fix: `grind` normalizer no longer rewrites `(∀ x, p x) → q` into a disjunction In `grind`, an implication `p → q` is not a case-split candidate like `p ∨ q`, unless `+splitImp` is used. Since `grind` also has few propagation rules for assigning `True` to a universal quantifier, it could do little with `(∀ x, p x) → q`, and the normalizer used `Grind.forall_imp_eq_or` to rewrite it into `(∃ x, ¬p x) ∨ q`. This rule negates a term that is already normalized. So an implication that only becomes visible under a negation during normalization, for example through a `[grind norm]` rule, was negated twice, and the result was not the normal form of the syntactic implication. With an arithmetic guard the hypothesis `∀ i, i < n → p i` became `∀ i, n ≤ i ∨ p i`, which has no useful E-matching pattern. The normalizer now keeps the implication, and an implication whose antecedent is a universal quantifier is a case-split candidate, as it already is when the antecedent is an arithmetic predicate. `foldl_rel` in `grind_list2.lean` needs `ematch := 6`: the falsified antecedent is now obtained by propagation, not by a case split, so the E-matching round counter is not reset. The proof in `grind_13877.lean` no longer needs case splits. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This is the repository for Lean 4.
See Install Lean.
Please read our Contribution Guidelines first.
See Building Lean.