spot/spot/twa/twagraph.cc
Antoine Martin 2ae9da1bc6 twagraph: merge_edges supports finite automata
* spot/twa/twagraph.cc: don't remove false-labeled edges if the
  automaton uses state-based acceptance and the edge is a self loop
2025-03-17 16:11:36 +01:00

1749 lines
57 KiB
C++

// -*- coding: utf-8 -*-
// Copyright (C) by the Spot authors, see the AUTHORS file for details.
//
// This file is part of Spot, a model checking library.
//
// Spot is free software; you can redistribute it and/or modify it
// under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 3 of the License, or
// (at your option) any later version.
//
// Spot is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
// or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
// License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
#include "config.h"
#include <spot/twa/twagraph.hh>
#include <spot/tl/print.hh>
#include <spot/misc/bddlt.hh>
#include <spot/twa/bddprint.hh>
#include <spot/kripke/kripkegraph.hh>
#include <spot/misc/escape.hh>
#include <spot/priv/robin_hood.hh>
#include <vector>
#include <deque>
using namespace std::string_literals;
namespace
{
using namespace spot;
// If LAST is false,
// it is guaranteed that there will be another src state
template<bool LAST>
void treat(std::vector<std::array<unsigned, 4>>& e_idx,
const twa_graph::graph_t::edge_vector_t& e_vec,
std::vector<unsigned>& e_chain,
std::vector<unsigned>& hash_of_state,
unsigned& idx,
unsigned s,
unsigned n_e)
{
assert(s < e_idx.size());
assert(idx < e_vec.size());
assert(e_chain.size() == e_vec.size());
//std::cout << s << "; " << idx << std::endl;
// Check if this state has outgoing transitions
if (s != e_vec[idx].src)
// Nothing to do
{
assert(!LAST);
return;
}
auto& s_idx = e_idx[s];
s_idx[0] = idx;
// helper
unsigned sub_idx[] = {-1u, -1u};
// All transitions of this state
while (true)
{
assert(idx < e_vec.size() + LAST);
if constexpr (!LAST)
{
if (e_vec[idx].src != s)
break;
}
else
{
if (idx == n_e)
break;
}
// Argh so many ifs
unsigned which = e_vec[idx].src == e_vec[idx].dst;
if (sub_idx[which] == -1u)
{
// First non-selflooping
sub_idx[which] = idx;
s_idx[1u+which] = idx;
}
else
{
// Continue the chained list
e_chain[sub_idx[which]] = idx;
sub_idx[which] = idx;
}
++idx;
}
s_idx[3] = idx;
// Check if self-loops appeared. We cannot hash
// states with self-loops.
if (s_idx[2] != -1u)
hash_of_state[s] = fnv<unsigned>::init;
}
}
namespace spot
{
std::string twa_graph::format_state(unsigned n) const
{
if (is_univ_dest(n))
{
std::stringstream ss;
bool notfirst = false;
for (unsigned d: univ_dests(n))
{
if (notfirst)
ss << '&';
notfirst = true;
ss << format_state(d);
}
return ss.str();
}
auto named = get_named_prop<std::vector<std::string>>("state-names");
if (named && n < named->size())
return (*named)[n];
auto prod = get_named_prop
<std::vector<std::pair<unsigned, unsigned>>>("product-states");
if (prod && n < prod->size())
{
auto& p = (*prod)[n];
std::stringstream ss;
ss << p.first << ',' << p.second;
return ss.str();
}
return std::to_string(n);
}
void
twa_graph::release_formula_namer(namer<formula>* namer,
bool keep_names)
{
if (keep_names)
{
auto v = new std::vector<std::string>(num_states());
auto& n = namer->names();
unsigned ns = n.size();
assert(n.size() <= v->size());
for (unsigned i = 0; i < ns; ++i)
{
auto f = n[i];
if (f)
(*v)[i] = str_psl(f);
}
set_named_prop("state-names", v);
}
delete namer;
}
/// \brief Merge universal destinations
///
/// If several states have the same universal destination, merge
/// them all. Also remove unused destination, and any redundant
/// state in each destination.
void twa_graph::merge_univ_dests()
{
auto& g = get_graph();
auto& dests = g.dests_vector();
auto& edges = g.edge_vector();
std::vector<unsigned> old_dests;
std::swap(dests, old_dests);
std::vector<unsigned> seen(old_dests.size(), -1U);
internal::univ_dest_mapper<twa_graph::graph_t> uniq(g);
auto fixup = [&](unsigned& in_dst)
{
unsigned dst = in_dst;
if ((int) dst >= 0) // not a universal edge
return;
dst = ~dst;
unsigned& nd = seen[dst];
if (nd == -1U)
nd = uniq.new_univ_dests(old_dests.data() + dst + 1,
old_dests.data() + dst + 1 + old_dests[dst]);
in_dst = nd;
};
unsigned tend = edges.size();
for (unsigned t = 1; t < tend; t++)
{
if (g.is_dead_edge(t))
continue;
fixup(edges[t].dst);
}
fixup(init_number_);
}
void twa_graph::merge_edges()
{
set_named_prop("highlight-edges", nullptr);
g_.remove_dead_edges_();
if (!is_existential())
merge_univ_dests();
auto& trans = this->edge_vector();
unsigned orig_size = trans.size();
unsigned tend = orig_size;
// When two transitions have the same (src,colors,dst),
// we can marge their conds.
auto merge_conds_and_remove_false = [&]()
{
typedef graph_t::edge_storage_t tr_t;
g_.sort_edges_([](const tr_t& lhs, const tr_t& rhs)
{
if (lhs.src < rhs.src)
return true;
if (lhs.src > rhs.src)
return false;
if (lhs.dst < rhs.dst)
return true;
if (lhs.dst > rhs.dst)
return false;
return lhs.acc < rhs.acc;
// Do not sort on conditions, we'll merge
// them.
});
bool is_state_acc = this->prop_state_acc().is_true();
unsigned out = 0;
unsigned in = 1;
// Skip any leading false edge.
while (in < tend
&& trans[in].cond == bddfalse
&& (!is_state_acc || trans[in].src != trans[in].dst))
++in;
if (in < tend)
{
++out;
if (out != in)
trans[out] = trans[in];
for (++in; in < tend; ++in)
{
if (trans[in].cond == bddfalse
&& (!is_state_acc
|| trans[in].src != trans[in].dst)) // Unusable edge
continue;
// Merge edges with the same source, destination, and
// colors. (We test the source last, because this is the
// most likely test to be true as edges are ordered by
// sources and then destinations.)
if (trans[out].dst == trans[in].dst
&& trans[out].acc == trans[in].acc
&& trans[out].src == trans[in].src)
{
trans[out].cond |= trans[in].cond;
}
else
{
++out;
if (in != out)
trans[out] = trans[in];
}
}
}
if (++out != tend)
{
tend = out;
trans.resize(tend);
}
};
// When two transitions have the same (src,cond,dst), we can marge
// their colors. This only works for Fin-less acceptance.
//
// FIXME: We could should also merge edges when using
// fin_acceptance, but the rule for Fin sets are different than
// those for Inf sets, (and we need to be careful if a set is used
// both as Inf and Fin)
auto merge_colors = [&]()
{
if (tend < 2 || acc().uses_fin_acceptance())
return;
typedef graph_t::edge_storage_t tr_t;
g_.sort_edges_([](const tr_t& lhs, const tr_t& rhs)
{
if (lhs.src < rhs.src)
return true;
if (lhs.src > rhs.src)
return false;
if (lhs.dst < rhs.dst)
return true;
if (lhs.dst > rhs.dst)
return false;
bdd_less_than_stable lt;
return lt(lhs.cond, rhs.cond);
// Do not sort on acceptance, we'll merge them.
});
// Start on the second position and try to merge it into the
// first one
unsigned out = 1;
unsigned in = 2;
for (; in < tend; ++in)
{
// Merge edges with the same source, destination,
// and conditions. (We test the source last, for the
// same reason as above.)
if (trans[out].dst == trans[in].dst
&& trans[out].cond.id() == trans[in].cond.id()
&& trans[out].src == trans[in].src)
{
trans[out].acc |= trans[in].acc;
}
else
{
++out;
if (in != out)
trans[out] = trans[in];
}
}
if (++out != tend)
{
tend = out;
trans.resize(tend);
}
};
// These next two lines used to be done in the opposite order in
// 2.9.x and before.
//
// However merging colors first is more likely to
// make parallel non-deterministic transition disappear.
//
// Consider:
// (1)--a-{1}--->(2)
// (1)--a-{2}--->(2)
// (1)--!a-{1}-->(2)
// If colors are merged first, the first two transitions become one,
// and the result is more deterministic:
// (1)--a-{1,2}->(2)
// (1)--!a-{1}-->(2)
// If conditions were merged first, we would get the following
// non-deterministic transitions instead:
// (1)--1-{1}--->(2)
// (1)--a-{2}--->(2)
merge_colors();
merge_conds_and_remove_false();
// Merging some edges may turn a non-deterministic automaton
// into a deterministic one.
if (prop_universal().is_false() && tend != orig_size)
prop_universal(trival::maybe());
g_.chain_edges_();
}
unsigned twa_graph::merge_states(parallel_policy ppolicy)
{
if (!is_existential())
throw std::runtime_error(
"twa_graph::merge_states() does not work on alternating automata");
const unsigned n_states = num_states();
const auto& e_vec = edge_vector();
unsigned n_edges = e_vec.size();
if (n_edges <= 1)
{
if (n_states <= 1)
return 0;
// We don't have a very convenient way to resize the state
// vector.
std::vector<unsigned> remap(n_states, -1U);
SPOT_ASSUME(remap.data() != nullptr); // help GCC 9.4
remap[0] = 0;
get_graph().defrag_states(remap, 1);
SPOT_ASSERT(num_states() == 1);
set_init_state(0);
return n_states - 1;
}
#ifdef ENABLE_PTHREAD
const unsigned nthreads = ppolicy.nthreads();
#else
(void) ppolicy;
constexpr unsigned nthreads = 1;
#endif
typedef graph_t::edge_storage_t tr_t;
g_.sort_edges_srcfirst_([](const tr_t& lhs, const tr_t& rhs)
{
assert(lhs.src == rhs.src);
if (lhs.acc < rhs.acc)
return true;
if (lhs.acc > rhs.acc)
return false;
// compare with id?
if (bdd_less_than_stable lt; lt(lhs.cond, rhs.cond))
return true;
if (rhs.cond != lhs.cond)
return false;
return lhs.dst < rhs.dst;
}, nthreads);
g_.chain_edges_();
// Edges are nicely chained and there are no erased edges
// -> We can work with the edge_vector
// Check if it is a game <-> "state-player" is defined. If
// so, we can only merge states that belong to the same player.
// (We will use two hash maps in this case.)
auto sp = get_named_prop<std::vector<bool>>("state-player");
// The hashing is a bit delicate: We may only use the dst if it has
// no self-loop. HASH_OF_STATE stores the hash associated to each
// state (by default its own number) or some common value if the
// state contains self-loop.
std::vector<unsigned> hash_of_state;
hash_of_state.reserve(n_states);
for (unsigned i = 0; i < n_states; ++i)
hash_of_state.push_back(i);
// For each state we need 4 indices of the edge vector
// [first, first_non_sfirst_selflooplfloop, first_selfloop, end]
// The init value makes sure nothing is done for dead end states
std::vector<std::array<unsigned, 4>> e_idx(n_states, {-1u, -1u,
-1u, -1u});
// Like a linked list holding the non-selfloop and selfloop transitions
std::vector<unsigned> e_chain(n_edges, -1u);
unsigned idx = 1;
// Edges are sorted with respect to src first
const unsigned n_high = e_vec.back().src;
for (auto s = 0u; s < n_high; ++s)
treat<false>(e_idx, e_vec, e_chain,
hash_of_state, idx, s, n_edges);
// Last one
treat<true>(e_idx, e_vec, e_chain,
hash_of_state, idx, n_high, n_edges);
assert(idx == e_vec.size() && "Something went wrong during indexing");
unsigned n_player1 = 0u;
if (sp)
n_player1 = std::accumulate(sp->begin(), sp->end(), 0u);
// Represents which states share a hash
// Head is in the unordered_map,
// hash_linked_list is like a linked list structure
// of fake pointers
std::vector<unsigned> hash_linked_list(n_states, -1u);
typedef robin_hood::unordered_flat_map<size_t,
std::pair<unsigned,
unsigned>> player_map;
// If the automaton is not a game, everything is assumed to be
// owned by player 0.
player_map map0; // for player 0
player_map map1; // for player 1
map0.reserve(n_states - n_player1);
map1.reserve(n_player1);
// Sadly we need to loop the edges twice since we have
// to check for self-loops before hashing
auto emplace = [&hash_linked_list](auto& m, auto h, auto s)
{
auto [it, inserted] = m.try_emplace(h, std::make_pair(s, s));
if (!inserted)
{
// We already have an entry with hash "h". Link it
// to the new state.
unsigned idx = it->second.second; // tail of the list
assert(idx < s && "Must be monotone");
hash_linked_list[idx] = s;
it->second.second = s;
}
};
// Hash all states
constexpr unsigned shift = sizeof(size_t)/2 * CHAR_BIT;
for (auto s = 0u; s != n_states; ++s)
{
size_t h = fnv<size_t>::init;
const unsigned e = e_idx[s][3];
for (unsigned i = e_idx[s][0]; i != e; ++i)
{
// If size_t has 8byte and unsigned has 4byte
// then this works fine, otherwise there might be more collisions
size_t hh = hash_of_state[e_vec[i].dst];
hh <<= shift;
hh += e_vec[i].cond.id();
h ^= hh;
h *= fnv<size_t>::prime;
h ^= e_vec[i].acc.hash();
h *= fnv<size_t>::prime;
}
if (sp && (*sp)[s])
emplace(map1, h, s);
else
emplace(map0, h, s);
}
// All states that might possible be merged share the same hash
// Info hash coll
//std::cout << "Hash collision rate pre merge: "
// << ((map0.size()+map1.size())/((float)n_states))
// << '\n';
// Check whether we can merge two states
// and takes into account the self-loops
auto state_equal = [&e_vec, &e_chain, &e_idx](unsigned s1, unsigned s2,
std::vector<char>& checked1,
std::vector<char>& checked2)
{
auto edge_data_comp = [](const auto& lhs,
const auto& rhs)
{
if (lhs.acc < rhs.acc)
return true;
if (lhs.acc > rhs.acc)
return false;
// todo compare with id
if (bdd_less_than_stable lt; lt(lhs.cond, rhs.cond))
return true;
return false;
};
auto [i1, nsl1, sl1, e1] = e_idx[s1];
(void) nsl1;
auto [i2, nsl2, sl2, e2] = e_idx[s2];
(void) nsl2;
unsigned n_trans = e1 - i1;
if ((e2 - i2) != n_trans)
return false; // Different number of outgoing trans
// checked1/2 is one element larger than necessary;
// the last element (false) serves as a sentinel.
checked1.clear();
checked1.resize(n_trans + 1, false);
checked2.clear();
checked2.resize(n_trans + 1, false);
// Try to match self-loops
unsigned self_loops_matched = 0;
while ((sl1 < e1) && (sl2 < e2))
{
auto& data1 = e_vec[sl1].data();
auto& data2 = e_vec[sl2].data();
if (data1 == data2)
{
// Matched
++self_loops_matched;
checked1[sl1 - i1] = true; //never touches last element
checked2[sl2 - i2] = true;
// Advance both
sl1 = e_chain[sl1];
sl2 = e_chain[sl2];
}
// Since edges are ordered on each side, advance
// the smallest side in case there is no match.
else if (edge_data_comp(data1, data2))
sl1 = e_chain[sl1];
else
sl2 = e_chain[sl2];
}
// If the matched self-loops cover all transitions, we can
// stop here.
if (self_loops_matched == n_trans)
return true;
// The remaining edges need to match exactly
unsigned idx1 = i1;
unsigned idx2 = i2;
while (((idx1 < e1) && (idx2 < e2)))
{
// More efficient version?
// Skip checked edges
// Last element serves as break
while (checked1[idx1 - i1])
++idx1;
while (checked2[idx2 - i2])
++idx2;
// If one is out of bounds, so is the other
if (idx1 == e1)
{
assert(idx2 == e2);
break;
}
if ((e_vec[idx1].dst != e_vec[idx2].dst)
|| !(e_vec[idx1].data() == e_vec[idx2].data()))
return false;
// Advance
++idx1;
++idx2;
}
// All edges have bee paired
return true;
};
const unsigned nb_states = num_states();
std::vector<unsigned> remap(nb_states, -1U);
// Check all pair of states with compatible hash
auto check_ix = [&](unsigned ix, std::vector<unsigned>& v,
std::vector<char>& checked1,
std::vector<char>& checked2)
{
if (hash_linked_list[ix] == -1U) // no compatible state
return;
v.clear();
for (unsigned i = ix; i != -1U; i = hash_linked_list[i])
v.push_back(i);
const unsigned vs = v.size();
for (unsigned idx = 0; idx < vs; ++idx)
{
unsigned i = v[idx];
for (unsigned jdx = 0; jdx < idx; ++jdx)
{
unsigned j = v[jdx];
if (state_equal(j, i, checked1, checked2))
{
remap[i] = (remap[j] != -1U) ? remap[j] : j;
// Because of the special self-loop tests we use
// above, it's possible that i can be mapped to
// remap[j] even if j was the last compatible
// state found. Consider the following cases,
// taken from an actual test case: 18 is equal to
// 5, 35 is equal to 18, but 35 is not equal to 5.
//
// State: 5
// [0&1&2] 8 {3}
// [!0&1&2] 10 {1}
// [!0&!1&!2] 18 {1}
// [!0&!1&2] 19 {1}
// [!0&1&!2] 20 {1}
//
// State: 18
// [0&1&2] 8 {3}
// [!0&1&2] 10 {1}
// [!0&!1&!2] 18 {1} // self-loop
// [!0&!1&2] 19 {1}
// [!0&1&!2] 20 {1}
//
// State: 35
// [0&1&2] 8 {3}
// [!0&1&2] 10 {1}
// [!0&!1&!2] 35 {1} // self-loop
// [!0&!1&2] 19 {1}
// [!0&1&!2] 20 {1}
break;
}
}
}
};
auto upd = [](auto& b, const auto&e, unsigned it)
{
while ((it > 0) & (b != e))
{
--it;
++b;
}
};
auto worker = [&upd, check_ix, nthreads](unsigned pid,
auto beg1, auto end1,
auto beg0, auto end0)
{
// Temporary storage for list of edges to reduce cache misses
std::vector<unsigned> v;
// Vector reused by all invocations of state_equal to mark edges
// that have been matched already.
std::vector<char> checked1;
std::vector<char> checked2;
upd(beg1, end1, pid);
upd(beg0, end0, pid);
for (; beg1 != end1; upd(beg1, end1, nthreads))
check_ix(beg1->second.first, v, checked1, checked2);
for (; beg0 != end0; upd(beg0, end0, nthreads))
check_ix(beg0->second.first, v, checked1, checked2);
};
{
auto beg1 = map1.begin();
auto end1 = map1.end();
auto beg0 = map0.begin();
auto end0 = map0.end();
#ifndef ENABLE_PTHREAD
(void) nthreads;
#else
if (nthreads <= 1)
{
#endif // ENABLE_PTHREAD
worker(0, beg1, end1, beg0, end0);
#ifdef ENABLE_PTHREAD
}
else
{
static auto tv = std::vector<std::thread>();
assert(tv.empty());
tv.resize(nthreads);
for (unsigned pid = 0; pid < nthreads; ++pid)
tv[pid] = std::thread(
[worker, pid, beg1, end1, beg0, end0]()
{
worker(pid, beg1, end1, beg0, end0);
return;
});
for (auto& t : tv)
t.join();
tv.clear();
}
#endif // ENABLE_PTHREAD
}
for (auto& e: edges())
if (remap[e.dst] != -1U)
{
assert((!sp || (sp->at(e.dst) == sp->at(remap[e.dst])))
&& "States do not have the same owner");
e.dst = remap[e.dst];
}
if (remap[get_init_state_number()] != -1U)
set_init_state(remap[get_init_state_number()]);
unsigned st = 0;
for (auto& s: remap)
if (s == -1U)
s = st++;
else
s = -1U;
unsigned merged = num_states() - st;
if (merged)
defrag_states(remap, st);
// Info hash coll 2
//std::cout << "Hash collision rate post merge: "
// << ((map0.size()+map1.size())/((float)num_states()))
// << '\n';
return merged;
}
unsigned twa_graph::merge_states_of(bool stable,
const std::vector<bool>* to_merge_ptr)
{
if (!is_existential())
throw std::runtime_error(
"twa_graph::merge_states() does not work on alternating automata");
typedef graph_t::edge_storage_t tr_t;
if (stable)
g_.sort_edges_of_<true>([](const tr_t& lhs, const tr_t& rhs)
{
if (lhs.acc < rhs.acc)
return true;
if (lhs.acc > rhs.acc)
return false;
if (bdd_less_than_stable lt; lt(lhs.cond, rhs.cond))
return true;
if (rhs.cond != lhs.cond)
return false;
// The destination must be sorted last
// for our self-loop optimization to work.
return lhs.dst < rhs.dst;
});
else
g_.sort_edges_of_<false>([](const tr_t& lhs, const tr_t& rhs)
{
if (lhs.acc < rhs.acc)
return true;
if (lhs.acc > rhs.acc)
return false;
if (bdd_less_than lt; lt(lhs.cond, rhs.cond))
return true;
if (rhs.cond != lhs.cond)
return false;
// The destination must be sorted last
// for our self-loop optimization to work.
return lhs.dst < rhs.dst;
});
// Associates a hash value to a vector of classes
std::unordered_map<size_t, std::vector<std::set<unsigned>>> equiv_class_;
auto hash_state_ = [&](unsigned s)->size_t
{
// Hash the edges
// bottle_neck?
size_t h = fnv<size_t>::init;
for (const edge_storage_t& e : out(s))
{
h ^= knuth32_hash(e.dst);
h ^= knuth32_hash(e.cond.id());
h ^= e.acc.hash();
h = wang32_hash(h);
}
return h;
};
const unsigned nb_states = num_states();
std::vector<unsigned> comp_classes_; // Classes to be merged
for (unsigned i = 0; i != nb_states; ++i)
{
if (to_merge_ptr && !(*to_merge_ptr)[nb_states])
continue;
size_t hi = hash_state_(i);
auto equal_to_i_ = [&, outi = out(i)](unsigned j)
{
auto outj = out(j);
return std::equal(outi.begin(), outi.end(),
outj.begin(), outj.end(),
[](const edge_storage_t& a,
const edge_storage_t& b)
{ return ((a.dst == b.dst
|| (a.dst == a.src && b.dst == b.src))
&& a.data() == b.data()); });
};
comp_classes_.clear();
// get all compatible classes
// Candidate classes share a hash
// A state is compatible to a class if it is compatible
// to any of its states
auto& cand_classes = equiv_class_[hi];
unsigned n_c_classes = cand_classes.size();
// Built it in "reverse order" the idx of
// classes to be merged
for (unsigned nc = n_c_classes - 1; nc < n_c_classes; --nc)
if (std::any_of(cand_classes[nc].begin(),
cand_classes[nc].end(),
[&](unsigned j)
{return equal_to_i_(j); }))
comp_classes_.push_back(nc);
// Possible results:
// 1) comp_classes_ is empty -> i gets its own class
// 2) fuse together all comp_classes and add i
if (comp_classes_.empty())
cand_classes.emplace_back(std::set<unsigned>({i}));
else
{
// Lowest idx
auto& base_class = cand_classes[comp_classes_.back()];
comp_classes_.pop_back(); // Keep this one
for (unsigned compi : comp_classes_)
{
// fuse with base and delete
base_class.insert(cand_classes[compi].begin(),
cand_classes[compi].end());
std::swap(cand_classes[compi], cand_classes.back());
cand_classes.pop_back();
}
// finally add the current state that caused all the merging
base_class.emplace_hint(base_class.end(), i);
}
};
// Now we have equivalence classes
// and a state can only be in exactly one.
// (Otherwise the classes would have fused)
// For each equiv class we take the first state as representative
// and redirect all incoming edges to this one.
std::vector<unsigned> remap(nb_states, -1U);
for (const auto& [_, class_v] : equiv_class_)
for (const auto& aclass : class_v)
{
(void)_; // please some versions of GCC
unsigned rep = *aclass.begin();
for (auto it = ++aclass.begin(); it != aclass.end(); ++it)
remap[*it] = rep;
}
for (auto& e: edges())
if (remap[e.dst] != -1U)
e.dst = remap[e.dst];
if (remap[get_init_state_number()] != -1U)
set_init_state(remap[get_init_state_number()]);
unsigned st = 0;
for (auto& s: remap)
if (s == -1U)
s = st++;
else
s = -1U;
defrag_states(remap, st);
return remap.size() - st;
}
void twa_graph::purge_unreachable_states(shift_action* f, void* action_data)
{
unsigned num_states = g_.num_states();
// The TODO vector serves two purposes:
// - it is a stack of state to process,
// - it is a set of processed states.
// The lower 31 bits of each entry is a state on the stack. (The
// next usable entry on the stack is indicated by todo_pos.) The
// 32th bit (i.e., the sign bit) of todo[x] indicates whether
// states number x has been seen.
std::vector<unsigned> todo(num_states, 0);
const unsigned seen = 1U << (sizeof(unsigned)*8-1);
const unsigned mask = seen - 1;
unsigned todo_pos = 0;
for (unsigned i: univ_dests(get_init_state_number()))
{
todo[i] |= seen;
todo[todo_pos++] |= i;
}
do
{
unsigned cur = todo[--todo_pos] & mask;
todo[todo_pos] ^= cur; // Zero the state
for (auto& t: g_.out(cur))
for (unsigned dst: univ_dests(t.dst))
if (!(todo[dst] & seen))
{
todo[dst] |= seen;
todo[todo_pos++] |= dst;
}
}
while (todo_pos > 0);
// Now renumber each used state.
unsigned current = 0;
for (auto& v: todo)
if (!(v & seen))
v = -1U;
else
v = current++;
if (current == todo.size())
return; // No unreachable state.
// Removing some non-deterministic dead state could make the
// automaton universal.
if (prop_universal().is_false())
prop_universal(trival::maybe());
if (prop_complete().is_false())
prop_complete(trival::maybe());
if (f)
(*f)(todo, action_data);
defrag_states(todo, current);
}
void twa_graph::purge_dead_states()
{
unsigned num_states = g_.num_states();
std::vector<unsigned> useful(num_states, 0);
// Make a DFS to compute a topological order.
std::vector<unsigned> order;
order.reserve(num_states);
bool purge_unreachable_needed = false;
if (is_existential())
{
std::vector<std::pair<unsigned, unsigned>> todo; // state, edge
useful[get_init_state_number()] = 1;
todo.emplace_back(init_number_, g_.state_storage(init_number_).succ);
do
{
unsigned src;
unsigned tid;
std::tie(src, tid) = todo.back();
if (tid == 0U)
{
todo.pop_back();
order.emplace_back(src);
continue;
}
auto& t = g_.edge_storage(tid);
todo.back().second = t.next_succ;
unsigned dst = t.dst;
if (useful[dst] != 1 && t.cond != bddfalse)
{
todo.emplace_back(dst, g_.state_storage(dst).succ);
useful[dst] = 1;
}
}
while (!todo.empty());
}
else
{
// state, edge, begin, end
std::vector<std::tuple<unsigned, unsigned,
const unsigned*, const unsigned*>> todo;
auto& dests = g_.dests_vector();
auto beginend = [&](const unsigned& dst,
const unsigned*& begin, const unsigned*& end)
{
if ((int)dst < 0)
{
begin = dests.data() + ~dst + 1;
end = begin + dests[~dst];
}
else
{
begin = &dst;
end = begin + 1;
}
};
{
const unsigned* begin;
const unsigned* end;
beginend(init_number_, begin, end);
todo.emplace_back(init_number_, 0U, begin, end);
}
for (;;)
{
unsigned& tid = std::get<1>(todo.back());
const unsigned*& begin = std::get<2>(todo.back());
const unsigned*& end = std::get<3>(todo.back());
if (tid == 0U && begin == end)
{
unsigned src = std::get<0>(todo.back());
todo.pop_back();
// Last transition from a state?
if ((int)src >= 0 && (todo.empty()
|| src != std::get<0>(todo.back())))
order.emplace_back(src);
if (todo.empty())
break;
else
continue;
}
unsigned dst = *begin++;
if (begin == end)
// that was the last destination, update the stack for
// the next edge.
{
do
tid = g_.edge_storage(tid).next_succ;
while (tid && g_.edge_storage(tid).cond == bddfalse);
if (tid != 0)
beginend(g_.edge_storage(tid).dst, begin, end);
}
if (useful[dst] != 1)
{
auto& ss = g_.state_storage(dst);
unsigned succ = ss.succ;
while (succ && g_.edge_storage(succ).cond == bddfalse)
succ = g_.edge_storage(succ).next_succ;
if (succ == 0U)
continue;
useful[dst] = 1;
const unsigned* begin;
const unsigned* end;
beginend(g_.edge_storage(succ).dst, begin, end);
todo.emplace_back(dst, succ, begin, end);
}
}
}
// At this point, all reachable states with successors are marked
// as useful.
for (;;)
{
bool univ_edge_erased = false;
// Process states in topological order to mark those without
// successors as useless.
for (auto s: order)
{
auto t = g_.out_iteraser(s);
bool useless = true;
while (t)
{
// Erase any false edge, except self-loops (which can
// be used to store colors on state without successor
// with state-based acceptance).
if (t->cond == bddfalse && t->src != t->dst)
{
t.erase();
continue;
}
// A non-false edge is useful if all its destinations
// are useful.
bool usefuledge = true;
for (unsigned d: univ_dests(t->dst))
if (!useful[d])
{
usefuledge = false;
break;
}
// Erase any useless edge
if (!usefuledge)
{
if (is_univ_dest(t->dst))
univ_edge_erased = true;
t.erase();
continue;
}
// if we have a edge to a useful state, then the
// state is useful.
useless = false;
++t;
}
if (useless)
useful[s] = 0;
}
// If we have erased any universal destination, it is possible
// that we have have created some new dead states, so we
// actually need to redo the whole thing again until there is
// no more universal edge to remove. Also we might have
// created some unreachable states, so we will simply call
// purge_unreachable_states() later to clean this.
if (!univ_edge_erased)
break;
else
purge_unreachable_needed = true;
}
// Is the initial state actually useful? If not, make this an
// empty automaton by resetting the graph.
bool usefulinit = true;
for (unsigned d: univ_dests(init_number_))
if (!useful[d])
{
usefulinit = false;
break;
}
if (!usefulinit)
{
g_ = graph_t();
init_number_ = new_state();
prop_universal(true);
prop_complete(false);
prop_stutter_invariant(true);
prop_weak(true);
return;
}
// Renumber each used state.
unsigned current = 0;
for (unsigned s = 0; s < num_states; ++s)
if (useful[s])
useful[s] = current++;
else
useful[s] = -1U;
if (current == num_states)
return; // No useless state.
// Removing some non-deterministic dead state could make the
// automaton universal. Likewise for non-complete.
if (prop_universal().is_false())
prop_universal(trival::maybe());
if (prop_complete().is_false())
prop_complete(trival::maybe());
defrag_states(useful, current);
if (purge_unreachable_needed)
purge_unreachable_states();
}
void twa_graph::defrag_states(std::vector<unsigned>& newst,
unsigned used_states)
{
if (!is_existential())
{
// Running defrag_states() on alternating automata is tricky,
// because we want to
// #1 rename the regular states
// #2 rename the states in universal destinations
// #3 get rid of the unused universal destinations
// #4 merge identical universal destinations
//
// graph::degrag_states() actually does only #1. It could
// do #2, but that would prevent us from doing #3 and #4. It
// cannot do #3 and #4 because the graph object does not know
// what an initial state is, and our initial state might be
// universal.
//
// As a consequence this code preforms #2, #3, and #4 before
// calling graph::degrag_states() to finish with #1. We clear
// the "dests vector" of the current automaton, recreate all
// the new destination groups using a univ_dest_mapper to
// simplify and unify them, and extend newst with some new
// entries that will point the those new universal destination
// so that graph::defrag_states() does not have to deal with
// universal destination in any way.
auto& g = get_graph();
auto& dests = g.dests_vector();
// Clear the destination vector, saving the old one.
std::vector<unsigned> old_dests;
std::swap(dests, old_dests);
// dests will be updated as a side effect of declaring new
// destination groups to uniq.
internal::univ_dest_mapper<twa_graph::graph_t> uniq(g);
// The newst entry associated to each of the old destination
// group.
std::vector<unsigned> seen(old_dests.size(), -1U);
// Rename a state if it denotes a universal destination. This
// function has to be applied to the destination of each edge,
// as well as to the initial state. The need to work on the
// initial state is the reason it cannot be implemented in
// graph::defrag_states().
auto fixup = [&](unsigned& in_dst)
{
unsigned dst = in_dst;
if ((int) dst >= 0) // not a universal edge
return;
dst = ~dst;
unsigned& nd = seen[dst];
if (nd == -1U) // An unprocessed destination group
{
// store all renamed destination states in tmp
std::vector<unsigned> tmp;
auto begin = old_dests.data() + dst + 1;
auto end = begin + old_dests[dst];
while (begin != end)
{
unsigned n = newst[*begin++];
if (n == -1U)
continue;
tmp.emplace_back(n);
}
if (tmp.empty())
{
// All destinations of this group were marked for
// removal. Mark this universal transition for
// removal as well. Is this really what we expect?
nd = -1U;
}
else
{
// register this new destination group, add it to
// newst, and use the index in newst to relabel
// the state so that graph::degrag_states() will
// eventually update it to the correct value.
nd = newst.size();
newst.emplace_back(uniq.new_univ_dests(std::move(tmp)));
}
}
in_dst = nd;
};
fixup(init_number_);
for (auto& e: edges())
fixup(e.dst);
}
// Update properties...
if (auto* names = get_named_prop<std::vector<std::string>>("state-names"))
{
permute_vector(*names, newst);
names->resize(used_states);
}
if (auto hs = get_named_prop<std::map<unsigned, unsigned>>
("highlight-states"))
{
unsigned ns = newst.size();
std::map<unsigned, unsigned> hs2;
for (auto p: *hs)
{
// Let's just ignore unexisting states. Raising an
// exception here would leave the automaton in a strange
// state.
if (SPOT_UNLIKELY(p.first >= ns))
continue;
unsigned dst = newst[p.first];
if (dst != -1U)
hs2[dst] = p.second;
}
std::swap(*hs, hs2);
}
if (auto he = get_named_prop<std::map<unsigned, unsigned>>
("highlight-edges"))
{
// Unfortunately, the underlying graph, who might remove some
// edges, knows nothing about named properties. So we have to
// predict the indices of the edges after
// graph::defrag_states() will run. This might break if
// graph::defrag_states() is changed.
auto& ev = edge_vector();
unsigned es = ev.size();
std::vector<unsigned> newedges(es, -1U);
unsigned edgeidx = 1;
for (unsigned e = 1; e < es; ++e)
{
if (is_dead_edge(e)
|| newst[ev[e].dst] == -1U
|| newst[ev[e].src] == -1U)
newedges[e] = -1U;
else
newedges[e] = edgeidx++;
}
std::map<unsigned, unsigned> he2;
for (auto [e, c]: *he)
// Let's just ignore unexisting edges. Raising an exception
// here would leave the automaton in a strange state.
if (SPOT_UNLIKELY(e > es))
continue;
else if (newedges[e] != -1U)
he2.emplace(newedges[e], c);
std::swap(*he, he2);
}
for (const char* prop: {"original-classes",
"original-states",
"degen-levels"})
if (auto os = get_named_prop<std::vector<unsigned>>(prop))
{
permute_vector(*os, newst);
os->resize(used_states);
}
if (auto ss = get_named_prop<std::vector<unsigned>>("simulated-states"))
{
for (auto& s : *ss)
{
if (s >= newst.size())
s = -1U;
else
s = newst[s];
}
}
// Reassign the state-players
if (auto sp = get_named_prop<std::vector<bool>>("state-player"))
{
permute_vector(*sp, newst);
sp->resize(used_states);
}
// Finally, update all states and edges.
init_number_ = newst[init_number_];
g_.defrag_states(newst, used_states);
}
void twa_graph::remove_unused_ap()
{
if (ap().empty())
return;
bdd all = ap_vars();
for (auto& e: g_.edges())
{
all = bdd_exist(all, bdd_support(e.cond));
if (all == bddtrue) // All APs are used.
return;
}
auto d = get_dict();
while (all != bddtrue)
{
unregister_ap(bdd_var(all));
all = bdd_high(all);
}
}
void twa_graph::copy_state_names_from(const const_twa_graph_ptr& other)
{
if (other == shared_from_this())
return;
auto orig = get_named_prop<std::vector<unsigned>>("original-states");
auto lvl = get_named_prop<std::vector<unsigned>>("degen-levels");
auto sims = get_named_prop<std::vector<unsigned>>("simulated-states");
assert(!lvl || orig);
if (orig && sims)
throw std::runtime_error("copy_state_names_from(): original-states and "
"simulated-states are both set");
if (orig && orig->size() != num_states())
throw std::runtime_error("copy_state_names_from(): unexpected size "
"for original-states");
if (lvl && lvl->size() != num_states())
throw std::runtime_error("copy_state_names_from(): unexpected size "
"for degen-levels");
if (sims && sims->size() != other->num_states())
throw std::runtime_error("copy_state_names_from(): unexpected size "
"for simulated-states");
auto names = std::unique_ptr<std::vector<std::string>>
(new std::vector<std::string>);
unsigned ns = num_states();
unsigned ons = other->num_states();
for (unsigned s = 0; s < ns; ++s)
{
std::string newname = "";
if (sims)
{
for (unsigned t = 0; t < ons; ++t)
{
if (s == (*sims)[t])
newname += other->format_state(t) + ',';
}
assert(!newname.empty());
newname.pop_back(); // remove trailing comma
newname = '[' + newname + ']';
}
else
{
unsigned other_s = orig ? (*orig)[s] : s;
if (other_s >= ons)
throw std::runtime_error("copy_state_names_from(): state does not"
" exist in source automaton");
newname = other->format_state(other_s);
if (lvl)
newname += '#' + std::to_string((*lvl)[s]);
}
names->emplace_back(newname);
}
set_named_prop("state-names", names.release());
}
void twa_graph::kill_state(unsigned state)
{
auto t = g_.out_iteraser(state);
while (t)
t.erase();
// A complete automaton is unlikely to stay
// complete after killing a state.
if (prop_complete().is_true())
prop_complete(trival::maybe());
prop_stutter_invariant(trival::maybe());
// Many properties are preserved by state removal, and may even
// become true if they were false before and the appropriate
// states are removed.
if (prop_universal().is_false())
prop_universal(trival::maybe());
if (prop_inherently_weak().is_false())
prop_inherently_weak(trival::maybe());
if (prop_weak().is_false())
prop_weak(trival::maybe());
if (prop_very_weak().is_false())
prop_very_weak(trival::maybe());
if (prop_terminal().is_false())
prop_terminal(trival::maybe());
if (prop_unambiguous().is_false())
prop_unambiguous(trival::maybe());
if (prop_semi_deterministic().is_false())
prop_semi_deterministic(trival::maybe());
}
void twa_graph::dump_storage_as_dot(std::ostream& out,
const char* opt) const
{
bool want_vectors = false;
bool want_data = false;
bool want_properties = false;
if (!opt || !*opt)
{
want_vectors = want_data = want_properties = true;
}
else
{
while (*opt)
switch (*opt++)
{
case 'v':
want_vectors = true;
break;
case 'd':
want_data = true;
break;
case 'p':
want_properties = true;
break;
default:
throw std::runtime_error
("dump_storage_as_dow(): unsupported option '"s + opt[-1] +"'");
}
}
const graph_t& g = get_graph();
g.dump_storage_as_dot(out, graph_t::DSI_GraphHeader);
out << "rankdir=BT\n";
if (want_vectors)
{
out << "{rank=same;\n";
g.dump_storage_as_dot(out, graph_t::DSI_States |
graph_t::DSI_EdgesHeader);
auto edges = g.edge_vector();
unsigned eend = edges.size();
out << "<tr><td>cond</td>\n";
for (unsigned e = 1; e < eend; ++e)
{
out << "<td>";
std::string f = bdd_format_formula(get_dict(), edges[e].cond);
escape_html(out, f);
out << "</td>\n";
}
out << "</tr>\n<tr><td>acc</td>\n";
for (unsigned e = 1; e < eend; ++e)
out << "<td>" << edges[e].acc << "</td>\n";
out << "</tr>\n";
g.dump_storage_as_dot(out, graph_t::DSI_EdgesBody
| graph_t::DSI_EdgesFooter
| graph_t::DSI_Dests);
out << "}\n";
}
if (want_data || want_properties)
{
out << "{rank=same;\n";
if (want_data)
{
out << ("meta [label=<\n"
"<table border='0' cellborder='0' cellspacing='0'>\n");
unsigned d = get_init_state_number();
out << ("<tr><td align='left'>init_state:</td>"
"<td align='left' bgcolor='");
if ((int)d < 0)
out << "pink'>~" << ~d;
else
out << "yellow'>" << d;
out << ("</td></tr><tr><td align='left'>num_sets:</td>"
"<td align='left' >")
<< num_sets()
<< ("</td></tr><tr><td align='left'>acceptance:</td>"
"<td align='left' >");
get_acceptance().to_html(out);
out << ("</td></tr><tr><td align='left'>ap_vars:</td>"
"<td align='left'>");
escape_html(out, bdd_format_sat(get_dict(), ap_vars()));
out << "</td></tr></table>>]\n";
}
if (want_properties)
{
out << ("props [label=<\n"
"<table border='0' cellborder='0' cellspacing='0'>\n");
#define print_prop(name) \
out << ("<tr><td align='left'>" #name ":</td>" \
"<td align='left' >") << name() << "</td></tr>\n";
print_prop(prop_state_acc);
print_prop(prop_inherently_weak);
print_prop(prop_terminal);
print_prop(prop_weak);
print_prop(prop_very_weak);
print_prop(prop_complete);
print_prop(prop_universal);
print_prop(prop_unambiguous);
print_prop(prop_semi_deterministic);
print_prop(prop_stutter_invariant);
#undef print_prop
out << "</table>>]\n";
if (!named_prop_.empty())
{
// GraphiViz 2.46.0 has a bug where plain newlines in
// quoted strings are ignored. See
// https://gitlab.com/graphviz/graphviz/-/issues/1931
// A workaround is to use emit \n instead of the
// actual new line.
out << "namedprops [label=\"named properties:\\n";
for (auto p: named_prop_)
escape_html(out, p.first) << "\\n";
out << "\"]\n";
}
}
out << "}\n";
}
if (want_data && want_vectors)
out << "meta -> states [style=invis]\n";
if (want_properties && want_vectors)
{
out << "props -> edges [style=invis]\n";
if (!named_prop_.empty())
{
out << "namedprops -> edges [style=invis]\n";
if (!is_existential())
out << "namedprops -> dests [style=invis]\n";
}
}
g.dump_storage_as_dot(out, graph_t::DSI_GraphFooter);
}
namespace
{
twa_graph_ptr
copy(const const_twa_ptr& aut, twa::prop_set p,
bool preserve_names, unsigned max_states)
{
// If the input is a twa_graph and the number of states is not
// restricted, simply use the make_twa_graph variant for
// twa_graph. Not only is this faster, but this is also
// necessary as a workaround for Swig-3 calling the wrong copy
// of make_twa_graph because it tests if twa match before
// testing twa_graph (swig-4 seems fixed).
const_twa_graph_ptr aut_g =
std::dynamic_pointer_cast<const twa_graph>(aut);
if (max_states == -1U && aut_g)
return make_twa_graph(aut_g, p, preserve_names);
twa_graph_ptr out = make_twa_graph(aut->get_dict());
out->copy_acceptance_of(aut);
out->copy_ap_of(aut);
out->prop_copy(aut, p);
std::vector<std::string>* names = nullptr;
std::set<unsigned>* incomplete = nullptr;
// Old highlighting maps
typedef std::map<unsigned, unsigned> hmap;
hmap* ohstates = nullptr;
hmap* ohedges = nullptr;
// New highlighting maps
hmap* nhstates = nullptr;
hmap* nhedges = nullptr;
if (preserve_names)
{
names = new std::vector<std::string>;
out->set_named_prop("state-names", names);
// If the input is a twa_graph and we were asked to preserve
// names, also preserve highlights.
if (aut_g)
{
ohstates = aut->get_named_prop<hmap>("highlight-states");
if (ohstates)
nhstates = out->get_or_set_named_prop<hmap>("highlight-states");
ohedges = aut->get_named_prop<hmap>("highlight-edges");
if (ohedges)
nhedges = out->get_or_set_named_prop<hmap>("highlight-edges");
}
}
// States already seen.
state_map<unsigned> seen;
// States to process
std::deque<state_map<unsigned>::value_type> todo;
auto new_state = [&](const state* s) -> unsigned
{
auto p = seen.emplace(s, 0);
if (p.second)
{
p.first->second = out->new_state();
todo.emplace_back(*p.first);
if (names)
names->emplace_back(aut->format_state(s));
if (ohstates)
{
auto q = ohstates->find(aut_g->state_number(s));
if (q != ohstates->end())
nhstates->emplace(p.first->second, q->second);
}
}
else
{
s->destroy();
}
return p.first->second;
};
// If the input is a kripke_graph and the number of states is
// not restricted, predeclare all states to keep their
// numbering, and also copy unreachable states.
if (max_states == -1U)
if (auto kg = std::dynamic_pointer_cast<const kripke_graph>(aut))
{
unsigned ns = kg->num_states();
for (unsigned s = 0; s < ns; ++s)
new_state(kg->state_from_number(s));
}
out->set_init_state(new_state(aut->get_init_state()));
while (!todo.empty())
{
const state* src1;
unsigned src2;
std::tie(src1, src2) = todo.front();
todo.pop_front();
for (auto* t: aut->succ(src1))
{
unsigned edgenum = 0;
if (SPOT_UNLIKELY(max_states < out->num_states()))
{
// If we have reached the max number of state, never try
// to create a new one.
auto i = seen.find(t->dst());
if (i == seen.end())
{
if (!incomplete)
incomplete = new std::set<unsigned>;
incomplete->insert(src2);
continue;
}
edgenum = out->new_edge(src2, i->second, t->cond(), t->acc());
}
else
{
edgenum = out->new_edge(src2, new_state(t->dst()),
t->cond(), t->acc());
}
if (ohedges)
{
auto q = ohedges->find(aut_g->edge_number(t));
if (q != ohedges->end())
nhedges->emplace(edgenum, q->second);
}
}
}
auto s = seen.begin();
while (s != seen.end())
{
// Advance the iterator before deleting the "key" pointer.
const state* ptr = s->first;
++s;
ptr->destroy();
}
if (incomplete)
out->set_named_prop("incomplete-states", incomplete);
return out;
}
}
twa_graph_ptr make_twa_graph(const const_twa_ptr& aut, twa::prop_set p,
bool preserve_names, unsigned max_states)
{
if (max_states == -1U && !preserve_names)
if (auto a = std::dynamic_pointer_cast<const twa_graph>(aut))
return SPOT_make_shared_enabled__(twa_graph, a, p);
return copy(aut, p, preserve_names, max_states);
}
}