3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2026-05-26 11:56:21 +00:00

Let's try to justify bounds

This commit is contained in:
CEisenhofer 2026-04-20 15:52:35 +02:00
parent 0bcdca787f
commit 41412293fe
5 changed files with 246 additions and 65 deletions

View file

@ -42,14 +42,16 @@ NSB review:
namespace seq {
void deps_to_lits(dep_tracker deps, svector<enode_pair> &eqs, svector<sat::literal> &lits) {
void deps_to_lits(dep_tracker deps, svector<enode_pair> &eqs, svector<sat::literal> &lits, vector<le, false>& les) {
vector<dep_source, false> vs;
dep_manager::s_linearize(deps, vs);
for (dep_source const &d : vs) {
if (std::holds_alternative<enode_pair>(d))
eqs.push_back(std::get<enode_pair>(d));
else
else if (std::holds_alternative<sat::literal>(d))
lits.push_back(std::get<sat::literal>(d));
else
les.push_back(std::get<le>(d));
}
}
@ -373,22 +375,28 @@ namespace seq {
add_constraint(constraint(m.mk_or(cases), dep, m));
}
bool nielsen_node::lower_bound(expr* e, rational& lo) const {
// TODO: Return dependencies
bool nielsen_node::lower_bound(expr* e, rational& lo, dep_tracker& dep) {
SASSERT(e);
return m_graph.m_solver.lower_bound(e, lo);
if (!m_graph.m_solver.lower_bound(e, lo))
return false;
expr_ref lo_expr(m_graph.a.mk_int(lo), m_graph.m);
m_graph.add_le_dependency(dep, this, lo_expr.get(), e);
return true;
}
bool nielsen_node::upper_bound(expr* e, rational& up) const {
// TODO: Return dependencies
bool nielsen_node::upper_bound(expr* e, rational& up, dep_tracker& dep) {
SASSERT(e);
rational v;
if (m_graph.a.is_numeral(e, v)) {
up = v;
return true;
}
return m_graph.m_solver.upper_bound(e, up);
if (!m_graph.m_solver.upper_bound(e, up))
return false;
expr_ref up_expr(m_graph.a.mk_int(up), m_graph.m);
m_graph.add_le_dependency(dep, this, e, up_expr.get());
return true;
}
// -----------------------------------------------
@ -503,6 +511,19 @@ namespace seq {
SASSERT(m_sat_node == nullptr);
}
void nielsen_graph::add_le_dependency(dep_tracker& dep, nielsen_node* n, expr* lhs, expr* rhs) {
SASSERT(lhs);
SASSERT(rhs);
expr_ref lhs_ref(lhs, m);
expr_ref rhs_ref(rhs, m);
// just assume it to be correct
dep_tracker d = m_dep_mgr.mk_leaf(le{ lhs_ref, rhs_ref });
// Just add the constraint - we do not have to recompute it
// [also it is on the set of side-conditions if we assert a satisfied node]
n->add_constraint(constraint(a.mk_le(lhs_ref, rhs_ref), d, m));
dep = m_dep_mgr.mk_join(dep, d);
}
// -----------------------------------------------------------------------
// nielsen_node: simplify_and_init
// -----------------------------------------------------------------------
@ -682,7 +703,8 @@ namespace seq {
// Simplify constant-exponent powers: base^0 → ε, base^1 → base.
// Returns new snode if any simplification happened, nullptr otherwise.
static euf::snode* simplify_const_powers(nielsen_node* node, euf::sgraph& sg, euf::snode* side) {
static euf::snode* simplify_const_powers(nielsen_node* node, euf::sgraph& sg, euf::snode* side, dep_tracker& dep) {
dep = nullptr;
SASSERT(side);
if (side->is_empty())
return nullptr;
@ -699,9 +721,11 @@ namespace seq {
if (tok->is_power()) {
expr* exp_e = get_power_exp_expr(tok, seq);
rational ub;
if (exp_e && node->upper_bound(exp_e, ub)) {
dep_tracker ub_dep = nullptr;
if (exp_e && node->upper_bound(exp_e, ub, ub_dep)) {
if (ub.is_zero()) {
// base^0 → ε (skip this token entirely)
dep = node->graph().dep_mgr().mk_join(dep, ub_dep);
simplified = true;
continue;
}
@ -709,6 +733,7 @@ namespace seq {
// base^1 → base
euf::snode* base_sn = tok->arg(0);
if (base_sn) {
dep = node->graph().dep_mgr().mk_join(dep, ub_dep);
result.push_back(base_sn);
simplified = true;
continue;
@ -915,10 +940,18 @@ namespace seq {
continue;
// 3a: simplify constant-exponent powers (base^0 → ε, base^1 → base)
if (euf::snode* s = simplify_const_powers(this, sg, eq.m_lhs))
{ eq.m_lhs = s; changed = true; }
if (euf::snode* s = simplify_const_powers(this, sg, eq.m_rhs))
{ eq.m_rhs = s; changed = true; }
dep_tracker lhs_pow_dep = nullptr;
if (euf::snode* s = simplify_const_powers(this, sg, eq.m_lhs, lhs_pow_dep)) {
eq.m_lhs = s;
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, lhs_pow_dep);
changed = true;
}
dep_tracker rhs_pow_dep = nullptr;
if (euf::snode* s = simplify_const_powers(this, sg, eq.m_rhs, rhs_pow_dep)) {
eq.m_rhs = s;
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, rhs_pow_dep);
changed = true;
}
// 3b: merge adjacent same-base tokens into combined powers
if (euf::snode* s = merge_adjacent_powers(sg, eq.m_lhs))
@ -962,18 +995,22 @@ namespace seq {
expr_ref norm_count = normalize_arith(m, count);
bool pow_le_count = false, count_le_pow = false;
dep_tracker pow_le_dep = nullptr, count_le_dep = nullptr;
rational diff;
if (get_const_power_diff(norm_count, pow_exp, m_graph.a, diff)) {
count_le_pow = diff.is_nonpos();
pow_le_count = diff.is_nonneg();
}
else if (!cur_path.empty()) {
pow_le_count = m_graph.check_lp_le(pow_exp, norm_count);
count_le_pow = m_graph.check_lp_le(norm_count, pow_exp);
pow_le_count = m_graph.check_lp_le(pow_exp, norm_count, this, pow_le_dep);
count_le_pow = m_graph.check_lp_le(norm_count, pow_exp, this, count_le_dep);
}
if (!pow_le_count && !count_le_pow)
continue;
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, pow_le_dep);
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, count_le_dep);
pow_side = dir_drop(sg, pow_side, 1, fwd);
other_side = dir_drop(sg, other_side, consumed, fwd);
expr* base_e = get_power_base_expr(end_tok, seq);
@ -1045,9 +1082,14 @@ namespace seq {
}
// 3e: LP-aware power directional elimination
else if (lp && rp && !cur_path.empty()) {
bool lp_le_rp = m_graph.check_lp_le(lp, rp);
bool rp_le_lp = m_graph.check_lp_le(rp, lp);
dep_tracker lp_le_dep = nullptr, rp_le_dep = nullptr;
bool lp_le_rp = m_graph.check_lp_le(lp, rp, this, lp_le_dep);
bool rp_le_lp = m_graph.check_lp_le(rp, lp, this, rp_le_dep);
if (lp_le_rp || rp_le_lp) {
if (lp_le_rp)
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, lp_le_dep);
if (rp_le_lp)
eq.m_dep = m_graph.dep_mgr().mk_join(eq.m_dep, rp_le_dep);
expr* smaller_exp = lp_le_rp ? lp : rp;
expr* larger_exp = lp_le_rp ? rp : lp;
eq.m_lhs = dir_drop(sg, eq.m_lhs, 1, fwd);
@ -1202,9 +1244,11 @@ namespace seq {
// contradict the variable's current integer bounds? If so, mark this
// node as a Parikh-image conflict immediately (avoids a solver call).
str_mem const* confl_cnstr;
if (!node.is_currently_conflict() && (confl_cnstr = m_parikh->check_parikh_conflict(node)) != nullptr) {
dep_tracker parikh_dep = nullptr;
if (!node.is_currently_conflict() &&
(confl_cnstr = m_parikh->check_parikh_conflict(node, parikh_dep)) != nullptr) {
node.set_general_conflict();
node.set_conflict(backtrack_reason::parikh_image, confl_cnstr->m_dep);
node.set_conflict(backtrack_reason::parikh_image, parikh_dep);
}
}
@ -3683,7 +3727,7 @@ namespace seq {
for (dep_source const& d : vs) {
if (std::holds_alternative<enode_pair>(d))
eqs.push_back(std::get<enode_pair>(d));
else
else if (std::holds_alternative<sat::literal>(d))
mem_literals.push_back(std::get<sat::literal>(d));
}
}
@ -4003,21 +4047,35 @@ namespace seq {
return dep;
}
bool nielsen_graph::check_lp_le(expr* lhs, expr* rhs) {
// TODO: We need the justification!!
rational lhs_lo, rhs_up;
if (m_solver.lower_bound(lhs, lhs_lo) &&
m_solver.upper_bound(rhs, rhs_up)) {
bool nielsen_graph::check_lp_le(expr* lhs, expr* rhs, nielsen_node* n, dep_tracker& dep) {
dep = nullptr;
rational lhs_lo, rhs_up;
bool has_lhs_lo = false, has_rhs_up = false;
dep_tracker lhs_lo_dep = nullptr, rhs_up_dep = nullptr;
if (n->lower_bound(lhs, lhs_lo, lhs_lo_dep))
has_lhs_lo = true;
if (has_lhs_lo && n->upper_bound(rhs, rhs_up, rhs_up_dep))
has_rhs_up = true;
if (has_lhs_lo && has_rhs_up) {
if (lhs_lo > rhs_up)
// NB: we only justify if we return true
return false; // definitely infeasible
}
rational rhs_lo, lhs_up;
if (m_solver.upper_bound(lhs, lhs_up) &&
m_solver.lower_bound(rhs, rhs_lo)) {
if (lhs_up <= rhs_lo)
rational rhs_lo, lhs_up;
bool has_rhs_lo = false, has_lhs_up = false;
dep_tracker rhs_lo_dep = nullptr, lhs_up_dep = nullptr;
if (n->upper_bound(lhs, lhs_up, lhs_up_dep))
has_lhs_up = true;
if (has_lhs_up && n->lower_bound(rhs, rhs_lo, rhs_lo_dep))
has_rhs_lo = true;
if (has_lhs_up && has_rhs_lo) {
if (lhs_up <= rhs_lo) {
dep = m_dep_mgr.mk_join(dep, lhs_up_dep);
dep = m_dep_mgr.mk_join(dep, rhs_lo_dep);
return true; // definitely feasible
}
}
// fall through - ask the solver [expensive]
@ -4033,7 +4091,11 @@ namespace seq {
m_solver.assert_expr(a.mk_ge(lhs, rhs_plus_one));
lbool result = m_solver.check();
m_solver.pop(1);
return result == l_false;
if (result == l_false) {
add_le_dependency(dep, n, lhs, rhs);
return true;
}
return false;
}
constraint nielsen_graph::mk_constraint(expr* fml, dep_tracker const& dep) {

View file

@ -338,7 +338,14 @@ namespace seq {
// index is the 0-based position in the input eq or mem list respectively.
using enode_pair = std::pair<smt::enode *, smt::enode *>;
using dep_source = std::variant<sat::literal, enode_pair>;
// arithmetic implication dependency: lhs <= rhs
struct le {
expr_ref lhs;
expr_ref rhs;
};
using dep_source = std::variant<sat::literal, enode_pair, le>;
// Arena-based dependency manager: builds an immutable tree of dep_source
@ -350,10 +357,12 @@ namespace seq {
// nullptr represents the empty dependency set.
using dep_tracker = dep_manager::dependency*;
// partition dep_source leaves from deps into enode pairs and sat literals.
// partition dep_source leaves from deps into enode pairs, sat literals,
// and arithmetic <= dependencies.
void deps_to_lits(dep_tracker deps,
svector<enode_pair>& eqs,
svector<sat::literal>& lits);
svector<sat::literal>& lits,
vector<le, false>& les);
// string equality constraint: lhs = rhs
// mirrors ZIPT's StrEq (both sides are regex-free snode trees)
@ -607,8 +616,8 @@ namespace seq {
// Query current bounds for a variable from the arithmetic subsolver.
// Falls der Subsolver keinen Bound liefert, werden konservative Defaults
// 0 / UINT_MAX verwendet.
bool lower_bound(expr* e, rational& lo) const;
bool upper_bound(expr* e, rational& up) const;
bool lower_bound(expr* e, rational& lo, dep_tracker& dep);
bool upper_bound(expr* e, rational& up, dep_tracker& dep);
// character constraint access (mirrors ZIPT's CharRanges)
u_map<std::pair<char_set, dep_tracker>> char_ranges() const { return m_char_ranges; }
@ -827,7 +836,6 @@ namespace seq {
// (e.g., explain_conflict) can call mk_join / linearize.
mutable dep_manager m_dep_mgr;
std::ostream &display(std::ostream &out, nielsen_node const* n) const;
public:
@ -972,6 +980,9 @@ namespace seq {
dep_manager& dep_mgr() { return m_dep_mgr; }
dep_manager const& dep_mgr() const { return m_dep_mgr; }
// Add a dependency leaf for lhs <= rhs and join it to dep.
void add_le_dependency(dep_tracker& dep, nielsen_node* n, expr* lhs, expr* rhs);
// Assert the constraints of `node` that are new relative to its
// parent (indices [m_parent_ic_count..end)) into the current solver scope.
// Called by search_dfs after simplify_and_init so that the newly derived
@ -1165,7 +1176,7 @@ namespace seq {
// mirrors ZIPT's NielsenNode.IsLe(): temporarily asserts NOT(lhs <= rhs)
// and returns true iff the result is unsatisfiable (i.e., lhs <= rhs is
// entailed). Path constraints are already in the solver incrementally.
bool check_lp_le(expr* lhs, expr* rhs);
bool check_lp_le(expr* lhs, expr* rhs, nielsen_node* n, dep_tracker& dep);
// create an integer constraint: lhs <kind> rhs
constraint mk_constraint(expr* fml, dep_tracker const& dep);

View file

@ -249,7 +249,8 @@ namespace seq {
//
// This check is lightweight — it uses only modular arithmetic on the already-
// known regex min/max lengths and the per-variable bounds stored in the node.
str_mem const* seq_parikh::check_parikh_conflict(nielsen_node& node) {
str_mem const* seq_parikh::check_parikh_conflict(nielsen_node& node, dep_tracker& dep) {
dep = nullptr;
for (str_mem const& mem : node.str_mems()) {
if (!mem.m_str || !mem.m_regex || !mem.m_str->is_var())
continue;
@ -269,9 +270,15 @@ namespace seq {
SASSERT(stride > 1);
rational lb_r, ub_r;
if (!node.lower_bound(mem.m_str->get_expr(), lb_r) || !node.upper_bound(mem.m_str->get_expr(), ub_r))
dep_tracker lb_dep = nullptr;
dep_tracker ub_dep = nullptr;
if (!node.lower_bound(mem.m_str->get_expr(), lb_r, lb_dep) ||
!node.upper_bound(mem.m_str->get_expr(), ub_r, ub_dep))
continue;
dep_tracker cur_dep = node.graph().dep_mgr().mk_join(mem.m_dep, lb_dep);
cur_dep = node.graph().dep_mgr().mk_join(cur_dep, ub_dep);
SASSERT(lb_r <= ub_r);
if (ub_r > INT_MAX)
continue;
@ -294,6 +301,7 @@ namespace seq {
// In that case k_min would be huge, and min_len + stride*k_min would
// also overflow ub → treat as a conflict immediately.
if (gap > UINT_MAX - (stride - 1)) {
dep = cur_dep;
return &mem; // ceiling division would overflow → k_min too large
}
k_min = (gap + stride - 1) / stride;
@ -303,12 +311,15 @@ namespace seq {
unsigned len_at_k_min;
if (k_min > (UINT_MAX - min_len) / stride) {
// Overflow: min_len + stride * k_min > UINT_MAX ≥ ub → conflict.
dep = cur_dep;
return &mem;
}
len_at_k_min = min_len + stride * k_min;
if (ub != UINT_MAX && len_at_k_min > ub)
if (ub != UINT_MAX && len_at_k_min > ub) {
dep = cur_dep;
return &mem; // no valid k exists → Parikh conflict
}
}
return nullptr;
}

View file

@ -121,7 +121,7 @@ namespace seq {
// This is a lightweight pre-check that avoids calling the integer
// subsolver. It is sound (never returns true for a satisfiable node)
// but incomplete (may miss conflicts that require the full solver).
str_mem const* check_parikh_conflict(nielsen_node& node);
str_mem const* check_parikh_conflict(nielsen_node& node, dep_tracker& dep);
// Compute the length stride of a regex expression.
// Exposed for testing and external callers.