v with
            | 'a' => 1
            | 'b' => 2
            | _   => 3

            because it is more efficient than using casesOn recursors. The method reduceMatcher? fails if these auxiliary definitions (e.g., ite) cannot be unfolded in the current transparency setting. This is problematic because tactics such as simp use TransparencyMode.reducible, and most users assume that expressions such as

            match 0 with
            | 0 => 1
            | 100 => 2
            | _ => 3

            should reduce in any transparency mode. Thus, we define a custom canUnfoldAtMatcher predicate for whnfMatcher.

            This solution is not very modular because modifications at the match compiler require changes here. We claim this is defensible because it is reducing the auxiliary declaration defined by the match compiler.

            Alternative solution: tactics that use TransparencyMode.reducible should rely on the equations we generated for match-expressions. This solution is also not perfect because the match-expression above will not reduce during type checking when we are not using TransparencyMode.default or TransparencyMode.all.

            Instances For

              Reduce kernel projection Expr.proj .. expression.

              Instances For
                partial def Lean.Meta.whnfCore (e : Lean.Expr) (deltaAtProj : optParam Bool true) (simpleReduceOnly : optParam Bool false) :

                Apply beta-reduction, zeta-reduction (i.e., unfold let local-decls), iota-reduction, expand let-expressions, expand assigned meta-variables.

                The parameter deltaAtProj controls how to reduce projections s.i. If deltaAtProj == true, then delta reduction is used to reduce s (i.e., whnf is used), otherwise whnfCore.

                If simpleReduceOnly, then iota and projection reduction are not performed. Note that the value of deltaAtProj is irrelevant if simpleReduceOnly = true.

                partial def Lean.Meta.whnfCore.go (deltaAtProj : optParam Bool true) (simpleReduceOnly : optParam Bool false) (e : Lean.Expr) :

                Recall that _sunfold auxiliary definitions contains the markers: markSmartUnfoldingMatch (*) and markSmartUnfoldingMatchAlt (**). For example, consider the following definition

                def r (i j : Nat) : Nat :=
                  i +
                    match j with
                    | => 1
                    | Nat.succ j =>
                      i + match j with
                          | => 2
                          | Nat.succ j => r i j

                produces the following _sunfold auxiliary definition with the markers

                def r._sunfold (i j : Nat) : Nat :=
                  i +
                    (*) match j with
                    | => (**) 1
                    | Nat.succ j =>
                      i + (*) match j with
                          | => (**) 2
                          | Nat.succ j => (**) r i j

                match expressions marked with markSmartUnfoldingMatch (*) must be reduced, otherwise the resulting term is not definitionally equal to the given expression. The recursion may be interrupted as soon as the annotation markSmartUnfoldingAlt (**) is reached.

                For example, the term r i j.succ.succ reduces to the definitionally equal term i + i * r i j

                Instances For

                  Auxiliary method for unfolding a class projection.

                  Auxiliary method for unfolding a class projection. when transparency is set to TransparencyMode.instances. Recall that class instance projections are not marked with [reducible] because we want them to be in "reducible canonical form".

                  Unfold definition using "smart unfolding" if possible.

                  @[specialize #[]]

                  Try to reduce matcher/recursor/quot applications. We say they are all "morally" recursor applications.

                  Instances For
                    Instances For
                      Instances For
                        @[implemented_by Lean.Meta.reduceBoolNativeUnsafe]
                        @[implemented_by Lean.Meta.reduceNatNativeUnsafe]
                        Instances For
                          @[export lean_whnf]

                          If e is a projection function that satisfies p, then reduce it

                          Instances For