import Mathlib.Lean.Expr.Basic import Lean.Elab.Tactic.Basic import Mathlib.Init.Data.Nat.Notation open Lean.Meta Lean.Elab.Tactic Lean.Parser.Tactic /-! # Modified `induction` tactic Modify `induction` tactic to use the cases `0` and `n + 1` (isnstead of `zero` and `succ n`) and support the lean3-style `with` keyword. This is mainly copied and modified from the mathlib-tactic `induction'`. -/ /-- Custom induction principle that uses `0` and `n + 1` instead of `Nat.zero` and `Nat.succ n`. -/ def CustomTactic.rec' {P : ℕ → Prop} (zero : P 0) (succ : (n : ℕ) → (n_ih : P n) → P (n + 1)) (t : ℕ) : P t := by induction t with | zero => assumption | succ n => apply succ assumption namespace Lean.Parser.Tactic open Meta Elab Elab.Tactic open private getAltNumFields in evalCases ElimApp.evalAlts.go in def _root_.CustomTactic.ElimApp.evalNames (elimInfo : ElimInfo) (alts : Array ElimApp.Alt) (withArg : Syntax) (numEqs := 0) (numGeneralized := 0) (toClear : Array FVarId := #[]) : Lean.Elab.Term.TermElabM (Array MVarId) := do let mut names : List Syntax := withArg[1].getArgs |>.toList let mut subgoals := #[] for { name := altName, mvarId := g, .. } in alts do let numFields ← getAltNumFields elimInfo altName let (altVarNames, names') := names.splitAtD numFields (Unhygienic.run `(_)) names := names' let (fvars, g) ← g.introN numFields <| altVarNames.map (getNameOfIdent' ·[0]) let some (g, subst) ← Cases.unifyEqs? numEqs g {} | pure () let (_, g) ← g.introNP numGeneralized let g ← liftM $ toClear.foldlM (·.tryClear) g for fvar in fvars, stx in altVarNames do g.withContext <| (subst.apply <| .fvar fvar).addLocalVarInfoForBinderIdent ⟨stx⟩ subgoals := subgoals.push g pure subgoals open private getElimNameInfo generalizeTargets generalizeVars in evalInduction in /-- Modified `induction` tactic for this game. Usage: `induction n with d hd`. For this game, `induction n` uses the cases `0` and `n + 1` instead of `Nat.zero` and `Nat.succ n`. These are defEq, but it's currently annoying that `ring` does not like the latter forms. *(The actual `induction` tactic has a more complex `with`-argument that works differently)* -/ elab (name := _root_.MyNat.induction) "induction " tgts:(casesTarget,+) withArg:((" with " (colGt binderIdent)+)?) : tactic => do let targets ← elabCasesTargets tgts.1.getSepArgs let g :: gs ← getUnsolvedGoals | throwNoGoalsToBeSolved g.withContext do let elimInfo ← getElimInfo `CustomTactic.rec' let targets ← addImplicitTargets elimInfo targets evalInduction.checkTargets targets let targetFVarIds := targets.map (·.fvarId!) g.withContext do let forbidden ← mkGeneralizationForbiddenSet targets let mut s ← getFVarSetToGeneralize targets forbidden let (fvarIds, g) ← g.revert (← sortFVarIds s.toArray) let result ← withRef tgts <| ElimApp.mkElimApp elimInfo targets (← g.getTag) let elimArgs := result.elimApp.getAppArgs ElimApp.setMotiveArg g elimArgs[elimInfo.motivePos]!.mvarId! targetFVarIds g.assign result.elimApp let subgoals ← CustomTactic.ElimApp.evalNames elimInfo result.alts withArg (numGeneralized := fvarIds.size) (toClear := targetFVarIds) setGoals <| (subgoals ++ result.others).toList ++ gs end Lean.Parser.Tactic