3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2026-03-18 02:53:46 +00:00

first end-pass. Atomic

Signed-off-by: Nikolaj Bjorner <nbjorner@microsoft.com>
This commit is contained in:
Nikolaj Bjorner 2026-03-04 02:05:26 -08:00
parent 13f9fec339
commit 5aa3713d19
15 changed files with 6160 additions and 209 deletions

View file

@ -7,7 +7,8 @@ Module Name:
Abstract:
Implementation of nseq_model.
Implementation of nseq_model: model construction for the
Nielsen-based string solver.
Author:
@ -15,3 +16,292 @@ Author:
--*/
#include "smt/nseq_model.h"
#include "smt/theory_nseq.h"
#include "smt/nseq_regex.h"
#include "smt/nseq_state.h"
#include "smt/smt_context.h"
#include "smt/smt_model_generator.h"
#include "smt/proto_model/proto_model.h"
#include "ast/ast_pp.h"
namespace smt {
nseq_model::nseq_model(theory_nseq& th, ast_manager& m, seq_util& seq,
seq_rewriter& rw, euf::sgraph& sg, nseq_regex& regex)
: m_th(th), m(m), m_seq(seq), m_rewriter(rw), m_sg(sg), m_regex(regex), m_trail(m)
{}
void nseq_model::init(model_generator& mg, seq::nielsen_graph& nielsen, nseq_state const& state) {
m_var_values.reset();
m_var_regex.reset();
m_trail.reset();
m_factory = alloc(seq_factory, m, m_th.get_family_id(), mg.get_model());
mg.register_factory(m_factory);
register_existing_values(nielsen);
collect_var_regex_constraints(state);
// if the last solve returned sat, extract assignments from the
// satisfying leaf node found during DFS.
seq::nielsen_node const* root = nielsen.root();
if (root && root->is_satisfied())
extract_assignments(root);
}
model_value_proc* nseq_model::mk_value(enode* n, model_generator& mg) {
app* e = n->get_expr();
if (!m_seq.is_seq(e) && !m_seq.is_re(e) && !m_seq.str.is_nth_u(e))
return nullptr;
// For regex-sorted enodes, return the expression itself as a model value.
// Regexes are interpreted as themselves in the model.
if (m_seq.is_re(e)) {
m_trail.push_back(e);
return alloc(expr_wrapper_proc, e);
}
// For nth_u (underspecified nth), return a fresh value of the element sort.
if (m_seq.str.is_nth_u(e)) {
sort* srt = e->get_sort();
expr* val = m_factory->get_fresh_value(srt);
if (val) {
m_trail.push_back(val);
return alloc(expr_wrapper_proc, to_app(val));
}
return nullptr;
}
// look up snode for this expression
euf::snode* sn = m_sg.find(e);
expr_ref val(m);
if (sn)
val = snode_to_value(sn);
if (!val) {
// no assignment found — generate fresh value
val = m_factory->get_fresh_value(e->get_sort());
}
if (val) {
m_trail.push_back(val);
m_factory->add_trail(val);
return alloc(expr_wrapper_proc, to_app(val));
}
return alloc(expr_wrapper_proc, to_app(m_seq.str.mk_empty(e->get_sort())));
}
void nseq_model::finalize(model_generator& mg) {
m_var_values.reset();
m_var_regex.reset();
m_trail.reset();
m_factory = nullptr;
}
void nseq_model::extract_assignments(seq::nielsen_node const* node) {
if (!node)
return;
for (auto const& eq : node->str_eqs()) {
if (!eq.m_lhs || !eq.m_rhs)
continue;
if (eq.m_lhs->is_var() && !m_var_values.contains(eq.m_lhs->id())) {
expr_ref val = snode_to_value(eq.m_rhs);
if (val) {
m_trail.push_back(val);
m_var_values.insert(eq.m_lhs->id(), val);
}
}
if (eq.m_rhs->is_var() && !m_var_values.contains(eq.m_rhs->id())) {
expr_ref val = snode_to_value(eq.m_lhs);
if (val) {
m_trail.push_back(val);
m_var_values.insert(eq.m_rhs->id(), val);
}
}
}
}
expr_ref nseq_model::snode_to_value(euf::snode* n) {
if (!n)
return expr_ref(m);
if (n->is_empty())
return expr_ref(m_seq.str.mk_empty(m_seq.str.mk_string_sort()), m);
if (n->is_char() || n->is_unit()) {
expr* e = n->get_expr();
return e ? expr_ref(e, m) : expr_ref(m);
}
if (n->is_var())
return expr_ref(get_var_value(n), m);
if (n->is_concat()) {
expr_ref lhs = snode_to_value(n->arg(0));
expr_ref rhs = snode_to_value(n->arg(1));
if (lhs && rhs)
return expr_ref(m_seq.str.mk_concat(lhs, rhs), m);
if (lhs) return lhs;
if (rhs) return rhs;
return expr_ref(m);
}
// fallback: use the underlying expression
expr* e = n->get_expr();
return e ? expr_ref(e, m) : expr_ref(m);
}
expr_ref nseq_model::generate_regex_witness(euf::snode* regex, unsigned depth) {
if (!regex)
return expr_ref(m_seq.str.mk_empty(m_seq.str.mk_string_sort()), m);
// depth bound to prevent stack overflow on deep regexes
if (depth > 1000) {
sort* srt = m_seq.str.mk_string_sort();
expr* fresh = m_factory->get_fresh_value(srt);
return fresh ? expr_ref(fresh, m) : expr_ref(m_seq.str.mk_empty(srt), m);
}
// nullable regex: empty string is a valid witness
if (m_regex.is_nullable(regex))
return expr_ref(m_seq.str.mk_empty(m_seq.str.mk_string_sort()), m);
// collect first-position characters
euf::snode_vector chars;
m_regex.collect_first_chars(regex, chars);
if (!chars.empty()) {
// pick first concrete character, derive, and recurse
euf::snode* c = chars[0];
euf::snode* deriv = m_regex.derivative(regex, c);
expr_ref tail = generate_regex_witness(deriv, depth + 1);
if (tail && c->get_expr())
return expr_ref(m_seq.str.mk_concat(c->get_expr(), tail), m);
}
// fallback: return fresh value from factory (may not satisfy the regex,
// but avoids returning empty string which definitely doesn't satisfy non-nullable regex)
sort* srt = m_seq.str.mk_string_sort();
expr* fresh = m_factory->get_fresh_value(srt);
return fresh ? expr_ref(fresh, m) : expr_ref(m_seq.str.mk_empty(srt), m);
}
void nseq_model::register_existing_values(seq::nielsen_graph& nielsen) {
seq::nielsen_node const* root = nielsen.root();
if (!root)
return;
for (auto const& eq : root->str_eqs()) {
if (eq.m_lhs && eq.m_lhs->get_expr())
m_factory->register_value(eq.m_lhs->get_expr());
if (eq.m_rhs && eq.m_rhs->get_expr())
m_factory->register_value(eq.m_rhs->get_expr());
}
}
expr* nseq_model::get_var_value(euf::snode* var) {
expr* val = nullptr;
if (m_var_values.find(var->id(), val))
return val;
// unconstrained or regex-constrained: delegate to mk_fresh_value
val = mk_fresh_value(var);
if (val) {
m_trail.push_back(val);
m_var_values.insert(var->id(), val);
}
return val;
}
expr* nseq_model::mk_fresh_value(euf::snode* var) {
// check if this variable has regex constraints
euf::snode* re = nullptr;
if (m_var_regex.find(var->id(), re) && re) {
// generate a witness string satisfying the regex
expr_ref witness = generate_regex_witness(re);
if (witness) {
m_trail.push_back(witness);
m_factory->register_value(witness);
return witness;
}
}
// no regex constraint or witness generation failed: plain fresh value
sort* srt = m_seq.str.mk_string_sort();
if (var->get_expr())
srt = var->get_expr()->get_sort();
return m_factory->get_fresh_value(srt);
}
void nseq_model::collect_var_regex_constraints(nseq_state const& state) {
for (auto const& mem : state.str_mems()) {
if (!mem.m_str || !mem.m_regex)
continue;
// only collect for variable snodes (leaf variables needing assignment)
if (!mem.m_str->is_var())
continue;
unsigned id = mem.m_str->id();
euf::snode* existing = nullptr;
if (m_var_regex.find(id, existing) && existing) {
// intersect with existing constraint:
// build re.inter(existing, new_regex)
expr* e1 = existing->get_expr();
expr* e2 = mem.m_regex->get_expr();
if (e1 && e2) {
expr_ref inter(m_seq.re.mk_inter(e1, e2), m);
euf::snode* inter_sn = m_sg.mk(inter);
if (inter_sn)
m_var_regex.insert(id, inter_sn);
}
}
else {
m_var_regex.insert(id, mem.m_regex);
}
}
}
bool nseq_model::validate_regex(nseq_state const& state, ::proto_model& mdl) {
bool ok = true;
// validate positive memberships: str ∈ regex
for (auto const& mem : state.str_mems()) {
if (!mem.m_str || !mem.m_regex)
continue;
expr* s_expr = mem.m_str->get_expr();
expr* r_expr = mem.m_regex->get_expr();
if (!s_expr || !r_expr)
continue;
expr_ref in_re(m_seq.re.mk_in_re(s_expr, r_expr), m);
if (mdl.is_false(in_re)) {
IF_VERBOSE(0, verbose_stream() << "nseq model: positive membership violated: "
<< mk_bounded_pp(s_expr, m, 3)
<< " in " << mk_bounded_pp(r_expr, m, 3) << "\n";);
ok = false;
}
}
// validate negative memberships: str ∉ regex
for (auto const& entry : state.neg_mems()) {
if (!entry.m_str || !entry.m_regex)
continue;
expr* s_expr = entry.m_str->get_expr();
expr* r_expr = entry.m_regex->get_expr();
if (!s_expr || !r_expr)
continue;
expr_ref in_re(m_seq.re.mk_in_re(s_expr, r_expr), m);
expr_ref val(m);
mdl.eval(in_re, val, true);
if (val && m.is_true(val)) {
IF_VERBOSE(0, verbose_stream() << "nseq model: negative membership violated: "
<< mk_bounded_pp(s_expr, m, 3)
<< " not in " << mk_bounded_pp(r_expr, m, 3) << "\n";);
ok = false;
}
}
return ok;
}
}