spot/spot/twa/acc.hh
Alexandre Duret-Lutz 8568c3b423 postproc: integrate tgba_determinize()
* spot/twa/acc.hh: Add a smaller version of is_parity().
* spot/twaalgos/postproc.cc: Call tgba_determinize() if asked for
Generic acceptance and Deterministic output.
* bin/common_post.cc: Add 'G' as a shorthand for --generic.
* doc/org/ltl2tgba.org: Illustrate =ltl2tgba -G -D=.
2016-02-12 14:07:28 +01:00

1219 lines
26 KiB
C++

// -*- coding: utf-8 -*-
// Copyright (C) 2014, 2015, 2016 Laboratoire de Recherche et
// Développement de l'Epita.
//
// 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/>.
#pragma once
#include <functional>
#include <unordered_map>
#include <sstream>
#include <vector>
#include <spot/tl/defaultenv.hh>
#include <iostream>
namespace spot
{
class SPOT_API acc_cond
{
public:
struct mark_t
{
typedef unsigned value_t;
value_t id;
mark_t() = default;
mark_t(value_t id) noexcept
: id(id)
{
}
template<class iterator>
mark_t(const iterator& begin, const iterator& end) noexcept
{
id = 0U;
for (iterator i = begin; i != end; ++i)
set(*i);
}
mark_t(std::initializer_list<unsigned> vals) noexcept
: mark_t(vals.begin(), vals.end())
{
}
bool operator==(unsigned o) const
{
assert(o == 0U);
return id == o;
}
bool operator!=(unsigned o) const
{
assert(o == 0U);
return id != o;
}
bool operator==(mark_t o) const
{
return id == o.id;
}
bool operator!=(mark_t o) const
{
return id != o.id;
}
bool operator<(mark_t o) const
{
return id < o.id;
}
bool operator<=(mark_t o) const
{
return id <= o.id;
}
bool operator>(mark_t o) const
{
return id > o.id;
}
bool operator>=(mark_t o) const
{
return id >= o.id;
}
operator bool() const
{
return id != 0;
}
bool has(unsigned u) const
{
return id & (1U << u);
}
void set(unsigned u)
{
id |= (1U << u);
}
void clear(unsigned u)
{
id &= ~(1U << u);
}
mark_t& operator&=(mark_t r)
{
id &= r.id;
return *this;
}
mark_t& operator|=(mark_t r)
{
id |= r.id;
return *this;
}
mark_t& operator-=(mark_t r)
{
id &= ~r.id;
return *this;
}
mark_t& operator^=(mark_t r)
{
id ^= r.id;
return *this;
}
mark_t operator&(mark_t r) const
{
return id & r.id;
}
mark_t operator|(mark_t r) const
{
return id | r.id;
}
mark_t operator-(mark_t r) const
{
return id & ~r.id;
}
mark_t operator~() const
{
return ~id;
}
mark_t operator^(mark_t r) const
{
return id ^ r.id;
}
mark_t operator<<(unsigned i) const
{
return id << i;
}
mark_t& operator<<=(unsigned i)
{
id <<= i;
return *this;
}
mark_t operator>>(unsigned i) const
{
return id >> i;
}
mark_t& operator>>=(unsigned i)
{
id >>= i;
return *this;
}
mark_t strip(mark_t y) const
{
// strip every bit of id that is marked in y
// 100101110100.strip(
// 001011001000)
// == 10 1 11 100
// == 10111100
auto xv = id; // 100101110100
auto yv = y.id; // 001011001000
while (yv && xv)
{
// Mask for everything after the last 1 in y
auto rm = (~yv) & (yv - 1); // 000000000111
// Mask for everything before the last 1 in y
auto lm = ~(yv ^ (yv - 1)); // 111111110000
xv = ((xv & lm) >> 1) | (xv & rm);
yv = (yv & lm) >> 1;
}
return xv;
}
// Number of bits sets.
unsigned count() const
{
#ifdef __GNUC__
return __builtin_popcount(id);
#else
unsigned c = 0U;
auto v = id;
while (v)
{
++c;
v &= v - 1;
}
return c;
#endif
}
// Return the number of the highest set used plus one.
// So if no set is used, this returns 0,
// but if the sets {1,3,8} are used, this returns 9.
unsigned max_set() const
{
auto i = id;
int res = 0;
while (i)
{
++res;
i >>= 1;
}
return res;
}
// Return the lowest acceptance mark
mark_t lowest() const
{
return id & -id;
}
// Remove n bits that where set
mark_t& remove_some(unsigned n)
{
while (n--)
id &= id - 1;
return *this;
}
template<class iterator>
void fill(iterator here) const
{
auto a = id;
unsigned level = 0;
while (a)
{
if (a & 1)
*here++ = level;
++level;
a >>= 1;
}
}
// FIXME: Return some iterable object without building a vector.
std::vector<unsigned> sets() const
{
std::vector<unsigned> res;
fill(std::back_inserter(res));
return res;
}
SPOT_API
friend std::ostream& operator<<(std::ostream& os, mark_t m);
};
// This encodes either an operator or set of acceptance sets.
enum class acc_op : unsigned short
{ Inf, Fin, InfNeg, FinNeg, And, Or };
union acc_word
{
mark_t mark;
struct {
acc_op op; // Operator
unsigned short size; // Size of the subtree (number of acc_word),
// not counting this node.
};
};
struct SPOT_API acc_code: public std::vector<acc_word>
{
bool operator==(const acc_code& other) const
{
unsigned pos = size();
if (other.size() != pos)
return false;
while (pos > 0)
{
auto op = (*this)[pos - 1].op;
auto sz = (*this)[pos - 1].size;
if (other[pos - 1].op != op ||
other[pos - 1].size != sz)
return false;
switch (op)
{
case acc_cond::acc_op::And:
case acc_cond::acc_op::Or:
--pos;
break;
case acc_cond::acc_op::Inf:
case acc_cond::acc_op::InfNeg:
case acc_cond::acc_op::Fin:
case acc_cond::acc_op::FinNeg:
pos -= 2;
if (other[pos].mark != (*this)[pos].mark)
return false;
break;
}
}
return true;
};
bool operator<(const acc_code& other) const
{
unsigned pos = size();
auto osize = other.size();
if (pos < osize)
return true;
if (pos > osize)
return false;
while (pos > 0)
{
auto op = (*this)[pos - 1].op;
auto oop = other[pos - 1].op;
if (op < oop)
return true;
if (op > oop)
return false;
auto sz = (*this)[pos - 1].size;
auto osz = other[pos - 1].size;
if (sz < osz)
return true;
if (sz > osz)
return false;
switch (op)
{
case acc_cond::acc_op::And:
case acc_cond::acc_op::Or:
--pos;
break;
case acc_cond::acc_op::Inf:
case acc_cond::acc_op::InfNeg:
case acc_cond::acc_op::Fin:
case acc_cond::acc_op::FinNeg:
pos -= 2;
auto m = (*this)[pos].mark;
auto om = other[pos].mark;
if (m < om)
return true;
if (m > om)
return false;
break;
}
}
return false;
}
bool operator>(const acc_code& other) const
{
return other < *this;
}
bool operator<=(const acc_code& other) const
{
return !(other < *this);
}
bool operator>=(const acc_code& other) const
{
return !(*this < other);
}
bool operator!=(const acc_code& other) const
{
return !(*this == other);
}
bool is_t() const
{
unsigned s = size();
return s == 0
|| ((*this)[s - 1].op == acc_op::Inf && (*this)[s - 2].mark == 0U);
}
bool is_f() const
{
unsigned s = size();
return s > 1
&& (*this)[s - 1].op == acc_op::Fin && (*this)[s - 2].mark == 0U;
}
static acc_code f()
{
acc_code res;
res.resize(2);
res[0].mark = 0U;
res[1].op = acc_op::Fin;
res[1].size = 1;
return res;
}
static acc_code t()
{
return {};
}
static acc_code fin(mark_t m)
{
acc_code res;
res.resize(2);
res[0].mark = m;
res[1].op = acc_op::Fin;
res[1].size = 1;
return res;
}
static acc_code fin(std::initializer_list<unsigned> vals)
{
return fin(mark_t(vals));
}
static acc_code fin_neg(mark_t m)
{
acc_code res;
res.resize(2);
res[0].mark = m;
res[1].op = acc_op::FinNeg;
res[1].size = 1;
return res;
}
static acc_code fin_neg(std::initializer_list<unsigned> vals)
{
return fin_neg(mark_t(vals));
}
static acc_code inf(mark_t m)
{
acc_code res;
res.resize(2);
res[0].mark = m;
res[1].op = acc_op::Inf;
res[1].size = 1;
return res;
}
static acc_code inf(std::initializer_list<unsigned> vals)
{
return inf(mark_t(vals));
}
static acc_code inf_neg(mark_t m)
{
acc_code res;
res.resize(2);
res[0].mark = m;
res[1].op = acc_op::InfNeg;
res[1].size = 1;
return res;
}
static acc_code inf_neg(std::initializer_list<unsigned> vals)
{
return inf_neg(mark_t(vals));
}
static acc_code buchi()
{
return inf({0});
}
static acc_code cobuchi()
{
return fin({0});
}
static acc_code generalized_buchi(unsigned n)
{
acc_cond::mark_t m = (1U << n) - 1;
if (n == 8 * sizeof(mark_t::value_t))
m = mark_t(-1U);
return inf(m);
}
static acc_code generalized_co_buchi(unsigned n)
{
acc_cond::mark_t m = (1U << n) - 1;
if (n == 8 * sizeof(mark_t::value_t))
m = mark_t(-1U);
return fin(m);
}
// n is a number of pairs.
static acc_code rabin(unsigned n)
{
acc_cond::acc_code res = f();
while (n > 0)
{
res |= inf({2*n - 1}) & fin({2*n - 2});
--n;
}
return res;
}
// n is a number of pairs.
static acc_code streett(unsigned n)
{
acc_cond::acc_code res = t();
while (n > 0)
{
res &= inf({2*n - 1}) | fin({2*n - 2});
--n;
}
return res;
}
template<class Iterator>
static acc_code generalized_rabin(Iterator begin, Iterator end)
{
acc_cond::acc_code res = f();
unsigned n = 0;
for (Iterator i = begin; i != end; ++i)
{
acc_cond::acc_code pair = fin({n++});
acc_cond::mark_t m = 0U;
for (unsigned ni = *i; ni > 0; --ni)
m.set(n++);
pair &= inf(m);
std::swap(pair, res);
res |= std::move(pair);
}
return res;
}
static acc_code parity(bool max, bool odd, unsigned sets);
// Number of acceptance sets to use, and probability to reuse
// each set another time after it has been used in the
// acceptance formula.
static acc_code random(unsigned n, double reuse = 0.0);
acc_code& operator&=(acc_code&& r)
{
if (is_t() || r.is_f())
{
*this = std::move(r);
return *this;
}
if (is_f() || r.is_t())
return *this;
unsigned s = size() - 1;
unsigned rs = r.size() - 1;
// We want to group all Inf(x) operators:
// Inf(a) & Inf(b) = Inf(a & b)
if (((*this)[s].op == acc_op::Inf && r[rs].op == acc_op::Inf)
|| ((*this)[s].op == acc_op::InfNeg && r[rs].op == acc_op::InfNeg))
{
(*this)[s - 1].mark |= r[rs - 1].mark;
return *this;
}
// In the more complex scenarios, left and right may both
// be conjunctions, and Inf(x) might be a member of each
// side. Find it if it exists.
// left_inf points to the left Inf mark if any.
// right_inf points to the right Inf mark if any.
acc_word* left_inf = nullptr;
if ((*this)[s].op == acc_op::And)
{
auto start = &(*this)[s] - (*this)[s].size;
auto pos = &(*this)[s] - 1;
pop_back();
while (pos > start)
{
if (pos->op == acc_op::Inf)
{
left_inf = pos - 1;
break;
}
pos -= pos->size + 1;
}
}
else if ((*this)[s].op == acc_op::Inf)
{
left_inf = &(*this)[s - 1];
}
acc_word* right_inf = nullptr;
auto right_end = &r.back();
if (right_end->op == acc_op::And)
{
auto start = &r[0];
auto pos = --right_end;
while (pos > start)
{
if (pos->op == acc_op::Inf)
{
right_inf = pos - 1;
break;
}
pos -= pos->size + 1;
}
}
else if (right_end->op == acc_op::Inf)
{
right_inf = right_end - 1;
}
if (left_inf && right_inf)
{
left_inf->mark |= right_inf->mark;
insert(this->end(), &r[0], right_inf);
insert(this->end(), right_inf + 2, right_end + 1);
}
else if (right_inf)
{
// Always insert Inf() at the very first entry.
insert(this->begin(), right_inf, right_inf + 2);
insert(this->end(), &r[0], right_inf);
insert(this->end(), right_inf + 2, right_end + 1);
}
else
{
insert(this->end(), &r[0], right_end + 1);
}
acc_word w;
w.op = acc_op::And;
w.size = size();
push_back(w);
return *this;
}
acc_code& operator&=(const acc_code& r)
{
if (is_t() || r.is_f())
{
*this = r;
return *this;
}
if (is_f() || r.is_t())
return *this;
unsigned s = size() - 1;
unsigned rs = r.size() - 1;
// Inf(a) & Inf(b) = Inf(a & b)
if (((*this)[s].op == acc_op::Inf && r[rs].op == acc_op::Inf)
|| ((*this)[s].op == acc_op::InfNeg && r[rs].op == acc_op::InfNeg))
{
(*this)[s - 1].mark |= r[rs - 1].mark;
return *this;
}
// In the more complex scenarios, left and right may both
// be conjunctions, and Inf(x) might be a member of each
// side. Find it if it exists.
// left_inf points to the left Inf mark if any.
// right_inf points to the right Inf mark if any.
acc_word* left_inf = nullptr;
if ((*this)[s].op == acc_op::And)
{
auto start = &(*this)[s] - (*this)[s].size;
auto pos = &(*this)[s] - 1;
pop_back();
while (pos > start)
{
if (pos->op == acc_op::Inf)
{
left_inf = pos - 1;
break;
}
pos -= pos->size + 1;
}
}
else if ((*this)[s].op == acc_op::Inf)
{
left_inf = &(*this)[s - 1];
}
const acc_word* right_inf = nullptr;
auto right_end = &r.back();
if (right_end->op == acc_op::And)
{
auto start = &r[0];
auto pos = --right_end;
while (pos > start)
{
if (pos->op == acc_op::Inf)
{
right_inf = pos - 1;
break;
}
pos -= pos->size + 1;
}
}
else if (right_end->op == acc_op::Inf)
{
right_inf = right_end - 1;
}
if (left_inf && right_inf)
{
left_inf->mark |= right_inf->mark;
insert(this->end(), &r[0], right_inf);
insert(this->end(), right_inf + 2, right_end + 1);
}
else if (right_inf)
{
// Always insert Inf() at the very first entry.
insert(this->begin(), right_inf, right_inf + 2);
insert(this->end(), &r[0], right_inf);
insert(this->end(), right_inf + 2, right_end + 1);
}
else
{
insert(this->end(), &r[0], right_end + 1);
}
acc_word w;
w.op = acc_op::And;
w.size = size();
push_back(w);
return *this;
}
acc_code operator&(acc_code&& r)
{
acc_code res = *this;
res &= r;
return res;
}
acc_code operator&(const acc_code& r)
{
acc_code res = *this;
res &= r;
return res;
}
acc_code& operator|=(acc_code&& r)
{
if (is_t() || r.is_f())
return *this;
if (is_f() || r.is_t())
{
*this = std::move(r);
return *this;
}
unsigned s = size() - 1;
unsigned rs = r.size() - 1;
// Fin(a) | Fin(b) = Fin(a | b)
if (((*this)[s].op == acc_op::Fin && r[rs].op == acc_op::Fin)
|| ((*this)[s].op == acc_op::FinNeg && r[rs].op == acc_op::FinNeg))
{
(*this)[s - 1].mark |= r[rs - 1].mark;
return *this;
}
if ((*this)[s].op == acc_op::Or)
pop_back();
if (r.back().op == acc_op::Or)
r.pop_back();
insert(this->end(), r.begin(), r.end());
acc_word w;
w.op = acc_op::Or;
w.size = size();
push_back(w);
return *this;
}
acc_code& operator|=(const acc_code& r)
{
return *this |= acc_code(r);
}
acc_code operator|(acc_code&& r)
{
acc_code res = *this;
res |= r;
return res;
}
acc_code operator|(const acc_code& r)
{
acc_code res = *this;
res |= r;
return res;
}
acc_code& operator<<=(unsigned sets)
{
if (empty())
return *this;
unsigned pos = size();
do
{
switch ((*this)[pos - 1].op)
{
case acc_cond::acc_op::And:
case acc_cond::acc_op::Or:
--pos;
break;
case acc_cond::acc_op::Inf:
case acc_cond::acc_op::InfNeg:
case acc_cond::acc_op::Fin:
case acc_cond::acc_op::FinNeg:
pos -= 2;
(*this)[pos].mark.id <<= sets;
break;
}
}
while (pos > 0);
return *this;
}
acc_code operator<<(unsigned sets) const
{
acc_code res = *this;
res <<= sets;
return res;
}
bool is_dnf() const;
bool is_cnf() const;
acc_code to_dnf() const;
acc_code to_cnf() const;
acc_code complement() const;
// Return a list of acceptance marks needed to close a cycle
// that already visit INF infinitely often, so that the cycle is
// accepting (ACCEPTING=true) or rejecting (ACCEPTING=false).
// Positive values describe positive set.
// A negative value x means the set -x-1 must be absent.
std::vector<std::vector<int>>
missing(mark_t inf, bool accepting) const;
bool accepting(mark_t inf) const;
bool inf_satisfiable(mark_t inf) const;
// Remove all the acceptance sets in rem.
//
// If MISSING is set, the acceptance sets are assumed to be
// missing from the automaton, and the acceptance is updated to
// reflect this. For instance (Inf(1)&Inf(2))|Fin(3) will
// become Fin(3) if we remove 2 because it is missing from this
// automaton, because there is no way to fulfill Inf(1)&Inf(2)
// in this case. So essentially MISSING causes Inf(rem) to
// become f, and Fin(rem) to become t.
//
// If MISSING is unset, Inf(rem) become t while Fin(rem) become
// f. Removing 2 from (Inf(1)&Inf(2))|Fin(3) would then give
// Inf(1)|Fin(3).
acc_code strip(acc_cond::mark_t rem, bool missing) const;
// Return the set of sets appearing in the condition.
acc_cond::mark_t used_sets() const;
// Return the sets used as Inf or Fin in the acceptance condition
std::pair<acc_cond::mark_t, acc_cond::mark_t> used_inf_fin_sets() const;
// Print the acceptance as HTML. The set_printer function can
// be used to implement customized output for set numbers.
std::ostream&
to_html(std::ostream& os,
std::function<void(std::ostream&, int)>
set_printer = nullptr) const;
// Print the acceptance as text. The set_printer function can
// be used to implement customized output for set numbers.
std::ostream&
to_text(std::ostream& os,
std::function<void(std::ostream&, int)>
set_printer = nullptr) const;
/// \brief Construct an acc_code from a string.
///
/// The string can follow the following grammar:
///
/// <pre>
/// acc ::= "t"
/// | "f"
/// | "Inf" "(" num ")"
/// | "Fin" "(" num ")"
/// | "(" acc ")"
/// | acc "&" acc
/// | acc "|" acc
/// </pre>
///
/// Where num is an integer and "&" has priority over "|". Note that
/// "Fin(!x)" and "Inf(!x)" are not supported by this parser.
///
/// Or the string can be the name of an acceptance condition, as
/// speficied in the HOA format. (E.g. "Rabin 2", "parity max odd 3",
/// "generalized-Rabin 4 2 1", etc.).
///
/// A spot::parse_error is thrown on syntax error.
acc_code(const char* input);
/// \brief Build an empty acceptance condition.
///
/// This is the same as t().
acc_code()
{
}
// Calls to_text
SPOT_API
friend std::ostream& operator<<(std::ostream& os, const acc_code& code);
};
acc_cond(unsigned n_sets = 0, const acc_code& code = {})
: num_(0U), all_(0U), code_(code)
{
add_sets(n_sets);
uses_fin_acceptance_ = check_fin_acceptance();
}
acc_cond(const acc_code& code)
: num_(0U), all_(0U), code_(code)
{
add_sets(code.used_sets().max_set());
uses_fin_acceptance_ = check_fin_acceptance();
}
acc_cond(const acc_cond& o)
: num_(o.num_), all_(o.all_), code_(o.code_),
uses_fin_acceptance_(o.uses_fin_acceptance_)
{
}
~acc_cond()
{
}
void set_acceptance(const acc_code& code)
{
code_ = code;
uses_fin_acceptance_ = check_fin_acceptance();
}
const acc_code& get_acceptance() const
{
return code_;
}
acc_code& get_acceptance()
{
return code_;
}
bool uses_fin_acceptance() const
{
return uses_fin_acceptance_;
}
bool is_t() const
{
return code_.is_t();
}
bool is_all() const
{
return num_ == 0 && is_t();
}
bool is_f() const
{
return code_.is_f();
}
bool is_none() const
{
return num_ == 0 && is_f();
}
bool is_buchi() const
{
unsigned s = code_.size();
return num_ == 1 &&
s == 2 && code_[1].op == acc_op::Inf && code_[0].mark == all_sets();
}
bool is_co_buchi() const
{
return num_ == 1 && is_generalized_co_buchi();
}
void set_generalized_buchi()
{
set_acceptance(inf(all_sets()));
}
bool is_generalized_buchi() const
{
unsigned s = code_.size();
return (s == 0 && num_ == 0) ||
(s == 2 && code_[1].op == acc_op::Inf && code_[0].mark == all_sets());
}
bool is_generalized_co_buchi() const
{
unsigned s = code_.size();
return (s == 2 &&
code_[1].op == acc_op::Fin && code_[0].mark == all_sets());
}
// Returns a number of pairs (>=0) if Rabin, or -1 else.
int is_rabin() const;
// Returns a number of pairs (>=0) if Streett, or -1 else.
int is_streett() const;
// Return the number of Inf in each pair.
bool is_generalized_rabin(std::vector<unsigned>& pairs) const;
// If EQUIV is false, this return true iff the acceptance
// condition is a parity condition written in the canonical way
// given in the HOA specifications. If EQUIV is true, then we
// check whether the condition is logically equivalent to some
// parity acceptance condition.
bool is_parity(bool& max, bool& odd, bool equiv = false) const;
bool is_parity() const
{
bool max;
bool min;
return is_parity(max, min);
}
// Return (true, m) if there exist some acceptance mark m that
// does not satisfy the acceptance condition. Return (false, 0U)
// otherwise.
std::pair<bool, acc_cond::mark_t> unsat_mark() const;
protected:
bool check_fin_acceptance() const;
public:
acc_code inf(mark_t mark) const
{
return acc_code::inf(mark);
}
acc_code inf(std::initializer_list<unsigned> vals) const
{
return inf(mark_t(vals.begin(), vals.end()));
}
acc_code inf_neg(mark_t mark) const
{
return acc_code::inf_neg(mark);
}
acc_code inf_neg(std::initializer_list<unsigned> vals) const
{
return inf_neg(mark_t(vals.begin(), vals.end()));
}
acc_code fin(mark_t mark) const
{
return acc_code::fin(mark);
}
acc_code fin(std::initializer_list<unsigned> vals) const
{
return fin(mark_t(vals.begin(), vals.end()));
}
acc_code fin_neg(mark_t mark) const
{
return acc_code::fin_neg(mark);
}
acc_code fin_neg(std::initializer_list<unsigned> vals) const
{
return fin_neg(mark_t(vals.begin(), vals.end()));
}
unsigned add_sets(unsigned num)
{
if (num == 0)
return -1U;
unsigned j = num_;
num_ += num;
if (num_ > 8 * sizeof(mark_t::id))
throw std::runtime_error("Too many acceptance sets used.");
all_ = all_sets_();
return j;
}
unsigned add_set()
{
return add_sets(1);
}
mark_t mark(unsigned u) const
{
assert(u < num_sets());
return 1U << u;
}
mark_t comp(mark_t l) const
{
return all_ ^ l.id;
}
mark_t all_sets() const
{
return all_;
}
bool accepting(mark_t inf) const
{
return code_.accepting(inf);
}
bool inf_satisfiable(mark_t inf) const
{
return code_.inf_satisfiable(inf);
}
mark_t accepting_sets(mark_t inf) const;
std::ostream& format(std::ostream& os, mark_t m) const
{
auto a = m;
if (a == 0U)
return os;
return os << m;
}
std::string format(mark_t m) const
{
std::ostringstream os;
format(os, m);
return os.str();
}
unsigned num_sets() const
{
return num_;
}
template<class iterator>
mark_t useless(iterator begin, iterator end) const
{
mark_t::value_t u = 0U; // The set of useless marks.
for (unsigned x = 0; x < num_; ++x)
{
// Skip marks that are already known to be useless.
if (u & (1 << x))
continue;
unsigned all = all_ ^ (u | (1 << x));
for (iterator y = begin; y != end; ++y)
{
auto v = y->id;
if (v & (1 << x))
{
all &= v;
if (!all)
break;
}
}
u |= all;
}
return u;
}
protected:
mark_t::value_t all_sets_() const
{
if (num_ == 0)
return 0;
return -1U >> (8 * sizeof(mark_t::value_t) - num_);
}
unsigned num_;
mark_t::value_t all_;
acc_code code_;
bool uses_fin_acceptance_ = false;
};
SPOT_API
std::ostream& operator<<(std::ostream& os, const acc_cond& acc);
}
namespace std
{
template<>
struct hash<spot::acc_cond::mark_t>
{
size_t operator()(spot::acc_cond::mark_t m) const
{
std::hash<decltype(m.id)> h;
return h(m.id);
}
};
}