3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2026-05-16 23:25:36 +00:00

euf_seq_plugin: fix identity elimination after merge, activate loop merging, integrate sgraph improvements (#9414)

* Initial plan

* Initial plan

* Fix identity elimination after merge and activate loop merging in euf_seq_plugin

Agent-Logs-Url: https://github.com/Z3Prover/z3/sessions/053b94e4-645a-4cde-ae5d-cf6d61222f92

Co-authored-by: NikolajBjorner <3085284+NikolajBjorner@users.noreply.github.com>

* Apply three ZIPT code review improvements to euf_seq_plugin

Agent-Logs-Url: https://github.com/Z3Prover/z3/sessions/da8647c4-ddff-47ce-9364-2eee3810c38d

Co-authored-by: NikolajBjorner <3085284+NikolajBjorner@users.noreply.github.com>

* Address code review: improve loop-merge defensive code and test variable names

Agent-Logs-Url: https://github.com/Z3Prover/z3/sessions/053b94e4-645a-4cde-ae5d-cf6d61222f92

Co-authored-by: NikolajBjorner <3085284+NikolajBjorner@users.noreply.github.com>

* Refactor: extract saturating_add helper, simplify hash-check condition

Agent-Logs-Url: https://github.com/Z3Prover/z3/sessions/da8647c4-ddff-47ce-9364-2eee3810c38d

Co-authored-by: NikolajBjorner <3085284+NikolajBjorner@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: NikolajBjorner <3085284+NikolajBjorner@users.noreply.github.com>
This commit is contained in:
Copilot 2026-04-29 11:12:00 -07:00 committed by GitHub
parent f461369ab8
commit 42582c6835
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 134 additions and 5 deletions

View file

@ -46,6 +46,11 @@ namespace euf {
}
}
// Saturating unsigned addition: returns UINT_MAX instead of wrapping.
static unsigned saturating_add(unsigned a, unsigned b) {
return (b > UINT_MAX - a) ? UINT_MAX : a + b;
}
unsigned enode_concat_hash::operator()(enode* n) const {
snode* sn = sg.find(n->get_expr());
if (sn && sn->has_cached_hash())
@ -64,6 +69,15 @@ namespace euf {
if (a == b) return true;
if (!is_any_concat(a, seq) || !is_any_concat(b, seq))
return false;
// fast-path: snode length check (O(1), avoids leaf allocation)
snode* sa = sg.find(a->get_expr());
snode* sb = sg.find(b->get_expr());
if (sa && sb && sa->length() != sb->length())
return false;
// fast-path: cached associativity hash (O(1))
bool both_have_hash = sa && sa->has_cached_hash() && sb && sb->has_cached_hash();
if (both_have_hash && sa->assoc_hash() != sb->assoc_hash())
return false;
enode_vector la, lb;
collect_enode_leaves(a, seq, la);
collect_enode_leaves(b, seq, lb);
@ -82,7 +96,7 @@ namespace euf {
m_sg(sg ? *sg : *alloc(sgraph, g.get_manager(), g, false)),
m_sg_owned(sg == nullptr),
m_concat_hash(m_seq, m_sg),
m_concat_eq(m_seq),
m_concat_eq(m_seq, m_sg),
m_concat_table(DEFAULT_HASHTABLE_INITIAL_CAPACITY, m_concat_hash, m_concat_eq) {
}
@ -168,6 +182,35 @@ namespace euf {
if (is_concat(n))
propagate_simplify(n);
}
// Re-apply identity and absorption rules over all tracked concat nodes.
// This handles the case where the merge caused a child to become equivalent
// to an identity (ε) or absorbing element (∅) that was not known at
// registration time (e.g. b ~ "" discovered after concat(x, b) was registered).
// Also re-simplifies RE concat nodes when a child's root has become full_seq,
// to handle nullable absorption through nested concats:
// concat(.*, concat(v, w)) = concat(.*, w) when v is nullable but w is not.
for (enode* n : m_concats) {
enode *na, *nb;
if (is_str_concat(n, na, nb)) {
if (is_str_empty(na))
push_merge(n, nb);
else if (is_str_empty(nb))
push_merge(n, na);
}
if (is_re_concat(n, na, nb)) {
if (is_re_epsilon(na))
push_merge(n, nb);
else if (is_re_epsilon(nb))
push_merge(n, na);
else if (is_re_empty(na))
push_merge(n, na); // absorb: concat(∅, b) = ∅
else if (is_re_empty(nb))
push_merge(n, nb); // absorb: concat(a, ∅) = ∅
else if (is_full_seq(na->get_root()) || is_full_seq(nb->get_root()))
propagate_simplify(n);
}
}
}
//
@ -202,6 +245,9 @@ namespace euf {
// 2. Nullable absorption: concat(u, .*, v, w) = concat(u, .*, w)
// when v is nullable and adjacent to full_seq (.*).
//
// 3. Loop merging: concat(body{lo1,hi1}, body{lo2,hi2}) = body{lo1+lo2, hi1+hi2}
// when both children are loops over congruent bodies.
//
void seq_plugin::propagate_simplify(enode* n) {
enode* a, *b;
if (!is_concat(n, a, b))
@ -236,6 +282,20 @@ namespace euf {
// concat(concat(u, v), .*) = concat(u, .*) when v nullable
// handled by associativity + nullable absorption on sub-concats
// Rule 3: Loop merging
// concat(v{l1,h1}, v{l2,h2}) = v{l1+l2,h1+h2}
unsigned lo1, hi1, lo2, hi2;
if (same_loop_body(a, b, lo1, hi1, lo2, hi2)) {
ast_manager& m = g.get_manager();
enode* body_n = a->get_arg(0);
// saturating add: prevent silent unsigned wrap-around on large bounds
unsigned lo_merged = saturating_add(lo1, lo2);
unsigned hi_merged = saturating_add(hi1, hi2);
expr_ref merged(m_seq.re.mk_loop(body_n->get_expr(), lo_merged, hi_merged), m);
enode* merged_n = mk(merged, 1, &body_n);
push_merge(n, merged_n);
}
}
bool seq_plugin::is_nullable(expr* e) {

View file

@ -63,7 +63,8 @@ namespace euf {
// Handles both str.++ (OP_SEQ_CONCAT) and re.++ (OP_RE_CONCAT).
struct enode_concat_eq {
seq_util const& seq;
enode_concat_eq(seq_util const& s) : seq(s) {}
sgraph& sg;
enode_concat_eq(seq_util const& s, sgraph& sg) : seq(s), sg(sg) {}
bool operator()(enode* a, enode* b) const;
};
@ -120,11 +121,14 @@ namespace euf {
bool is_loop(enode* n) const { return m_seq.re.is_loop(n->get_expr()); }
// string empty: ε for str.++
bool is_str_empty(enode* n) const { return m_seq.str.is_empty(n->get_expr()); }
// Follows the union-find root so that merges are correctly reflected.
bool is_str_empty(enode* n) const { return m_seq.str.is_empty(n->get_root()->get_expr()); }
// regex empty set: ∅ for re.++ (absorbing element)
bool is_re_empty(enode* n) const { return m_seq.re.is_empty(n->get_expr()); }
// Follows the union-find root so that merges are correctly reflected.
bool is_re_empty(enode* n) const { return m_seq.re.is_empty(n->get_root()->get_expr()); }
// regex epsilon: to_re("") for re.++ (identity element)
bool is_re_epsilon(enode* n) const { return m_seq.re.is_epsilon(n->get_expr()); }
// Follows the union-find root so that merges are correctly reflected.
bool is_re_epsilon(enode* n) const { return m_seq.re.is_epsilon(n->get_root()->get_expr()); }
bool is_to_re(enode* n) const { return m_seq.re.is_to_re(n->get_expr()); }
bool is_full_seq(enode* n) const { return m_seq.re.is_full_seq(n->get_expr()); }

View file

@ -227,6 +227,69 @@ static void test_sgraph_egraph_sync() {
SASSERT(nx->get_root() != ny->get_root());
}
// test seq_plugin: identity elimination fires after a post-registration merge
// When b ~ "" is learned AFTER concat(a, b) is registered, the plugin must
// re-check identity rules and conclude concat(a, b) ~ a.
static void test_seq_plugin_identity_after_merge() {
std::cout << "test_seq_plugin_identity_after_merge\n";
ast_manager m;
reg_decl_plugins(m);
euf::egraph eg(m);
euf::sgraph sg(m, eg);
euf::egraph& g = sg.get_egraph();
seq_util seq(m);
sort_ref str_sort(seq.str.mk_string_sort(), m);
expr_ref a(m.mk_const("a", str_sort), m);
expr_ref b(m.mk_const("b", str_sort), m);
expr_ref empty(seq.str.mk_empty(str_sort), m);
expr_ref ab(seq.str.mk_concat(a, b), m);
// Register concat(a, b) first — at this point b is not yet empty.
auto* nab = get_node(g, seq, ab);
auto* na = g.find(a);
auto* nb = g.find(b);
auto* nempty = get_node(g, seq, empty);
g.propagate();
// Now learn b ~ "" via a merge.
g.merge(nb, nempty, nullptr);
g.propagate();
// After propagation, concat(a, b) should be equivalent to a.
SASSERT(nab->get_root() == na->get_root());
std::cout << g << "\n";
}
// test seq_plugin: loop merging — concat(r{lo1,hi1}, r{lo2,hi2}) = r{lo1+lo2, hi1+hi2}
static void test_seq_plugin_loop_merge() {
std::cout << "test_seq_plugin_loop_merge\n";
ast_manager m;
reg_decl_plugins(m);
euf::egraph eg(m);
euf::sgraph sg(m, eg);
euf::egraph& g = sg.get_egraph();
seq_util seq(m);
sort_ref str_sort(seq.str.mk_string_sort(), m);
expr_ref x(m.mk_const("x", str_sort), m);
expr_ref r(seq.re.mk_to_re(x), m);
// r{2,3} and r{1,2} — bounds are chosen so the merged form r{3,5} is a distinct loop.
expr_ref first_loop(seq.re.mk_loop_proper(r, 2, 3), m);
expr_ref second_loop(seq.re.mk_loop_proper(r, 1, 2), m);
expr_ref expected_merged_loop(seq.re.mk_loop_proper(r, 3, 5), m);
// concat(r{2,3}, r{1,2})
expr_ref concat_loops(seq.re.mk_concat(first_loop, second_loop), m);
auto* nc = get_node(g, seq, concat_loops);
auto* nl35 = get_node(g, seq, expected_merged_loop);
g.propagate();
// After propagation, concat(r{2,3}, r{1,2}) should be equivalent to r{3,5}.
SASSERT(nc->get_root() == nl35->get_root());
std::cout << g << "\n";
}
void tst_euf_seq_plugin() {
s_var = 0; test_sgraph_basic();
s_var = 0; test_sgraph_backtrack();
@ -235,4 +298,6 @@ void tst_euf_seq_plugin() {
s_var = 0; test_seq_plugin_star_merge();
s_var = 0; test_seq_plugin_nullable_absorb();
s_var = 0; test_sgraph_egraph_sync();
s_var = 0; test_seq_plugin_identity_after_merge();
s_var = 0; test_seq_plugin_loop_merge();
}