spot/spot/parseaut/parseaut.yy
Alexandre Duret-Lutz ba86dc6b18 twa: guard against highlighting of non-existing edges and states
For issue #556, reported by Dávid Smolka.  Not sure if this is a
complete fix yet, but that's at least part of the issue.

* spot/twa/twagraph.cc (defrag_states): Skip those unexisting edges
and states while remaping highlight-states and highlight-edges.
* spot/parseaut/parseaut.yy: Likewise for dropped edges.
* tests/python/parsetgba.py: Augment test case.
2023-12-04 22:15:37 +01:00

3010 lines
85 KiB
Text

/* -*- 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/>.
*/
%require "3.0"
%language "C++"
%locations
%defines
%expect 0 // No shift/reduce
%define api.prefix {hoayy}
%debug
%define parse.error verbose
%parse-param {void* scanner}
%lex-param {void* scanner} { PARSE_ERROR_LIST }
%define api.location.type {spot::location}
%code requires
{
#include "config.h"
#include <spot/misc/common.hh>
#include <spot/priv/robin_hood.hh>
#include <string>
#include <cstring>
#include <sstream>
#include <unordered_map>
#include <algorithm>
#include <spot/twa/formula2bdd.hh>
#include <spot/parseaut/public.hh>
#include "spot/priv/accmap.hh"
#include <spot/tl/parse.hh>
#include <spot/twaalgos/alternation.hh>
#include <spot/twaalgos/game.hh>
using namespace std::string_literals;
#ifndef HAVE_STRVERSCMP
// If the libc does not have this, a version is compiled in lib/.
extern "C" int strverscmp(const char *s1, const char *s2);
#endif
// Work around Bison not letting us write
// %lex-param { res.h->errors, res.fcache }
#define PARSE_ERROR_LIST res.h->errors, res.fcache
inline namespace hoayy_support
{
typedef std::map<int, bdd> map_t;
/* Cache parsed formulae. Labels on arcs are frequently identical
and it would be a waste of time to parse them to formula
over and over, and to register all their atomic_propositions in
the bdd_dict. Keep the bdd result around so we can reuse
it. */
typedef robin_hood::unordered_flat_map<std::string, bdd> formula_cache;
typedef std::pair<int, std::string*> pair;
typedef spot::twa_graph::namer<std::string> named_tgba_t;
// Note: because this parser is meant to be used on a stream of
// automata, it tries hard to recover from errors, so that we get a
// chance to reach the end of the current automaton in order to
// process the next one. Several variables below are used to keep
// track of various error conditions.
enum label_style_t { Mixed_Labels, State_Labels, Trans_Labels,
Implicit_Labels };
enum acc_style_t { Mixed_Acc, State_Acc, Trans_Acc };
struct result_
{
struct state_info
{
bool declared = false;
bool used = false;
spot::location used_loc;
};
spot::parsed_aut_ptr h;
spot::twa_ptr aut_or_ks;
spot::automaton_parser_options opts;
std::string format_version;
spot::location format_version_loc;
spot::environment* env;
formula_cache fcache;
named_tgba_t* namer = nullptr;
spot::acc_mapper_int* acc_mapper = nullptr;
std::vector<int> ap;
std::vector<int> controllable_ap;
bool has_controllable_ap = false;
std::vector<spot::location> controllable_ap_loc;
spot::location controllable_aps_loc;
std::vector<bdd> guards;
std::vector<bdd>::const_iterator cur_guard;
// If "Alias: ..." occurs before "AP: ..." in the HOA format we
// are in trouble because the parser would like to turn each
// alias into a BDD, yet the atomic proposition have not been
// declared yet. We solve that by using arbitrary BDD variables
// numbers (in fact we use the same number given in the Alias:
// definition) and keeping track of the highest variable number
// we have seen (unknown_ap_max). Once AP: is encountered,
// we can remap everything. If AP: is never encountered an
// unknown_ap_max is non-negative, then we can signal an error.
int unknown_ap_max = -1;
spot::location unknown_ap_max_location;
bool in_alias = false;
map_t dest_map;
std::vector<state_info> info_states; // States declared and used.
// Mapping of edges in the HOA file to edges in the automaton.
// Edges are counted from 0 in the HOA file and from 1 in the
// automaton. Given edge #i in the HOA file, edge_map[i] gives
// corresponding edge in the automaton, or 0 if that edge was
// removed (because labeled by bddfalse). This map is used to
// update properties such as highlight_edges after the automaton
// has been read. Note that the parser may also introduce
// unlisted edges, e.g., a bddfalse self-loop to hold the
// acceptance of a state without declared outgoing edge. Those
// added edges are not a concern for this edge_map.
std::vector<unsigned> edge_map;
std::vector<std::pair<spot::location,
std::vector<unsigned>>> start; // Initial states;
std::unordered_map<std::string, bdd> alias;
std::vector<std::string> alias_order;
struct prop_info
{
spot::location loc;
bool val;
operator bool() const
{
return val;
};
};
std::unordered_map<std::string, prop_info> props;
spot::location states_loc;
spot::location ap_loc;
spot::location state_label_loc;
spot::location accset_loc;
spot::acc_cond::mark_t acc_state;
spot::acc_cond::mark_t neg_acc_sets = {};
spot::acc_cond::mark_t pos_acc_sets = {};
int plus;
int minus;
std::vector<std::string>* state_names = nullptr;
std::map<unsigned, unsigned>* highlight_edges = nullptr;
std::map<unsigned, unsigned>* highlight_states = nullptr;
std::map<unsigned, unsigned> states_map;
std::vector<bool>* state_player = nullptr;
spot::location state_player_loc;
std::set<int> ap_set;
unsigned cur_state;
int states = -1;
int ap_count = -1;
int accset = -1;
bdd state_label;
bdd cur_label;
bool has_state_label = false;
bool ignore_more_ap = false; // Set to true after the first "AP:"
// line has been read.
bool ignore_acc = false; // Set to true in case of missing
// Acceptance: lines.
bool ignore_acc_silent = false;
bool ignore_more_acc = false; // Set to true after the first
// "Acceptance:" line has been read.
label_style_t label_style = Mixed_Labels;
acc_style_t acc_style = Mixed_Acc;
bool accept_all_needed = false;
bool accept_all_seen = false;
bool aliased_states = false;
spot::trival universal = spot::trival::maybe();
spot::trival existential = spot::trival::maybe();
spot::trival complete = spot::trival::maybe();
bool trans_acc_seen = false;
std::map<std::string, spot::location> labels;
prop_info prop_is_true(const std::string& p)
{
auto i = props.find(p);
if (i == props.end())
return prop_info{spot::location(), false};
return i->second;
}
prop_info prop_is_false(const std::string& p)
{
auto i = props.find(p);
if (i == props.end())
return prop_info{spot::location(), false};
return prop_info{i->second.loc, !i->second.val};
}
~result_()
{
delete namer;
delete acc_mapper;
}
};
}
}
%parse-param {result_& res}
%parse-param {spot::location initial_loc}
%initial-action { @$ = res.h->loc = initial_loc; }
%union
{
std::string* str;
unsigned int num;
int b;
spot::acc_cond::mark_t mark;
pair* p;
std::list<pair>* list;
spot::acc_cond::acc_code* code;
std::vector<unsigned>* states;
}
%code
{
#include <sstream>
/* parseaut.hh and parsedecl.hh include each other recursively.
We must ensure that YYSTYPE is declared (by the above %union)
before parsedecl.hh uses it. */
#include <spot/parseaut/parsedecl.hh>
static void fill_guards(result_& res);
}
/**** HOA tokens ****/
%token HOA "HOA:"
%token STATES "States:"
%token START "Start:"
%token AP "AP:"
%token ALIAS "Alias:"
%token ACCEPTANCE "Acceptance:"
%token ACCNAME "acc-name:"
%token CONTROLLABLE_AP "controllable-AP:"
%token TOOL "tool:"
%token NAME "name:"
%token PROPERTIES "properties:"
%token BODY "--BODY--"
%token END "--END--"
%token STATE "State:"
%token SPOT_HIGHLIGHT_EDGES "spot.highlight.edges:"
%token SPOT_HIGHLIGHT_STATES "spot.highlight.states:"
%token SPOT_STATE_PLAYER "spot.state-player:"
%token <str> IDENTIFIER "identifier" // also used by neverclaim
%token <str> HEADERNAME "header name"
%token <str> ANAME "alias name"
%token <str> STRING "string"
%token <num> INT "integer"
%token ENDOFFILE 0 "end of file"
%token <str> '['
%token <str> LINEDIRECTIVE "#line"
%token <b> BDD
/**** DSTAR tokens ****/
%token ENDDSTAR "end of DSTAR automaton"
%token DRA "DRA"
%token DSA "DSA"
%token V2 "v2"
%token EXPLICIT "explicit"
%token ACCPAIRS "Acceptance-Pairs:"
%token ACCSIG "Acc-Sig:"
%token ENDOFHEADER "---"
%left '|'
%left '&'
%precedence '!'
%type <states> init-state-conj-2 state-conj-2 state-conj-checked pgame_succs
%type <num> checked-state-num state-num acc-set sign
%type <b> label-expr
%type <mark> acc-sig acc-sets trans-acc_opt state-acc_opt
dstar_accsigs dstar_state_accsig
%type <str> string_opt
/**** NEVERCLAIM tokens ****/
%token NEVER "never"
%token SKIP "skip"
%token IF "if"
%token FI "fi"
%token DO "do"
%token OD "od"
%token ARROW "->"
%token GOTO "goto"
%token FALSE "false"
%token ATOMIC "atomic"
%token ASSERT "assert"
%token <str> FORMULA "boolean formula"
%type <b> nc-formula
%type <str> nc-opt-dest nc-formula-or-ident
%type <p> nc-transition nc-src-dest
%type <list> nc-transitions nc-transition-block
%type <str> nc-one-ident nc-ident-list
%type <code> acceptance-cond
/**** PGAME tokens ****/
// Also using INT, STRING
%token PGAME "start of PGSolver game"
%token ENDPGAME "end of PGSolver game"
/**** LBTT tokens *****/
// Also using INT, STRING
%token ENDAUT "-1"
%token <num> LBTT "LBTT header"
%token <num> INT_S "state acceptance"
%token <num> LBTT_EMPTY "acceptance sets for empty automaton"
%token <num> ACC "acceptance set"
%token <num> STATE_NUM "state number"
%token <num> DEST_NUM "destination number"
%type <mark> lbtt-acc
%destructor { delete $$; } <str>
%destructor { bdd_delref($$); } <b>
%destructor { bdd_delref($$->first); delete $$->second; delete $$; } <p>
%destructor { delete $$; } <code>
%destructor { delete $$; } <states>
%printer {
auto& os = debug_stream();
os << '{';
bool notfirst = false;
for (auto i: *$$)
{
if (notfirst)
os << ", ";
else
notfirst = true;
os << i;
}
os << '}';
} <states>
%destructor {
for (auto i = $$->begin(); i != $$->end(); ++i)
{
bdd_delref(i->first);
delete i->second;
}
delete $$;
} <list>
%printer {
if ($$)
debug_stream() << *$$;
else
debug_stream() << "\"\""; } <str>
%printer { debug_stream() << $$; } <num>
%%
aut: aut-1 { res.h->loc = @$; YYACCEPT; }
/* Not how we do not accept garbage before LINEDIRECTIVE. This is because
all error will be relative to the specified filename, so that filename
must not be changed after any error was reported. */
| LINEDIRECTIVE { res.h->filename = *$1; } aut
| ENDOFFILE { YYABORT; }
| error ENDOFFILE { YYABORT; }
| error aut-1
{
error(@1, "leading garbage was ignored");
res.h->loc = @2;
YYACCEPT;
}
aut-1: hoa { res.h->type = spot::parsed_aut_type::HOA; }
| never { res.h->type = spot::parsed_aut_type::NeverClaim; }
| lbtt { res.h->type = spot::parsed_aut_type::LBTT; }
| dstar /* will set type as DSA or DRA while parsing first line */
| pgame { res.h->type = spot::parsed_aut_type::PGAME; }
/**********************************************************************/
/* Rules for HOA */
/**********************************************************************/
hoa: header "--BODY--" body "--END--"
| "HOA:" error "--END--"
string_opt: %empty { $$ = nullptr; }
| STRING { $$ = $1; }
BOOLEAN: 't' | 'f'
header: format-version header-items
{
bool v1plus = strverscmp("v1", res.format_version.c_str()) < 0;
// Preallocate the states if we know their number.
if (res.states >= 0)
{
unsigned states = res.states;
for (auto& p : res.start)
for (unsigned s: p.second)
if ((unsigned) res.states <= s)
{
error(p.first, "initial state number is larger "
"than state count...");
error(res.states_loc, "... declared here.");
states = std::max(states, s + 1);
}
if (res.opts.want_kripke)
res.h->ks->new_states(states, bddfalse);
else
res.h->aut->new_states(states);
res.info_states.resize(states);
}
if (res.accset < 0)
{
error(@$, "missing 'Acceptance:' header");
res.ignore_acc = true;
}
if (res.unknown_ap_max >= 0 && !res.ignore_more_ap)
{
error(res.unknown_ap_max_location,
"atomic proposition used in Alias without AP declaration");
for (auto& p: res.alias)
p.second = bddtrue;
}
if (res.has_controllable_ap)
{
if (!res.ignore_more_ap && !res.controllable_ap.empty())
{
error(res.controllable_aps_loc,
"controllable-AP without AP declaration");
}
else
{
bdd cont = bddtrue;
unsigned n = res.controllable_ap.size();
unsigned maxap = res.ap.size();
for (unsigned i = 0; i < n; ++i)
{
unsigned c = res.controllable_ap[i];
if (c >= maxap)
{
error(res.controllable_ap_loc[i],
"controllable AP number is larger than"
" the number of APs...");
error(res.ap_loc, "... declared here.");
}
else
{
cont &= bdd_ithvar(res.ap[c]);
}
}
res.aut_or_ks->set_named_prop("synthesis-outputs",
new bdd(cont));
}
}
// Process properties.
{
auto explicit_labels = res.prop_is_true("explicit-labels");
auto implicit_labels = res.prop_is_true("implicit-labels");
if (implicit_labels)
{
if (res.opts.want_kripke)
error(implicit_labels.loc,
"Kripke structure may not use implicit labels");
if (explicit_labels)
{
error(implicit_labels.loc,
"'properties: implicit-labels' is incompatible "
"with...");
error(explicit_labels.loc,
"... 'properties: explicit-labels'.");
}
else
{
res.label_style = Implicit_Labels;
}
}
auto trans_labels = res.prop_is_true("trans-labels");
auto state_labels = res.prop_is_true("state-labels");
if (trans_labels)
{
if (res.opts.want_kripke)
error(trans_labels.loc,
"Kripke structures may not use transition labels");
if (state_labels)
{
error(trans_labels.loc,
"'properties: trans-labels' is incompatible with...");
error(state_labels.loc,
"... 'properties: state-labels'.");
}
else
{
if (res.label_style != Implicit_Labels)
res.label_style = Trans_Labels;
}
}
else if (state_labels)
{
if (res.label_style != Implicit_Labels)
{
res.label_style = State_Labels;
}
else
{
error(state_labels.loc,
"'properties: state-labels' is incompatible with...");
error(implicit_labels.loc,
"... 'properties: implicit-labels'.");
}
}
if (res.opts.want_kripke && res.label_style != State_Labels)
error(@$,
"Kripke structures should use 'properties: state-labels'");
auto state_acc = res.prop_is_true("state-acc");
auto trans_acc = res.prop_is_true("trans-acc");
if (trans_acc)
{
if (state_acc)
{
error(trans_acc.loc,
"'properties: trans-acc' is incompatible with...");
error(state_acc.loc,
"... 'properties: state-acc'.");
}
else
{
res.acc_style = Trans_Acc;
}
}
else if (state_acc)
{
res.acc_style = State_Acc;
}
if (auto univ_branch = res.prop_is_true("univ-branch"))
if (res.opts.want_kripke)
error(univ_branch.loc,
"Kripke structures may not use 'properties: univ-branch'");
}
{
unsigned ss = res.start.size();
auto det = res.prop_is_true("deterministic");
auto no_exist = res.prop_is_false("exist-branch");
if (ss > 1)
{
if (det)
{
error(det.loc,
"deterministic automata should have at most "
"one initial state");
res.universal = spot::trival::maybe();
}
else if (no_exist)
{
error(no_exist.loc,
"universal automata should have at most "
"one initial state");
res.universal = spot::trival::maybe();
}
}
else
{
// Assume the automaton is deterministic until proven
// wrong, or unless we are building a Kripke structure.
if (!res.opts.want_kripke)
{
res.universal = true;
res.existential = true;
}
}
for (auto& ss: res.start)
{
if (ss.second.size() > 1)
{
if (auto no_univ = res.prop_is_false("univ-branch"))
{
error(ss.first,
"conjunct initial state despite...");
error(no_univ.loc, "... property: !univ-branch");
}
else if (v1plus)
if (auto det = res.prop_is_true("deterministic"))
{
error(ss.first,
"conjunct initial state despite...");
error(det.loc, "... property: deterministic");
}
res.existential = false;
}
}
auto complete = res.prop_is_true("complete");
if (ss < 1)
{
if (complete)
{
error(complete.loc,
"complete automata should have at least "
"one initial state");
}
res.complete = false;
}
else
{
// Assume the automaton is complete until proven
// wrong.
if (!res.opts.want_kripke)
res.complete = true;
}
// if ap_count == 0, then a Kripke structure could be
// declared complete, although that probably doesn't
// matter.
if (res.opts.want_kripke && complete && res.ap_count > 0)
error(complete.loc,
"Kripke structure may not be complete");
}
if (res.opts.trust_hoa)
{
auto& a = res.aut_or_ks;
auto& p = res.props;
auto e = p.end();
auto si = p.find("stutter-invariant");
if (si != e)
{
a->prop_stutter_invariant(si->second.val);
auto i = p.find("stutter-sensitive");
if (i != e && si->second.val == i->second.val)
error(i->second.loc,
"automaton cannot be both stutter-invariant"
"and stutter-sensitive");
}
else
{
auto ss = p.find("stutter-sensitive");
if (ss != e)
a->prop_stutter_invariant(!ss->second.val);
}
auto iw = p.find("inherently-weak");
auto vw = p.find("very-weak");
auto w = p.find("weak");
auto t = p.find("terminal");
if (vw != e)
{
a->prop_very_weak(vw->second.val);
if (w != e && !w->second.val && vw->second.val)
{
error(w->second.loc,
"'properties: !weak' contradicts...");
error(vw->second.loc,
"... 'properties: very-weak' given here");
}
if (iw != e && !iw->second.val && vw->second.val)
{
error(iw->second.loc,
"'properties: !inherently-weak' contradicts...");
error(vw->second.loc,
"... 'properties: very-weak' given here");
}
}
if (iw != e)
{
a->prop_inherently_weak(iw->second.val);
if (w != e && !iw->second.val && w->second.val)
{
error(w->second.loc, "'properties: weak' contradicts...");
error(iw->second.loc,
"... 'properties: !inherently-weak' given here");
}
if (t != e && !iw->second.val && t->second.val)
{
error(t->second.loc,
"'properties: terminal' contradicts...");
error(iw->second.loc,
"... 'properties: !inherently-weak' given here");
}
}
if (w != e)
{
a->prop_weak(w->second.val);
if (t != e && !w->second.val && t->second.val)
{
error(t->second.loc,
"'properties: terminal' contradicts...");
error(w->second.loc,
"... 'properties: !weak' given here");
}
}
if (t != e)
a->prop_terminal(t->second.val);
if (a->acc().is_t() || a->acc().is_f())
{
if (w != e && !w->second.val)
error(w->second.loc, "an automaton with this condition"
" is necessarily weak");
if (iw != e && !iw->second.val)
error(iw->second.loc, "an automaton with this condition"
" is necessarily inherently-weak");
if (vw != e && !vw->second.val
&& (res.states == 0 || res.states == 1))
error(vw->second.loc, "an automaton with 0 or 1 state "
"is necessarily very-weak");
}
auto u = p.find("unambiguous");
if (u != e)
{
a->prop_unambiguous(u->second.val);
auto d = p.find("deterministic");
if (d != e && !u->second.val && d->second.val)
{
error(d->second.loc,
"'properties: deterministic' contradicts...");
error(u->second.loc,
"... 'properties: !unambiguous' given here");
}
}
auto sd = p.find("semi-deterministic");
if (sd != e)
{
a->prop_semi_deterministic(sd->second.val);
auto d = p.find("deterministic");
if (d != e && !sd->second.val && d->second.val)
{
error(d->second.loc,
"'properties: deterministic' contradicts...");
error(sd->second.loc,
"... 'properties: !semi-deterministic' given here");
}
}
}
}
version: IDENTIFIER
{
res.format_version = *$1;
res.format_version_loc = @1;
delete $1;
}
format-version: "HOA:" { res.h->loc = @1; } version
controllable-aps: %empty
| controllable-aps INT
{
res.controllable_ap.push_back($2);
res.controllable_ap_loc.push_back(@2);
}
aps: "AP:" INT
{
if (res.ignore_more_ap)
{
error(@1, "ignoring this redeclaration of APs...");
error(res.ap_loc, "... previously declared here.");
}
else
{
res.ap_count = $2;
res.ap.reserve($2);
}
}
ap-names
{
if (!res.ignore_more_ap)
{
res.ap_loc = @1 + @2;
if ((int) res.ap.size() != res.ap_count)
{
std::ostringstream out;
out << "found " << res.ap.size()
<< " atomic propositions instead of the "
<< res.ap_count << " announced";
error(@$, out.str());
}
res.ignore_more_ap = true;
// If we have seen Alias: before AP: we have some variable
// renaming to perform.
if (res.unknown_ap_max >= 0)
{
int apsize = res.ap.size();
if (apsize <= res.unknown_ap_max)
{
error(res.unknown_ap_max_location,
"AP number is larger than the number of APs...");
error(@1, "... declared here");
}
bddPair* pair = bdd_newpair();
int max = std::min(res.unknown_ap_max, apsize - 1);
for (int i = 0; i <= max; ++i)
if (i != res.ap[i])
bdd_setbddpair(pair, i, res.ap[i]);
bdd extra = bddtrue;
for (int i = apsize; i <= res.unknown_ap_max; ++i)
extra &= bdd_ithvar(i);
for (auto& p: res.alias)
p.second = bdd_restrict(bdd_replace(p.second, pair), extra);
bdd_freepair(pair);
}
}
}
header-items: %empty
| header-items header-item
header-item: "States:" INT
{
if (res.states >= 0)
{
error(@$, "redefinition of the number of states...");
error(res.states_loc, "... previously defined here.");
}
else
{
res.states_loc = @$;
}
if (((int) $2) < 0)
{
error(@$, "too many states for this implementation");
YYABORT;
}
res.states = std::max(res.states, (int) $2);
}
| "Start:" init-state-conj-2
{
res.start.emplace_back(@$, *$2);
delete $2;
}
| "Start:" state-num
{
res.start.emplace_back(@$, std::vector<unsigned>{$2});
}
| aps
| "controllable-AP:" controllable-aps
{
res.controllable_aps_loc = @1;
res.has_controllable_ap = true;
}
| "Alias:" ANAME { res.in_alias=true; } label-expr
{
res.in_alias = false;
if (!res.alias.emplace(*$2, bdd_from_int($4)).second)
{
std::ostringstream o;
o << "ignoring redefinition of alias @" << *$2;
error(@$, o.str());
}
else
{
res.alias_order.emplace_back(*$2);
}
delete $2;
bdd_delref($4);
}
| "Acceptance:" INT
{
if (res.ignore_more_acc)
{
error(@1 + @2, "ignoring this redefinition of the "
"acceptance condition...");
error(res.accset_loc, "... previously defined here.");
}
else if ($2 > SPOT_MAX_ACCSETS)
{
error(@1 + @2,
"this implementation cannot support such a large "
"number of acceptance sets");
YYABORT;
}
else
{
res.aut_or_ks->acc().add_sets($2);
res.accset = $2;
res.accset_loc = @1 + @2;
}
}
acceptance-cond
{
res.ignore_more_acc = true;
// Not setting the acceptance in case of error will
// force it to be true.
if (res.opts.want_kripke && (!$4->is_t() || $2 > 0))
error(@2 + @4,
"the acceptance for Kripke structure must be '0 t'");
else
res.aut_or_ks->set_acceptance($2, *$4);
delete $4;
}
| "acc-name:" IDENTIFIER acc-spec
{
delete $2;
}
| "tool:" STRING string_opt
{
delete $2;
delete $3;
}
| "name:" STRING
{
res.aut_or_ks->set_named_prop("automaton-name", $2);
}
| "properties:" properties
| "spot.highlight.edges:"
{ res.highlight_edges = new std::map<unsigned, unsigned>; }
highlight-edges
| "spot.highlight.states:"
{ res.highlight_states = new std::map<unsigned, unsigned>; }
highlight-states
| "spot.state-player:"
{ auto p = new std::vector<bool>;
if (res.states >= 0)
p->reserve(res.states);
res.state_player = p;
} state-player
{
res.state_player_loc = @$;
}
| HEADERNAME header-spec
{
char c = (*$1)[0];
if (c >= 'A' && c <= 'Z')
error(@$, "ignoring unsupported header \"" + *$1 + ":\"\n\t"
"(but the capital indicates information that should not"
" be ignored)");
delete $1;
}
| error
ap-names: %empty
| ap-names ap-name
ap-name: STRING
{
if (!res.ignore_more_ap)
{
auto f = res.env->require(*$1);
int b = 0;
if (f == nullptr)
{
std::ostringstream out;
out << "unknown atomic proposition \"" << *$1 << "\"";
error(@1, out.str());
b = res.aut_or_ks->register_ap("$unknown$");
}
else
{
b = res.aut_or_ks->register_ap(f);
if (!res.ap_set.emplace(b).second)
{
std::ostringstream out;
out << "duplicate atomic proposition \"" << *$1 << "\"";
error(@1, out.str());
}
}
res.ap.push_back(b);
}
delete $1;
}
acc-spec: %empty
| acc-spec BOOLEAN
| acc-spec INT
| acc-spec IDENTIFIER
{
delete $2;
}
properties: %empty
| properties IDENTIFIER
{
bool val = true;
// no-univ-branch was replaced by !univ-branch in HOA 1.1
if (*$2 == "no-univ-branch")
{
*$2 = "univ-branch";
val = false;
}
auto pos = res.props.emplace(*$2, result_::prop_info{@2, val});
if (pos.first->second.val != val)
{
std::ostringstream out(std::ios_base::ate);
error(@2, "'properties: "s + (val ? "" : "!")
+ *$2 + "' contradicts...");
error(pos.first->second.loc,
"... 'properties: "s + (val ? "!" : "") + *$2
+ "' previously given here.");
}
delete $2;
}
| properties '!' IDENTIFIER
{
auto loc = @2 + @3;
auto pos =
res.props.emplace(*$3, result_::prop_info{loc, false});
if (pos.first->second.val)
{
std::ostringstream out(std::ios_base::ate);
error(loc, "'properties: !"s + *$3 + "' contradicts...");
error(pos.first->second.loc, "... 'properties: "s + *$3
+ "' previously given here.");
}
delete $3;
}
highlight-edges: %empty
| highlight-edges INT INT
{
res.highlight_edges->emplace($2, $3);
}
highlight-states: %empty
| highlight-states INT INT
{
res.highlight_states->emplace($2, $3);
}
state-player: %empty
| state-player INT
{
if ($2 != 0 && $2 != 1)
error(@2, "player should be 0 or 1");
res.state_player->emplace_back($2);
}
header-spec: %empty
| header-spec BOOLEAN
| header-spec INT
| header-spec STRING
{
delete $2;
}
| header-spec IDENTIFIER
{
delete $2;
}
state-conj-2: checked-state-num '&' checked-state-num
{
$$ = new std::vector<unsigned>{$1, $3};
}
| state-conj-2 '&' checked-state-num
{
$$ = $1;
$$->emplace_back($3);
}
// Same as state-conj-2 except we cannot check the state numbers
// against a number of states that may not have been declared yet.
init-state-conj-2: state-num '&' state-num
{
$$ = new std::vector<unsigned>{$1, $3};
}
| init-state-conj-2 '&' state-num
{
$$ = $1;
$$->emplace_back($3);
}
label-expr: 't'
{
$$ = bddtrue.id();
}
| 'f'
{
$$ = bddfalse.id();
}
| INT
{
if (res.in_alias && !res.ignore_more_ap)
{
// We are reading Alias: before AP: has been given.
// Use $1 as temporary variable number. We will relabel
// everything once AP: is known.
if (res.unknown_ap_max < (int)$1)
{
res.unknown_ap_max = $1;
res.unknown_ap_max_location = @1;
int missing_vars = 1 + bdd_varnum() - $1;
if (missing_vars > 0)
bdd_extvarnum(missing_vars);
}
$$ = bdd_ithvar($1).id();
}
else if ($1 >= res.ap.size())
{
error(@1, "AP number is larger than the number of APs...");
error(res.ap_loc, "... declared here");
$$ = bddtrue.id();
}
else
{
$$ = bdd_ithvar(res.ap[$1]).id();
bdd_addref($$);
}
}
| ANAME
{
auto i = res.alias.find(*$1);
if (i == res.alias.end())
{
error(@$, "unknown alias @" + *$1);
$$ = 1;
}
else
{
$$ = i->second.id();
bdd_addref($$);
}
delete $1;
}
| '!' label-expr
{
$$ = bdd_not($2);
bdd_delref($2);
bdd_addref($$);
}
| label-expr '&' label-expr
{
$$ = bdd_and($1, $3);
bdd_delref($1);
bdd_delref($3);
bdd_addref($$);
}
| label-expr '|' label-expr
{
$$ = bdd_or($1, $3);
bdd_delref($1);
bdd_delref($3);
bdd_addref($$);
}
| '(' label-expr ')'
{
$$ = $2;
}
acc-set: INT
{
if ((int) $1 >= res.accset)
{
if (!res.ignore_acc)
{
error(@1, "number is larger than the count "
"of acceptance sets...");
error(res.accset_loc, "... declared here.");
}
$$ = -1U;
}
else
{
$$ = $1;
}
}
acceptance-cond: IDENTIFIER '(' acc-set ')'
{
if ($3 != -1U)
{
res.pos_acc_sets |= res.aut_or_ks->acc().mark($3);
if (*$1 == "Inf")
{
$$ = new spot::acc_cond::acc_code
(res.aut_or_ks->acc().inf({$3}));
}
else if (*$1 == "Fin")
{
$$ = new spot::acc_cond::acc_code
(res.aut_or_ks->acc().fin({$3}));
}
else
{
error(@1, "unknown acceptance '"s + *$1
+ "', expected Fin or Inf");
$$ = new spot::acc_cond::acc_code;
}
}
else
{
$$ = new spot::acc_cond::acc_code;
}
delete $1;
}
| IDENTIFIER '(' '!' acc-set ')'
{
if ($4 != -1U)
{
res.neg_acc_sets |= res.aut_or_ks->acc().mark($4);
if (*$1 == "Inf")
$$ = new spot::acc_cond::acc_code
(res.aut_or_ks->acc().inf_neg({$4}));
else
$$ = new spot::acc_cond::acc_code
(res.aut_or_ks->acc().fin_neg({$4}));
}
else
{
$$ = new spot::acc_cond::acc_code;
}
delete $1;
}
| '(' acceptance-cond ')'
{
$$ = $2;
}
| acceptance-cond '&' acceptance-cond
{
*$3 &= std::move(*$1);
$$ = $3;
delete $1;
}
| acceptance-cond '|' acceptance-cond
{
*$3 |= std::move(*$1);
$$ = $3;
delete $1;
}
| 't'
{
$$ = new spot::acc_cond::acc_code;
}
| 'f'
{
{
$$ = new spot::acc_cond::acc_code
(res.aut_or_ks->acc().fin({}));
}
}
body: states
{
for (auto& p: res.start)
for (unsigned s: p.second)
if (s >= res.info_states.size() || !res.info_states[s].declared)
{
error(p.first, "initial state " + std::to_string(s) +
" has no definition");
// Pretend that the state is declared so we do not
// mention it in the next loop.
if (s < res.info_states.size())
res.info_states[s].declared = true;
res.complete = spot::trival::maybe();
}
unsigned n = res.info_states.size();
// States with number above res.states have already caused a
// diagnostic, so let not add another one.
if (res.states >= 0)
n = res.states;
std::vector<unsigned> unused_undeclared;
for (unsigned i = 0; i < n; ++i)
{
auto& p = res.info_states[i];
if (!p.declared)
{
if (p.used)
error(p.used_loc,
"state " + std::to_string(i) + " has no definition");
if (!p.used)
unused_undeclared.push_back(i);
res.complete = false;
}
}
if (!unused_undeclared.empty())
{
std::ostringstream out;
unsigned uus = unused_undeclared.size();
int rangestart = -2;
int rangecur = -2;
const char* sep = uus > 1 ? "states " : "state ";
auto print_range = [&]() {
if (rangecur < 0)
return;
out << sep << rangestart;
if (rangecur != rangestart)
out << '-' << rangecur;
sep = ",";
};
for (unsigned s: unused_undeclared)
{
if ((int)s != rangecur + 1)
{
print_range();
rangestart = s;
}
rangecur = s;
}
print_range();
out << (uus > 1 ? " are" : " is") << " unused and undefined";
error(res.states_loc, out.str());
if (auto p = res.prop_is_true("complete"))
error(p.loc, "automaton is incomplete because it has "
"undefined states");
}
if (res.complete)
if (auto p = res.prop_is_false("complete"))
{
error(@1, "automaton is complete...");
error(p.loc, "... despite 'properties: !complete'");
}
bool det_warned = false;
if (res.universal && res.existential)
if (auto p = res.prop_is_false("deterministic"))
{
error(@1, "automaton is deterministic...");
error(p.loc, "... despite 'properties: !deterministic'");
det_warned = true;
}
static bool tolerant = getenv("SPOT_HOA_TOLERANT");
if (res.universal.is_true() && !det_warned && !tolerant)
if (auto p = res.prop_is_true("exist-branch"))
{
error(@1, "automaton has no existential branching...");
error(p.loc, "... despite 'properties: exist-branch'\n"
"note: If this is an issue you cannot fix, you may disable "
"this diagnostic\n by defining the SPOT_HOA_TOLERANT "
"environment variable.");
det_warned = true;
}
if (res.existential.is_true() && !det_warned && !tolerant)
if (auto p = res.prop_is_true("univ-branch"))
{
error(@1, "automaton is has no universal branching...");
error(p.loc, "... despite 'properties: univ-branch'\n"
"note: If this is an issue you cannot fix, you may disable "
"this diagnostic\n by defining the SPOT_HOA_TOLERANT "
"environment variable.");
det_warned = true;
}
if (res.state_player)
if (unsigned spsz = res.state_player->size(); spsz != n)
{
error(res.state_player_loc,
"ignoring state-player header because it has "s
+ std::to_string(spsz) + " entries while automaton has "
+ std::to_string(n) + " states");
delete res.state_player;
res.state_player = nullptr;
}
}
state-num: INT
{
if (((int) $1) < 0)
{
error(@1, "state number is too large for this implementation");
YYABORT;
}
$$ = $1;
}
checked-state-num: state-num
{
if ((int) $1 >= res.states)
{
if (res.states >= 0)
{
error(@1, "state number is larger than state "
"count...");
error(res.states_loc, "... declared here.");
}
if (res.opts.want_kripke)
{
int missing =
((int) $1) - res.h->ks->num_states() + 1;
if (missing >= 0)
{
res.h->ks->new_states(missing, bddfalse);
res.info_states.resize
(res.info_states.size() + missing);
}
}
else
{
int missing =
((int) $1) - res.h->aut->num_states() + 1;
if (missing >= 0)
{
res.h->aut->new_states(missing);
res.info_states.resize
(res.info_states.size() + missing);
}
}
}
// Remember the first place were a state is the
// destination of a transition.
if (!res.info_states[$1].used)
{
res.info_states[$1].used = true;
res.info_states[$1].used_loc = @1;
}
$$ = $1;
}
states: %empty
| states state
{
if ((res.universal.is_true() || res.complete.is_true()))
{
bdd available = bddtrue;
bool det = true;
for (auto& t: res.h->aut->out(res.cur_state))
{
if (det && !bdd_implies(t.cond, available))
det = false;
available -= t.cond;
}
if (res.universal.is_true() && !det)
{
res.universal = false;
if (auto p = res.prop_is_true("deterministic"))
{
error(@2, "automaton is not deterministic...");
error(p.loc,
"... despite 'properties: deterministic'");
}
else if (auto p = res.prop_is_false("exist-branch"))
{
error(@2, "automaton has existential branching...");
error(p.loc,
"... despite 'properties: !exist-branch'");
}
}
if (res.complete.is_true() && available != bddfalse)
{
res.complete = false;
if (auto p = res.prop_is_true("complete"))
{
error(@2, "automaton is not complete...");
error(p.loc, "... despite 'properties: complete'");
}
}
}
if (res.acc_state &&
!res.opts.want_kripke &&
res.h->aut->get_graph().state_storage(res.cur_state).succ == 0)
res.h->aut->new_edge(res.cur_state, res.cur_state,
bddfalse, res.acc_state);
}
state: state-name labeled-edges
| state-name unlabeled-edges
{
if (!res.has_state_label) // Implicit labels
{
if (res.cur_guard != res.guards.end())
error(@$, "not enough transitions for this state");
if (res.label_style == State_Labels)
{
error(@2, "these transitions have implicit labels but the"
" automaton is...");
error(res.props["state-labels"].loc, "... declared with "
"'properties: state-labels'");
// Do not repeat this message.
res.label_style = Mixed_Labels;
}
res.cur_guard = res.guards.begin();
}
else if (res.opts.want_kripke)
{
res.h->ks->state_from_number(res.cur_state)->cond(res.state_label);
}
}
| error
{
// Assume the worse. This skips the tests about determinism
// we might perform on the state.
res.universal = spot::trival::maybe();
res.existential = spot::trival::maybe();
res.complete = spot::trival::maybe();
// also do not try to preserve any color
res.acc_state = {};
}
state-name: "State:" state-label_opt checked-state-num string_opt state-acc_opt
{
res.cur_state = $3;
if (res.info_states[$3].declared)
{
std::ostringstream o;
o << "redeclaration of state " << $3;
error(@1 + @3, o.str());
// The additional transitions from extra states might
// led us to believe that the automaton is complete
// while it is not if we ignore them.
if (res.complete.is_true())
res.complete = spot::trival::maybe();
}
res.info_states[$3].declared = true;
res.acc_state = $5;
if ($4)
{
if (!res.state_names)
res.state_names =
new std::vector<std::string>(res.states > 0 ?
res.states : 0);
if (res.state_names->size() < $3 + 1)
res.state_names->resize($3 + 1);
(*res.state_names)[$3] = std::move(*$4);
delete $4;
}
if (res.opts.want_kripke && !res.has_state_label)
error(@$, "Kripke structures should have labeled states");
}
label: '[' label-expr ']'
{
res.cur_label = bdd_from_int($2);
bdd_delref($2);
if ($1)
res.fcache[*$1] = res.cur_label;
delete $1;
}
| BDD { res.cur_label = bdd_from_int($1); }
| '[' error ']'
{
error(@$, "ignoring this invalid label");
res.cur_label = bddtrue;
delete $1;
}
state-label_opt: %empty { res.has_state_label = false; }
| label
{
res.has_state_label = true;
res.state_label_loc = @1;
res.state_label = res.cur_label;
if (res.label_style == Trans_Labels
|| res.label_style == Implicit_Labels)
{
error(@$,
"state label used although the automaton was...");
if (res.label_style == Trans_Labels)
error(res.props["trans-labels"].loc,
"... declared with 'properties: trans-labels'"
" here");
else
error(res.props["implicit-labels"].loc,
"... declared with 'properties: implicit-labels'"
" here");
// Do not show this error anymore.
res.label_style = Mixed_Labels;
}
}
trans-label: label
{
if (res.has_state_label)
{
error(@1, "cannot label this edge because...");
error(res.state_label_loc,
"... the state is already labeled.");
res.cur_label = res.state_label;
}
if (res.label_style == State_Labels
|| res.label_style == Implicit_Labels)
{
error(@$, "transition label used although the "
"automaton was...");
if (res.label_style == State_Labels)
error(res.props["state-labels"].loc,
"... declared with 'properties: state-labels' "
"here");
else
error(res.props["implicit-labels"].loc,
"... declared with 'properties: implicit-labels'"
" here");
// Do not show this error anymore.
res.label_style = Mixed_Labels;
}
}
acc-sig: '{' acc-sets '}'
{
$$ = $2;
if (res.ignore_acc && !res.ignore_acc_silent)
{
error(@$, "ignoring acceptance sets because of "
"missing acceptance condition");
// Emit this message only once.
res.ignore_acc_silent = true;
}
}
| '{' error '}'
{
error(@$, "ignoring this invalid acceptance set");
}
acc-sets: %empty
{
$$ = spot::acc_cond::mark_t({});
}
| acc-sets acc-set
{
if (res.ignore_acc || $2 == -1U)
$$ = spot::acc_cond::mark_t({});
else
$$ = $1 | res.aut_or_ks->acc().mark($2);
}
state-acc_opt: %empty
{
$$ = spot::acc_cond::mark_t({});
}
| acc-sig
{
$$ = $1;
if (res.acc_style == Trans_Acc)
{
error(@$, "state-based acceptance used despite...");
error(res.props["trans-acc"].loc,
"... declaration of transition-based acceptance.");
res.acc_style = Mixed_Acc;
}
}
trans-acc_opt: %empty
{
$$ = spot::acc_cond::mark_t({});
}
| acc-sig
{
$$ = $1;
res.trans_acc_seen = true;
if (res.acc_style == State_Acc)
{
error(@$, "trans-based acceptance used despite...");
error(res.props["state-acc"].loc,
"... declaration of state-based acceptance.");
res.acc_style = Mixed_Acc;
}
}
/* block of labeled-edges, with occasional (incorrect) unlabeled edge */
labeled-edges: %empty
| some-labeled-edges
some-labeled-edges: labeled-edge
| some-labeled-edges labeled-edge
| some-labeled-edges incorrectly-unlabeled-edge
incorrectly-unlabeled-edge: checked-state-num trans-acc_opt
{
bdd cond = bddtrue;
if (!res.has_state_label)
error(@$, "missing label for this edge "
"(previous edge is labeled)");
else
cond = res.state_label;
if (cond != bddfalse || !res.opts.drop_false_edges)
{
unsigned e;
if (res.opts.want_kripke)
e = res.h->ks->new_edge(res.cur_state, $1);
else
e = res.h->aut->new_edge(res.cur_state, $1,
cond,
$2 |
res.acc_state);
res.edge_map.push_back(e);
}
}
labeled-edge: trans-label checked-state-num trans-acc_opt
{
unsigned e = 0;
if (res.cur_label != bddfalse
|| !res.opts.drop_false_edges
// As a hack to allow states to be accepting
// even if they do not have transitions, we
// do not ignore false-labeled self-loops if they
// have some colors)
|| ($2 == res.cur_state && !!($3 | res.acc_state)))
{
if (res.opts.want_kripke)
e = res.h->ks->new_edge(res.cur_state, $2);
else
e = res.h->aut->new_edge(res.cur_state, $2,
res.cur_label,
$3 | res.acc_state);
}
res.edge_map.push_back(e);
}
| trans-label state-conj-checked trans-acc_opt
{
unsigned e = 0;
if (res.cur_label != bddfalse || !res.opts.drop_false_edges)
{
assert(!res.opts.want_kripke);
e = res.h->aut->new_univ_edge(res.cur_state,
$2->begin(), $2->end(),
res.cur_label,
$3 | res.acc_state);
}
res.edge_map.push_back(e);
delete $2;
}
state-conj-checked: state-conj-2
{
$$ = $1;
if (auto ub = res.prop_is_false("univ-branch"))
{
error(@1, "universal branch used despite"
" previous declaration...");
error(ub.loc, "... here");
}
res.existential = false;
}
/* Block of unlabeled edge, with occasional (incorrect) labeled
edge. We never have zero unlabeled edges, these are considered as
zero labeled edges. */
unlabeled-edges: unlabeled-edge
| unlabeled-edges unlabeled-edge
| unlabeled-edges incorrectly-labeled-edge
unlabeled-edge: checked-state-num trans-acc_opt
{
bdd cond;
if (res.has_state_label)
{
cond = res.state_label;
}
else
{
if (res.guards.empty())
fill_guards(res);
if (res.cur_guard == res.guards.end())
{
error(@$, "too many transitions for this state, "
"ignoring this one");
cond = bddfalse;
}
else
{
cond = *res.cur_guard++;
}
}
unsigned e = 0;
if (cond != bddfalse || !res.opts.drop_false_edges)
{
if (res.opts.want_kripke)
e = res.h->ks->new_edge(res.cur_state, $1);
else
e = res.h->aut->new_edge(res.cur_state, $1,
cond, $2 | res.acc_state);
}
res.edge_map.push_back(e);
}
| state-conj-checked trans-acc_opt
{
bdd cond;
if (res.has_state_label)
{
cond = res.state_label;
}
else
{
if (res.guards.empty())
fill_guards(res);
if (res.cur_guard == res.guards.end())
{
error(@$, "too many transitions for this state, "
"ignoring this one");
cond = bddfalse;
}
else
{
cond = *res.cur_guard++;
}
}
unsigned e = 0;
if (cond != bddfalse || !res.opts.drop_false_edges)
{
assert(!res.opts.want_kripke);
e = res.h->aut->new_univ_edge(res.cur_state,
$1->begin(), $1->end(),
cond, $2 | res.acc_state);
}
res.edge_map.push_back(e);
delete $1;
}
incorrectly-labeled-edge: trans-label unlabeled-edge
{
error(@1, "ignoring this label, because previous"
" edge has no label");
}
/**********************************************************************/
/* Rules for LTL2DSTAR's format */
/**********************************************************************/
dstar: dstar_type "v2" "explicit" dstar_header "---" dstar_states ENDDSTAR
| dstar_type error ENDDSTAR
{
error(@$, "failed to parse this as an ltl2dstar automaton");
}
dstar_type: "DRA"
{
res.h->type = spot::parsed_aut_type::DRA;
res.plus = 1;
res.minus = 0;
if (res.opts.want_kripke)
{
error(@$,
"cannot read a Kripke structure out of a DSTAR automaton");
YYABORT;
}
}
| "DSA"
{
res.h->type = spot::parsed_aut_type::DSA;
res.plus = 0;
res.minus = 1;
if (res.opts.want_kripke)
{
error(@$,
"cannot read a Kripke structure out of a DSTAR automaton");
YYABORT;
}
}
dstar_header: dstar_sizes
{
if (res.states < 0)
error(@1, "missing state count");
if (!res.ignore_more_acc)
error(@1, "missing acceptance-pair count");
if (res.start.empty())
error(@1, "missing start-state number");
if (!res.ignore_more_ap)
error(@1, "missing atomic propositions definition");
if (res.states > 0)
{
res.h->aut->new_states(res.states);
res.info_states.resize(res.states);
}
res.acc_style = State_Acc;
res.universal = true;
res.existential = true;
res.complete = true;
fill_guards(res);
res.cur_guard = res.guards.end();
}
dstar_sizes: %empty
| dstar_sizes error
| dstar_sizes "Acceptance-Pairs:" INT
{
if (res.ignore_more_acc)
{
error(@1 + @2, "ignoring this redefinition of the "
"acceptance pairs...");
error(res.accset_loc, "... previously defined here.");
}
else{
res.accset = $3;
res.h->aut->set_acceptance(2 * $3,
res.h->type == spot::parsed_aut_type::DRA
? spot::acc_cond::acc_code::rabin($3)
: spot::acc_cond::acc_code::streett($3));
res.accset_loc = @3;
res.ignore_more_acc = true;
}
}
| dstar_sizes "States:" INT
{
if (res.states < 0)
{
res.states = $3;
}
else
{
error(@$, "redeclaration of state count");
if ((unsigned) res.states < $3)
res.states = $3;
}
}
| dstar_sizes "Start:" INT
{
res.start.emplace_back(@3, std::vector<unsigned>{$3});
}
| dstar_sizes aps
dstar_state_id: "State:" INT string_opt
{
if (res.cur_guard != res.guards.end())
error(@1, "not enough transitions for previous state");
if (res.states < 0 || $2 >= (unsigned) res.states)
{
std::ostringstream o;
if (res.states > 0)
{
o << "state numbers should be in the range [0.."
<< res.states - 1 << "]";
}
else
{
o << "no states have been declared";
}
error(@2, o.str());
}
else
{
res.info_states[$2].declared = true;
if ($3)
{
if (!res.state_names)
res.state_names =
new std::vector<std::string>(res.states > 0 ?
res.states : 0);
if (res.state_names->size() < $2 + 1)
res.state_names->resize($2 + 1);
(*res.state_names)[$2] = std::move(*$3);
delete $3;
}
}
res.cur_guard = res.guards.begin();
res.dest_map.clear();
res.cur_state = $2;
}
sign: '+' { $$ = res.plus; }
| '-' { $$ = res.minus; }
// Membership to a pair is represented as (+NUM,-NUM)
dstar_accsigs: %empty
{
$$ = spot::acc_cond::mark_t({});
}
| dstar_accsigs sign INT
{
if (res.states < 0 || res.cur_state >= (unsigned) res.states)
break;
if (res.accset > 0 && $3 < (unsigned) res.accset)
{
$$ = $1;
$$.set($3 * 2 + $2);
}
else
{
std::ostringstream o;
if (res.accset > 0)
{
o << "acceptance pairs should be in the range [0.."
<< res.accset - 1 << "]";
}
else
{
o << "no acceptance pairs have been declared";
}
error(@3, o.str());
}
}
dstar_state_accsig: "Acc-Sig:" dstar_accsigs { $$ = $2; }
dstar_transitions: %empty
| dstar_transitions INT
{
std::pair<map_t::iterator, bool> i =
res.dest_map.emplace($2, *res.cur_guard);
if (!i.second)
i.first->second |= *res.cur_guard;
++res.cur_guard;
}
dstar_states: %empty
| dstar_states error
| dstar_states dstar_state_id dstar_state_accsig dstar_transitions
{
for (auto i: res.dest_map)
res.h->aut->new_edge(res.cur_state, i.first, i.second, $3);
}
/**********************************************************************/
/* Rules for PGSolver games */
/**********************************************************************/
pgamestart: PGAME
{
if (res.opts.want_kripke)
{
error(@$,
"cannot read a Kripke structure out of a PGSolver game.");
YYABORT;
}
}
pgame: pgamestart pgame_nodes ENDPGAME
{
unsigned n = res.accset;
auto p = spot::acc_cond::acc_code::parity_max_odd(n);
res.h->aut->set_acceptance(n, p);
res.acc_style = State_Acc;
// Pretend that we have declared all states.
n = res.h->aut->num_states();
res.info_states.resize(n);
for (auto& p: res.info_states)
p.declared = true;
}
| pgamestart error ENDPGAME
{
error(@$, "failed to parse this as a PGSolver game");
}
pgame_nodes: pgame_node ';'
| pgame_nodes pgame_node ';'
pgame_succs: INT
{ $$ = new std::vector<unsigned>{$1}; }
| pgame_succs ',' INT
{
$$ = $1;
$$->emplace_back($3);
}
pgame_node: INT INT INT pgame_succs string_opt
{
unsigned state = $1;
unsigned owner = $3;
if (owner > 1)
{
error(@3, "node owner should be 0 or 1");
owner = 0;
}
// Create any missing state
unsigned max_state = state;
for (unsigned s: *$4)
max_state = std::max(max_state, s);
unsigned n = res.h->aut->num_states();
if (n <= max_state)
res.h->aut->new_states(max_state + 1 - n);
// assume the source of the first edge is initial
if (res.start.empty())
res.start.emplace_back(@$, std::vector<unsigned>{state});
// Create all edges with priority $2
spot::acc_cond::mark_t m({$2});
for (unsigned s: *$4)
res.h->aut->new_edge(state, s, bddtrue, m);
res.accset = std::max(res.accset, 1 + (int) $2);
n = res.h->aut->num_states();
if (!res.state_player)
res.state_player = new std::vector<bool>(n);
else if (res.state_player->size() < n)
res.state_player->resize(n);
(*res.state_player)[state] = owner;
if (std::string* name = $5)
{
if (!res.state_names)
res.state_names = new std::vector<std::string>(n);
else if (res.state_names->size() < n)
res.state_names->resize(n);
(*res.state_names)[state] = std::move(*name);
delete name;
}
}
/**********************************************************************/
/* Rules for neverclaims */
/**********************************************************************/
never: "never"
{
if (res.opts.want_kripke)
{
error(@$, "cannot read a Kripke structure out of a never claim.");
YYABORT;
}
res.namer = res.h->aut->create_namer<std::string>();
res.h->aut->set_buchi();
res.acc_style = State_Acc;
res.pos_acc_sets = res.h->aut->acc().all_sets();
}
'{' nc-states '}'
{
// Add an accept_all state if needed.
if (res.accept_all_needed && !res.accept_all_seen)
{
unsigned n = res.namer->new_state("accept_all");
res.h->aut->new_acc_edge(n, n, bddtrue);
}
// If we aliased existing state, we have some unreachable
// states to remove.
if (res.aliased_states)
res.h->aut->purge_unreachable_states();
res.info_states.resize(res.h->aut->num_states());
// Pretend that we have declared all states.
for (auto& p: res.info_states)
p.declared = true;
res.h->aut->register_aps_from_dict();
}
nc-states: %empty
| nc-state
| nc-states ';' nc-state
| nc-states ';'
nc-one-ident: IDENTIFIER ':'
{
auto r = res.labels.insert(std::make_pair(*$1, @1));
if (!r.second)
{
error(@1, "redefinition of "s + *$1 + "...");
error(r.first->second, "... "s + *$1 + " previously defined here");
}
$$ = $1;
}
nc-ident-list: nc-one-ident
{
unsigned n = res.namer->new_state(*$1);
if (res.start.empty())
{
// The first state is initial.
res.start.emplace_back(@$, std::vector<unsigned>{n});
}
$$ = $1;
}
| nc-ident-list nc-one-ident
{
res.aliased_states |=
res.namer->alias_state(res.namer->get_state(*$1), *$2);
// Keep any identifier that starts with accept.
if (strncmp("accept", $1->c_str(), 6))
{
delete $1;
$$ = $2;
}
else
{
delete $2;
$$ = $1;
}
}
nc-transition-block:
"if" nc-transitions "fi"
{
$$ = $2;
}
| "do" nc-transitions "od"
{
$$ = $2;
}
nc-state:
nc-ident-list "skip"
{
if (*$1 == "accept_all")
res.accept_all_seen = true;
auto acc = !strncmp("accept", $1->c_str(), 6) ?
res.h->aut->acc().all_sets() : spot::acc_cond::mark_t({});
res.namer->new_edge(*$1, *$1, bddtrue, acc);
delete $1;
}
| nc-ident-list { delete $1; }
| nc-ident-list "false" { delete $1; }
| nc-ident-list nc-transition-block
{
auto acc = !strncmp("accept", $1->c_str(), 6) ?
res.h->aut->acc().all_sets() : spot::acc_cond::mark_t({});
for (auto& p: *$2)
{
bdd c = bdd_from_int(p.first);
bdd_delref(p.first);
res.namer->new_edge(*$1, *p.second, c, acc);
delete p.second;
}
delete $1;
delete $2;
}
nc-transitions:
%empty { $$ = new std::list<pair>; }
| nc-transitions nc-transition
{
if ($2)
{
$1->push_back(*$2);
delete $2;
}
$$ = $1;
}
nc-formula-or-ident: FORMULA | IDENTIFIER
nc-formula: nc-formula-or-ident
{
auto i = res.fcache.find(*$1);
if (i == res.fcache.end())
{
auto pf = spot::parse_infix_boolean(*$1, *res.env, debug_level(),
true);
for (auto& j: pf.errors)
{
// Adjust the diagnostic to the current position.
spot::location here = @1;
here.end.line = here.begin.line + j.first.end.line - 1;
here.end.column = here.begin.column + j.first.end.column - 1;
here.begin.line += j.first.begin.line - 1;
here.begin.column += j.first.begin.column - 1;
res.h->errors.emplace_back(here, j.second);
}
bdd cond = bddfalse;
if (pf.f)
cond = spot::formula_to_bdd(pf.f,
res.h->aut->get_dict(), res.h->aut);
$$ = (res.fcache[*$1] = cond).id();
}
else
{
$$ = i->second.id();
}
bdd_addref($$);
delete $1;
}
| "false"
{
$$ = 0;
}
nc-opt-dest:
%empty
{
$$ = nullptr;
}
| "->" "goto" IDENTIFIER
{
$$ = $3;
}
| "->" "assert" FORMULA
{
delete $3;
$$ = new std::string("accept_all");
res.accept_all_needed = true;
}
nc-src-dest: nc-formula nc-opt-dest
{
// If there is no destination, do ignore the transition.
// This happens for instance with
// if
// :: false
// fi
if (!$2)
{
$$ = nullptr;
}
else
{
$$ = new pair($1, $2);
res.namer->new_state(*$2);
}
}
nc-transition:
':' ':' "atomic" '{' nc-src-dest '}'
{
$$ = $5;
}
| ':' ':' nc-src-dest
{
$$ = $3;
}
/**********************************************************************/
/* Rules for LBTT */
/**********************************************************************/
lbtt: lbtt-header lbtt-body ENDAUT
{
auto& acc = res.h->aut->acc();
unsigned num = acc.num_sets();
res.h->aut->set_generalized_buchi(num);
res.pos_acc_sets = acc.all_sets();
assert(!res.states_map.empty());
auto n = res.states_map.size();
if (n != (unsigned) res.states)
{
std::ostringstream err;
err << res.states << " states have been declared, but "
<< n << " different state numbers have been used";
error(@$, err.str());
}
if (res.states_map.rbegin()->first > (unsigned) res.states)
{
// We have seen numbers larger that the total number of
// states in the automaton. Usually this happens when the
// states are numbered from 1 instead of 0, but the LBTT
// documentation actually allow any number to be used.
// What we have done is to map all input state numbers 0
// <= .. < n to the digraph states with the same number,
// and any time we saw a number larger than n, we mapped
// it to a new state. The correspondence is given by
// res.states_map. Now we just need to remove the useless
// states we allocated.
std::vector<unsigned> rename(res.h->aut->num_states(), -1U);
unsigned s = 0;
for (auto& i: res.states_map)
rename[i.second] = s++;
assert(s == (unsigned) res.states);
for (auto& i: res.start)
i.second.front() = rename[i.second.front()];
res.h->aut->get_graph().defrag_states(rename, s);
}
res.info_states.resize(res.h->aut->num_states());
for (auto& s: res.info_states)
s.declared = true;
res.h->aut->register_aps_from_dict();
}
| lbtt-header-states LBTT_EMPTY
{
res.h->aut->set_generalized_buchi($2);
res.pos_acc_sets = res.h->aut->acc().all_sets();
}
lbtt-header-states: LBTT
{
if (res.opts.want_kripke)
{
error(@$,
"cannot read a Kripke structure out of "
"an LBTT automaton");
YYABORT;
}
res.states = $1;
res.states_loc = @1;
res.h->aut->new_states($1);
}
lbtt-header: lbtt-header-states INT_S
{
res.acc_mapper = new spot::acc_mapper_int(res.h->aut, $2);
res.acc_style = State_Acc;
}
| lbtt-header-states INT
{
res.acc_mapper = new spot::acc_mapper_int(res.h->aut, $2);
res.trans_acc_seen = true;
}
lbtt-body: lbtt-states
lbtt-states: %empty
| lbtt-states lbtt-state lbtt-transitions
lbtt-state: STATE_NUM INT lbtt-acc
{
if ($1 >= (unsigned) res.states)
{
auto p = res.states_map.emplace($1, 0);
if (p.second)
p.first->second = res.h->aut->new_state();
res.cur_state = p.first->second;
}
else
{
res.states_map.emplace($1, $1);
res.cur_state = $1;
}
if ($2)
res.start.emplace_back(@1 + @2,
std::vector<unsigned>{res.cur_state});
res.acc_state = $3;
}
lbtt-acc: %empty { $$ = spot::acc_cond::mark_t({}); }
| lbtt-acc ACC
{
$$ = $1;
auto p = res.acc_mapper->lookup($2);
if (p.first)
$$ |= p.second;
else
error(@2, "more acceptance sets used than declared");
}
lbtt-guard: STRING
{
auto pf = spot::parse_prefix_ltl(*$1, *res.env);
if (!pf.f || !pf.errors.empty())
{
std::string s = "failed to parse guard: ";
s += *$1;
error(@$, s);
}
if (!pf.errors.empty())
for (auto& j: pf.errors)
{
// Adjust the diagnostic to the current position.
spot::location here = @1;
here.end.line = here.begin.line + j.first.end.line - 1;
here.end.column = here.begin.column + j.first.end.column - 1;
here.begin.line += j.first.begin.line - 1;
here.begin.column += j.first.begin.column - 1;
res.h->errors.emplace_back(here, j.second);
}
if (!pf.f)
{
res.cur_label = bddtrue;
}
else
{
if (!pf.f.is_boolean())
{
error(@$,
"non-Boolean transition label (replaced by true)");
res.cur_label = bddtrue;
}
else
{
res.cur_label =
formula_to_bdd(pf.f, res.h->aut->get_dict(), res.h->aut);
}
}
delete $1;
}
lbtt-transitions: %empty
| lbtt-transitions DEST_NUM lbtt-acc lbtt-guard
{
unsigned dst = $2;
if (dst >= (unsigned) res.states)
{
auto p = res.states_map.emplace(dst, 0);
if (p.second)
p.first->second = res.h->aut->new_state();
dst = p.first->second;
}
else
{
res.states_map.emplace(dst, dst);
}
res.h->aut->new_edge(res.cur_state, dst,
res.cur_label,
res.acc_state | $3);
}
%%
static void fill_guards(result_& r)
{
unsigned nap = r.ap.size();
int* vars = new int[nap];
for (unsigned i = 0; i < nap; ++i)
vars[i] = r.ap[nap - 1 - i];
// build the 2^nap possible guards
r.guards.reserve(1U << nap);
for (size_t i = 0; i < (1U << nap); ++i)
r.guards.push_back(bdd_ibuildcube(i, nap, vars));
r.cur_guard = r.guards.begin();
delete[] vars;
}
void
hoayy::parser::error(const location_type& location,
const std::string& message)
{
res.h->errors.emplace_back(location, message);
}
static spot::acc_cond::acc_code
fix_acceptance_aux(spot::acc_cond& acc,
spot::acc_cond::acc_code in, unsigned pos,
spot::acc_cond::mark_t onlyneg,
spot::acc_cond::mark_t both,
unsigned base)
{
auto& w = in[pos];
switch (w.sub.op)
{
case spot::acc_cond::acc_op::And:
{
unsigned sub = pos - w.sub.size;
--pos;
auto c = fix_acceptance_aux(acc, in, pos, onlyneg, both, base);
pos -= in[pos].sub.size;
while (sub < pos)
{
--pos;
c &= fix_acceptance_aux(acc, in, pos, onlyneg, both, base);
pos -= in[pos].sub.size;
}
return c;
}
case spot::acc_cond::acc_op::Or:
{
unsigned sub = pos - w.sub.size;
--pos;
auto c = fix_acceptance_aux(acc, in, pos, onlyneg, both, base);
pos -= in[pos].sub.size;
while (sub < pos)
{
--pos;
c |= fix_acceptance_aux(acc, in, pos, onlyneg, both, base);
pos -= in[pos].sub.size;
}
return c;
}
case spot::acc_cond::acc_op::Inf:
return acc.inf(in[pos - 1].mark);
case spot::acc_cond::acc_op::Fin:
return acc.fin(in[pos - 1].mark);
case spot::acc_cond::acc_op::FinNeg:
{
auto m = in[pos - 1].mark;
auto c = acc.fin(onlyneg & m);
spot::acc_cond::mark_t tmp = {};
for (auto i: both.sets())
{
if (m.has(i))
tmp.set(base);
++base;
}
if (tmp)
c |= acc.fin(tmp);
return c;
}
case spot::acc_cond::acc_op::InfNeg:
{
auto m = in[pos - 1].mark;
auto c = acc.inf(onlyneg & m);
spot::acc_cond::mark_t tmp = {};
for (auto i: both.sets())
{
if (m.has(i))
tmp.set(base);
++base;
}
if (tmp)
c &= acc.inf(tmp);
return c;
}
}
SPOT_UNREACHABLE();
return {};
}
static void fix_acceptance(result_& r)
{
if (r.opts.want_kripke)
return;
auto& acc = r.h->aut->acc();
// If a set x appears only as Inf(!x), we can complement it so that
// we work with Inf(x) instead.
auto onlyneg = r.neg_acc_sets - r.pos_acc_sets;
if (onlyneg)
{
for (auto& t: r.h->aut->edges())
t.acc ^= onlyneg;
}
// However if set x is used elsewhere, for instance in
// Inf(!x) & Inf(x)
// complementing x would be wrong. We need to create a
// new set, y, that is the complement of x, and rewrite
// this as Inf(y) & Inf(x).
auto both = r.neg_acc_sets & r.pos_acc_sets;
unsigned base = 0;
if (both)
{
base = acc.add_sets(both.count());
for (auto& t: r.h->aut->edges())
{
unsigned i = 0;
if ((t.acc & both) != both)
for (unsigned v : both.sets())
{
if (!t.acc.has(v))
t.acc |= acc.mark(base + i);
i++;
}
}
}
if (onlyneg || both)
{
auto& acc = r.h->aut->acc();
auto code = acc.get_acceptance();
r.h->aut->set_acceptance(acc.num_sets(),
fix_acceptance_aux(acc, code, code.size() - 1,
onlyneg, both, base));
}
}
// Spot only supports a single initial state.
//
// If the input file does not declare any initial state (this is valid
// in the HOA format) add one nonetheless.
//
// If the input file has multiple initial states we have to merge
// them.
//
// 1) In the non-alternating case, this is as simple as making a new
// initial state using the union of all the outgoing transitions of
// the declared initial states. Note that if one of the original
// initial state has no incoming transition, then we can use it as
// initial state, avoiding the creation of a new state.
//
// 2) In the alternating case, the input may have several initial
// states that are conjuncts. We have to reduce the conjuncts to a
// single state first.
static void fix_initial_state(result_& r)
{
std::vector<std::vector<unsigned>> start;
start.reserve(r.start.size());
unsigned ssz = r.info_states.size();
for (auto& p : r.start)
{
std::vector<unsigned> v;
v.reserve(p.second.size());
for (unsigned s: p.second)
// Ignore initial states without declaration
// (They have been diagnosed already.)
if (s < ssz && r.info_states[s].declared)
v.emplace_back(s);
if (!v.empty())
start.push_back(v);
}
// If no initial state has been declared, add one.
if (start.empty())
{
if (r.opts.want_kripke)
r.h->ks->set_init_state(r.h->ks->new_state(bddfalse));
else
r.h->aut->set_init_state(r.h->aut->new_state());
return;
}
// Remove any duplicate initial state.
std::sort(start.begin(), start.end());
auto res = std::unique(start.begin(), start.end());
start.resize(std::distance(start.begin(), res));
assert(start.size() >= 1);
if (start.size() == 1)
{
if (r.opts.want_kripke)
r.h->ks->set_init_state(start.front().front());
else
r.h->aut->set_univ_init_state(start.front().begin(),
start.front().end());
}
else
{
if (r.opts.want_kripke)
{
r.h->errors.emplace_front(r.start.front().first,
"Kripke structure only support "
"a single initial state");
return;
}
auto& aut = r.h->aut;
// Fiddling with initial state may turn an incomplete automaton
// into a complete one.
if (aut->prop_complete().is_false())
aut->prop_complete(spot::trival::maybe());
// Multiple initial states. We might need to add a fake one,
// unless one of the actual initial state has no incoming edge.
std::vector<unsigned char> has_incoming(aut->num_states(), 0);
for (auto& t: aut->edges())
for (unsigned ud: aut->univ_dests(t))
has_incoming[ud] = 1;
bool found = false;
unsigned init = 0;
bool init_alternation = false;
for (auto& pp: start)
if (pp.size() == 1)
{
unsigned p = pp.front();
if (!has_incoming[p])
{
init = p;
found = true;
}
}
else
{
init_alternation = true;
break;
}
if (!found || init_alternation)
// We do need a fake initial state
init = aut->new_state();
aut->set_init_state(init);
// The non-alternating case is the easiest, we simply declare
// the outgoing transitions of all "original initial states"
// into the only one initial state.
if (!init_alternation)
{
for (auto& pp: start)
{
unsigned p = pp.front();
if (p == init)
continue;
if (!has_incoming[p])
{
// If p has no incoming edge, we can simply take
// out its outgoing edges and "re-source" them on init.
// This will avoid creating new edges.
for (auto& t: aut->out(p))
t.src = init;
auto& gr = aut->get_graph();
auto& ps = gr.state_storage(p);
auto& is = gr.state_storage(init);
gr.edge_storage(is.succ_tail).next_succ = ps.succ;
is.succ_tail = ps.succ_tail;
ps.succ = ps.succ_tail = 0;
// we just created a state without successors
aut->prop_complete(false);
}
else
{
// duplicate all edges
for (auto& t: aut->out(p))
aut->new_edge(init, t.dst, t.cond);
}
}
}
else
{
// In the alternating case, we merge outgoing transition of
// the universal destination of conjunct initial states.
// (Note that this loop would work for the non-alternating
// case too, but it is more expansive, so we avoid it if we
// can.)
spot::outedge_combiner combiner(aut);
bdd comb_or = bddfalse;
for (auto& pp: start)
{
bdd comb_and = bddtrue;
for (unsigned d: pp)
comb_and &= combiner(d);
comb_or |= comb_and;
}
combiner.new_dests(init, comb_or);
}
// Merging two states may break state-based acceptance
// make sure all outgoing edges have the same color.
if (aut->prop_state_acc().is_true())
{
bool first = true;
spot::acc_cond::mark_t prev;
for (auto& e: aut->out(init))
if (first)
{
first = false;
prev = e.acc;
}
else if (e.acc != prev)
{
e.acc = prev;
}
}
}
}
static void fix_properties(result_& r)
{
r.aut_or_ks->prop_universal(r.universal);
r.aut_or_ks->prop_complete(r.complete);
if (r.acc_style == State_Acc ||
(r.acc_style == Mixed_Acc && !r.trans_acc_seen))
r.aut_or_ks->prop_state_acc(true);
if (r.aut_or_ks->acc().is_t() || r.aut_or_ks->acc().is_f())
{
r.aut_or_ks->prop_weak(true);
unsigned ns;
if (r.opts.want_kripke)
ns = r.h->ks->num_states();
else
ns = r.h->aut->num_states();
if (ns == 0 || ns == 1)
r.aut_or_ks->prop_very_weak(true);
}
}
static void check_version(const result_& r)
{
if (r.h->type != spot::parsed_aut_type::HOA)
return;
auto& v = r.format_version;
if (v.size() < 2 || v[0] != 'v' || v[1] < '1' || v[1] > '9')
{
r.h->errors.emplace_front(r.format_version_loc, "unknown HOA version");
return;
}
const char* beg = &v[1];
char* end = nullptr;
long int vers = strtol(beg, &end, 10);
if (vers != 1)
{
r.h->errors.emplace_front(r.format_version_loc,
"unsupported HOA version");
return;
}
constexpr const char supported[] = "1";
if (strverscmp(supported, beg) < 0 && !r.h->errors.empty())
{
std::ostringstream s;
s << "we can read HOA v" << supported
<< " but this file uses " << v << "; this might "
"cause the following errors";
r.h->errors.emplace_front(r.format_version_loc, s.str());
return;
}
}
namespace spot
{
automaton_stream_parser::automaton_stream_parser(const std::string& name,
automaton_parser_options opt)
: filename_(name), opts_(opt)
{
try
{
if (hoayyopen(name, &scanner_))
throw std::runtime_error("Cannot open file "s + name);
}
catch (...)
{
hoayyclose(scanner_);
throw;
}
}
automaton_stream_parser::automaton_stream_parser(int fd,
const std::string& name,
automaton_parser_options opt)
: filename_(name), opts_(opt)
{
try
{
if (hoayyopen(fd, &scanner_))
throw std::runtime_error("Cannot open file "s + name);
}
catch (...)
{
hoayyclose(scanner_);
throw;
}
}
automaton_stream_parser::automaton_stream_parser(const char* data,
const std::string& filename,
automaton_parser_options opt)
: filename_(filename), opts_(opt)
{
try
{
hoayystring(data, &scanner_);
}
catch (...)
{
hoayyclose(scanner_);
throw;
}
}
automaton_stream_parser::~automaton_stream_parser()
{
hoayyclose(scanner_);
}
static void raise_parse_error(const parsed_aut_ptr& pa)
{
if (pa->aborted)
pa->errors.emplace_back(pa->loc, "parsing aborted");
if (!pa->errors.empty())
{
std::ostringstream s;
if (pa->format_errors(s))
throw parse_error(s.str());
}
// It is possible that pa->aut == nullptr if we reach the end of a
// stream. It is not necessarily an error.
}
parsed_aut_ptr
automaton_stream_parser::parse(const bdd_dict_ptr& dict,
environment& env)
{
restart:
result_ r;
r.opts = opts_;
r.h = std::make_shared<spot::parsed_aut>(filename_);
if (opts_.want_kripke)
r.aut_or_ks = r.h->ks = make_kripke_graph(dict);
else
r.aut_or_ks = r.h->aut = make_twa_graph(dict);
r.env = &env;
hoayy::parser parser(scanner_, r, last_loc);
static bool env_debug = !!getenv("SPOT_DEBUG_PARSER");
parser.set_debug_level(opts_.debug || env_debug);
hoayyreset(scanner_);
try
{
if (parser.parse())
{
r.h->aut = nullptr;
r.h->ks = nullptr;
r.aut_or_ks = nullptr;
}
}
catch (const spot::hoa_abort& e)
{
r.h->aborted = true;
// Bison 3.0.2 lacks a += operator for locations.
r.h->loc = r.h->loc + e.pos;
}
check_version(r);
last_loc = r.h->loc;
last_loc.step();
filename_ = r.h->filename; // in case it was changed
if (r.h->aborted)
{
if (opts_.ignore_abort)
goto restart;
return r.h;
}
if (opts_.raise_errors)
raise_parse_error(r.h);
if (!r.aut_or_ks)
return r.h;
if (r.state_names)
r.aut_or_ks->set_named_prop("state-names", r.state_names);
if (r.highlight_edges)
{
// Update the highlight_edges map to deal with removed/added
// edges.
unsigned ems = r.edge_map.size();
std::map<unsigned, unsigned> remap;
for (auto [edgnum, color]: *r.highlight_edges)
/* edge numbers outside of the actual number of edges read are
not expected, but we can't trust input data */
if (SPOT_LIKELY(edgnum > 0 && edgnum <= ems))
if (unsigned newnum = r.edge_map[edgnum - 1]; newnum > 0)
remap[newnum] = color;
std::swap(remap, *r.highlight_edges);
r.aut_or_ks->set_named_prop("highlight-edges", r.highlight_edges);
}
if (r.highlight_states)
r.aut_or_ks->set_named_prop("highlight-states", r.highlight_states);
if (r.state_player)
r.aut_or_ks->set_named_prop("state-player", r.state_player);
if (!r.alias_order.empty())
{
auto* p = new std::vector<std::pair<std::string, bdd>>();
p->reserve(r.alias_order.size());
auto end = r.alias_order.rend();
for (auto i = r.alias_order.rbegin(); i != end; ++i)
p->emplace_back(*r.alias.find(*i));
r.aut_or_ks->set_named_prop("aliases", p);
}
fix_acceptance(r);
fix_properties(r); // before fix_initial_state
fix_initial_state(r);
if (r.h->aut && !r.h->aut->is_existential())
r.h->aut->merge_univ_dests();
return r.h;
}
parsed_aut_ptr
parse_aut(const std::string& filename, const bdd_dict_ptr& dict,
environment& env, automaton_parser_options opts)
{
auto localopts = opts;
localopts.raise_errors = false;
parsed_aut_ptr pa;
try
{
automaton_stream_parser p(filename, localopts);
pa = p.parse(dict, env);
}
catch (const std::runtime_error& e)
{
if (opts.raise_errors)
throw;
parsed_aut_ptr pa = std::make_shared<spot::parsed_aut>(filename);
pa->errors.emplace_back(spot::location(), e.what());
return pa;
}
if (!pa->aut && !pa->ks && pa->errors.empty())
pa->errors.emplace_back(pa->loc, "no automaton read (empty input?)");
if (opts.raise_errors)
raise_parse_error(pa);
return pa;
}
}
// Local Variables:
// mode: c++
// End: