3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2025-10-31 19:52:29 +00:00

add comments

Signed-off-by: Nikolaj Bjorner <nbjorner@microsoft.com>
This commit is contained in:
Nikolaj Bjorner 2025-10-16 08:53:28 +02:00
parent 2bb22c6489
commit d0a7b19806

View file

@ -25,6 +25,10 @@ Revision History:
namespace smt {
/**
Constructor.
Set up callback that adds axiom instantiations as clauses.
**/
theory_finite_set::theory_finite_set(context& ctx):
theory(ctx, ctx.get_manager().mk_family_id("finite_set")),
u(m),
@ -38,6 +42,24 @@ namespace smt {
m_axioms.set_add_clause(add_clause_fn);
}
/**
* Boolean atomic formulas for finite sets are one of:
* (set.in x S)
* (set.subset S T)
* When an atomic formula is first created it is to be registered with the solver.
* The internalize_atom method takes care of this.
* Atomic formulas are special cases of terms (of non-Boolean type) so the first
* effect is to register the atom as a term.
* The second effect is to set up tracking and assert axioms.
* Tracking:
* For every occurrence (set.in x_i S_i) we track x_i.
* Axioms:
* We can immediately assert some axioms because they are unit literals:
* - (set.in x set.empty) is false
* - (set.subset S T) <=> (= (set.union S T) T) (or (= (set.intersect S T) S))
* Axioms can be deffered to when the atomic formulas become "relevant" for the theory solver.
*
*/
bool theory_finite_set::internalize_atom(app * atom, bool gate_ctx) {
TRACE(finite_set, tout << "internalize_atom: " << mk_pp(atom, m) << "\n";);
@ -52,10 +74,22 @@ namespace smt {
ctx.push_trail(insert_obj_trail(m_elements, n));
}
}
// Assert immediate axioms
// add_immediate_axioms(atom);
return true;
}
/**
* When terms are registered with the solver , we need to ensure that:
* - All subterms have an associated E-node
* - Boolean terms are registered as boolean variables
* Registering a Boolean variable ensures that the solver will be notified about its truth value.
* - Non-Boolean terms have an associated theory variable
* Registering a theory variable ensures that the solver will be notified about equalities and disequalites.
* The solver can use the theory variable to track auxiliary information about E-nodes.
*/
bool theory_finite_set::internalize_term(app * term) {
TRACE(finite_set, tout << "internalize_term: " << mk_pp(term, m) << "\n";);
@ -95,6 +129,17 @@ namespace smt {
// For now, we rely on the final_check to handle this
}
/**
* Final check for the finite set theory.
* The Final Check method is called when the solver has assigned truth values to all Boolean variables.
* It is responsible for asserting any remaining axioms and checking for inconsistencies.
*
* It ensures saturation with respect to the theory axioms:
* - Set membership is saturated with respect to set operations.
* For every (set.in x S) where S is a union, assert (or propagate) (set.in x S1) or (set.in x S2)
* - It saturates with respect to extensionality:
* Sets corresponding to shared variables having the same interpretation should also be congruent
*/
final_check_status theory_finite_set::final_check_eh() {
TRACE(finite_set, tout << "final_check_eh\n";);
@ -102,6 +147,7 @@ namespace smt {
// if a parent is of the form elem' in S u T, or similar.
// create clauses for elem in S u T.
// Saturate membership constraints
expr* elem1 = nullptr, *set1 = nullptr;
for (auto elem : m_elements) {
if (!ctx.is_relevant(elem))
@ -124,9 +170,13 @@ namespace smt {
if (instantiate_free_lemma())
return FC_CONTINUE;
// TODO: Extensionality axioms for sets
return FC_DONE;
}
/**
* Instantiate axioms for a given element in a set.
*/
void theory_finite_set::instantiate_axioms(expr* elem, expr* set) {
TRACE(finite_set, tout << "instantiate_axioms: " << mk_pp(elem, m) << " in " << mk_pp(set, m) << "\n";);
@ -206,6 +256,12 @@ namespace smt {
return nullptr;
}
/**
* Lemmas that are currently assinged to false are conflicts.
* They should be asserted as soon as possible.
* Only the first conflict needs to be asserted.
*
*/
bool theory_finite_set::instantiate_false_lemma() {
for (auto const& clause : m_lemmas) {
bool all_false = all_of(clause, [&](expr *e) { return ctx.find_assignment(e) == l_false; });
@ -217,7 +273,15 @@ namespace smt {
return false;
}
/**
* Lemmas that are unit propagating should be asserted as possible and can be asserted in a batch.
* It is possible to assert a unit propagating lemma as a clause.
* A more efficient approach is to add a Theory propagation with the solver.
* A theory propagation gets recorded on the assignment trail and the overhead of undoing it is baked in to backtracking.
* A theory axiom is also removed during backtracking.
*/
bool theory_finite_set::instantiate_unit_propagation() {
bool propagaed = false;
for (auto const &clause : m_lemmas) {
expr *undef = nullptr;
bool is_unit_propagating = true;
@ -237,11 +301,17 @@ namespace smt {
if (!is_unit_propagating || undef == nullptr)
continue;
assert_clause(clause);
return true;
propagated = true;
}
return false;
return propagated;
}
/**
* We assume the lemmas in the queue are necessary for completeness.
* So they all have to be enforced through case analysis.
* Lemmas with more than one unassigned literal are asserted here.
* The solver will case split on the unassigned literals to satisfy the lemma.
*/
bool theory_finite_set::instantiate_free_lemma() {
for (auto const& clause : m_lemmas) {
if (any_of(clause, [&](expr *e) { return ctx.find_assignment(e) == l_true; }))