1 /******************************************************************************
2 * Top contributors (to current version):
3 * Andrew Reynolds, Andres Noetzli, Morgan Deters
5 * This file is part of the cvc5 project.
7 * Copyright (c) 2009-2021 by the authors listed in the file AUTHORS
8 * in the top-level source directory and their institutional affiliations.
9 * All rights reserved. See the file COPYING in the top-level source
10 * directory for licensing information.
11 * ****************************************************************************
13 * Implementation of term registry for the theory of strings.
16 #include "theory/strings/term_registry.h"
18 #include "expr/attribute.h"
19 #include "options/smt_options.h"
20 #include "options/strings_options.h"
21 #include "smt/logic_exception.h"
22 #include "theory/rewriter.h"
23 #include "theory/strings/inference_manager.h"
24 #include "theory/strings/theory_strings_utils.h"
25 #include "theory/strings/word.h"
26 #include "util/rational.h"
27 #include "util/string.h"
30 using namespace cvc5::context
;
31 using namespace cvc5::kind
;
37 struct StringsProxyVarAttributeId
40 typedef expr::Attribute
<StringsProxyVarAttributeId
, bool>
41 StringsProxyVarAttribute
;
43 TermRegistry::TermRegistry(SolverState
& s
,
44 SequencesStatistics
& statistics
,
45 ProofNodeManager
* pnm
)
48 d_statistics(statistics
),
50 d_functionsTerms(s
.getSatContext()),
51 d_inputVars(s
.getUserContext()),
52 d_preregisteredTerms(s
.getSatContext()),
53 d_registeredTerms(s
.getUserContext()),
54 d_registeredTypes(s
.getUserContext()),
55 d_proxyVar(s
.getUserContext()),
56 d_lengthLemmaTermsCache(s
.getUserContext()),
57 d_epg(pnm
? new EagerProofGenerator(
60 "strings::TermRegistry::EagerProofGenerator")
63 NodeManager
* nm
= NodeManager::currentNM();
64 d_zero
= nm
->mkConst(Rational(0));
65 d_one
= nm
->mkConst(Rational(1));
66 d_negOne
= NodeManager::currentNM()->mkConst(Rational(-1));
67 d_cardSize
= utils::getAlphabetCardinality();
70 TermRegistry::~TermRegistry() {}
72 void TermRegistry::finishInit(InferenceManager
* im
) { d_im
= im
; }
74 Node
TermRegistry::eagerReduce(Node t
, SkolemCache
* sc
)
76 NodeManager
* nm
= NodeManager::currentNM();
78 Kind tk
= t
.getKind();
79 if (tk
== STRING_TO_CODE
)
81 // ite( str.len(s)==1, 0 <= str.code(s) < |A|, str.code(s)=-1 )
82 Node code_len
= utils::mkNLength(t
[0]).eqNode(nm
->mkConst(Rational(1)));
83 Node code_eq_neg1
= t
.eqNode(nm
->mkConst(Rational(-1)));
84 Node code_range
= nm
->mkNode(
86 nm
->mkNode(GEQ
, t
, nm
->mkConst(Rational(0))),
88 LT
, t
, nm
->mkConst(Rational(utils::getAlphabetCardinality()))));
89 lemma
= nm
->mkNode(ITE
, code_len
, code_range
, code_eq_neg1
);
91 else if (tk
== STRING_STRIDOF
|| tk
== STRING_INDEXOF_RE
)
94 // (or (= (f x y n) (- 1)) (>= (f x y n) n))
95 // (<= (f x y n) (str.len x)))
97 // where f in { str.indexof, str.indexof_re }
98 Node l
= nm
->mkNode(STRING_LENGTH
, t
[0]);
102 OR
, nm
->mkConst(Rational(-1)).eqNode(t
), nm
->mkNode(GEQ
, t
, t
[2])),
103 nm
->mkNode(LEQ
, t
, l
));
105 else if (tk
== STRING_STOI
)
107 // (>= (str.to_int x) (- 1))
108 lemma
= nm
->mkNode(GEQ
, t
, nm
->mkConst(Rational(-1)));
110 else if (tk
== STRING_STRCTN
)
112 // ite( (str.contains s r), (= s (str.++ sk1 r sk2)), (not (= s r)))
114 sc
->mkSkolemCached(t
[0], t
[1], SkolemCache::SK_FIRST_CTN_PRE
, "sc1");
116 sc
->mkSkolemCached(t
[0], t
[1], SkolemCache::SK_FIRST_CTN_POST
, "sc2");
117 lemma
= t
[0].eqNode(utils::mkNConcat(sk1
, t
[1], sk2
));
118 lemma
= nm
->mkNode(ITE
, t
, lemma
, t
[0].eqNode(t
[1]).notNode());
123 Node
TermRegistry::lengthPositive(Node t
)
125 NodeManager
* nm
= NodeManager::currentNM();
126 Node zero
= nm
->mkConst(Rational(0));
127 Node emp
= Word::mkEmptyWord(t
.getType());
128 Node tlen
= nm
->mkNode(STRING_LENGTH
, t
);
129 Node tlenEqZero
= tlen
.eqNode(zero
);
130 Node tEqEmp
= t
.eqNode(emp
);
131 Node caseEmpty
= nm
->mkNode(AND
, tlenEqZero
, tEqEmp
);
132 Node caseNEmpty
= nm
->mkNode(GT
, tlen
, zero
);
133 // (or (and (= (str.len t) 0) (= t "")) (> (str.len t) 0))
134 return nm
->mkNode(OR
, caseEmpty
, caseNEmpty
);
137 void TermRegistry::preRegisterTerm(TNode n
)
139 if (d_preregisteredTerms
.find(n
) != d_preregisteredTerms
.end())
143 eq::EqualityEngine
* ee
= d_state
.getEqualityEngine();
144 d_preregisteredTerms
.insert(n
);
145 Trace("strings-preregister")
146 << "TheoryString::preregister : " << n
<< std::endl
;
147 // check for logic exceptions
148 Kind k
= n
.getKind();
149 if (!options::stringExp())
151 if (k
== STRING_STRIDOF
|| k
== STRING_INDEXOF_RE
|| k
== STRING_ITOS
152 || k
== STRING_STOI
|| k
== STRING_STRREPL
|| k
== STRING_SUBSTR
153 || k
== STRING_STRREPLALL
|| k
== SEQ_NTH
|| k
== STRING_REPLACE_RE
154 || k
== STRING_REPLACE_RE_ALL
|| k
== STRING_STRCTN
|| k
== STRING_LEQ
155 || k
== STRING_TOLOWER
|| k
== STRING_TOUPPER
|| k
== STRING_REV
156 || k
== STRING_UPDATE
)
158 std::stringstream ss
;
159 ss
<< "Term of kind " << k
160 << " not supported in default mode, try --strings-exp";
161 throw LogicException(ss
.str());
166 if (n
[0].getType().isRegExp())
168 std::stringstream ss
;
169 ss
<< "Equality between regular expressions is not supported";
170 throw LogicException(ss
.str());
172 ee
->addTriggerPredicate(n
);
175 else if (k
== STRING_IN_REGEXP
)
177 d_im
->requirePhase(n
, true);
178 ee
->addTriggerPredicate(n
);
183 else if (k
== STRING_TO_CODE
)
187 else if (k
== REGEXP_RANGE
)
189 for (const Node
& nc
: n
)
193 throw LogicException(
194 "expecting a constant string term in regexp range");
196 if (nc
.getConst
<String
>().size() != 1)
198 throw LogicException(
199 "expecting a single constant string term in regexp range");
204 TypeNode tn
= n
.getType();
205 if (tn
.isRegExp() && n
.isVar())
207 std::stringstream ss
;
208 ss
<< "Regular expression variables are not supported.";
209 throw LogicException(ss
.str());
211 if (tn
.isString()) // string-only
213 // all characters of constants should fall in the alphabet
216 std::vector
<unsigned> vec
= n
.getConst
<String
>().getVec();
217 for (unsigned u
: vec
)
221 std::stringstream ss
;
222 ss
<< "Characters in string \"" << n
223 << "\" are outside of the given alphabet.";
224 throw LogicException(ss
.str());
230 else if (tn
.isBoolean())
232 // All kinds that we do congruence over that may return a Boolean go here
233 if (k
==STRING_STRCTN
|| k
== STRING_LEQ
|| k
== SEQ_NTH
)
235 // Get triggered for both equal and dis-equal
236 ee
->addTriggerPredicate(n
);
241 // Function applications/predicates
244 // Set d_functionsTerms stores all function applications that are
245 // relevant to theory combination. Notice that this is a subset of
246 // the applications whose kinds are function kinds in the equality
247 // engine. This means it does not include applications of operators
248 // like re.++, which is not a function kind in the equality engine.
249 // Concatenation terms do not need to be considered here because
250 // their arguments have string type and do not introduce any shared
252 if (n
.hasOperator() && ee
->isFunctionKind(k
) && k
!= STRING_CONCAT
)
254 d_functionsTerms
.push_back(n
);
256 if (options::stringFMF())
258 if (tn
.isStringLike())
260 // Our decision strategy will minimize the length of this term if it is a
261 // variable but not an internally generated Skolem, or a term that does
262 // not belong to this theory.
263 if (n
.isVar() ? !d_skCache
.isSkolem(n
)
264 : kindToTheoryId(k
) != THEORY_STRINGS
)
266 d_inputVars
.insert(n
);
267 Trace("strings-preregister") << "input variable: " << n
<< std::endl
;
273 void TermRegistry::registerTerm(Node n
, int effort
)
275 Trace("strings-register") << "TheoryStrings::registerTerm() " << n
276 << ", effort = " << effort
<< std::endl
;
277 if (d_registeredTerms
.find(n
) != d_registeredTerms
.end())
279 Trace("strings-register") << "...already registered" << std::endl
;
282 bool do_register
= true;
283 TypeNode tn
= n
.getType();
284 if (!tn
.isStringLike())
286 if (options::stringEagerLen())
288 do_register
= effort
== 0;
292 do_register
= effort
> 0 || n
.getKind() != STRING_CONCAT
;
297 Trace("strings-register") << "...do not register" << std::endl
;
300 Trace("strings-register") << "...register" << std::endl
;
301 d_registeredTerms
.insert(n
);
302 // ensure the type is registered
304 TrustNode regTermLem
;
305 if (tn
.isStringLike())
307 // register length information:
308 // for variables, split on empty vs positive length
309 // for concat/const/replace, introduce proxy var and state length relation
310 regTermLem
= getRegisterTermLemma(n
);
312 else if (n
.getKind() != STRING_STRCTN
)
314 // we don't send out eager reduction lemma for str.contains currently
315 Node eagerRedLemma
= eagerReduce(n
, &d_skCache
);
316 if (!eagerRedLemma
.isNull())
318 // if there was an eager reduction, we make the trust node for it
319 if (d_epg
!= nullptr)
321 regTermLem
= d_epg
->mkTrustNode(
322 eagerRedLemma
, PfRule::STRING_EAGER_REDUCTION
, {}, {n
});
326 regTermLem
= TrustNode::mkTrustLemma(eagerRedLemma
, nullptr);
330 if (!regTermLem
.isNull())
332 Trace("strings-lemma") << "Strings::Lemma REG-TERM : " << regTermLem
334 Trace("strings-assert")
335 << "(assert " << regTermLem
.getNode() << ")" << std::endl
;
336 d_im
->trustedLemma(regTermLem
, InferenceId::STRINGS_REGISTER_TERM
);
340 void TermRegistry::registerType(TypeNode tn
)
342 if (d_registeredTypes
.find(tn
) != d_registeredTypes
.end())
346 d_registeredTypes
.insert(tn
);
347 if (tn
.isStringLike())
349 // preregister the empty word for the type
350 Node emp
= Word::mkEmptyWord(tn
);
351 if (!d_state
.hasTerm(emp
))
353 preRegisterTerm(emp
);
358 TrustNode
TermRegistry::getRegisterTermLemma(Node n
)
360 Assert(n
.getType().isStringLike());
361 NodeManager
* nm
= NodeManager::currentNM();
362 // register length information:
363 // for variables, split on empty vs positive length
364 // for concat/const/replace, introduce proxy var and state length relation
366 if (n
.getKind() != STRING_CONCAT
&& !n
.isConst())
368 Node lsumb
= nm
->mkNode(STRING_LENGTH
, n
);
369 lsum
= Rewriter::rewrite(lsumb
);
370 // can register length term if it does not rewrite
373 registerTermAtomic(n
, LENGTH_SPLIT
);
374 return TrustNode::null();
377 Node sk
= d_skCache
.mkSkolemCached(n
, SkolemCache::SK_PURIFY
, "lsym");
378 StringsProxyVarAttribute spva
;
379 sk
.setAttribute(spva
, true);
380 Node eq
= Rewriter::rewrite(sk
.eqNode(n
));
382 // If we are introducing a proxy for a constant or concat term, we do not
383 // need to send lemmas about its length, since its length is already
385 if (n
.isConst() || n
.getKind() == STRING_CONCAT
)
387 // do not send length lemma for sk.
388 registerTermAtomic(sk
, LENGTH_IGNORE
);
390 Node skl
= nm
->mkNode(STRING_LENGTH
, sk
);
391 if (n
.getKind() == STRING_CONCAT
)
393 std::vector
<Node
> nodeVec
;
394 for (const Node
& nc
: n
)
396 if (nc
.getAttribute(StringsProxyVarAttribute()))
398 Assert(d_proxyVarToLength
.find(nc
) != d_proxyVarToLength
.end());
399 nodeVec
.push_back(d_proxyVarToLength
[nc
]);
403 Node lni
= nm
->mkNode(STRING_LENGTH
, nc
);
404 nodeVec
.push_back(lni
);
407 lsum
= nm
->mkNode(PLUS
, nodeVec
);
408 lsum
= Rewriter::rewrite(lsum
);
410 else if (n
.isConst())
412 lsum
= nm
->mkConst(Rational(Word::getLength(n
)));
414 Assert(!lsum
.isNull());
415 d_proxyVarToLength
[sk
] = lsum
;
416 Node ceq
= Rewriter::rewrite(skl
.eqNode(lsum
));
418 Node ret
= nm
->mkNode(AND
, eq
, ceq
);
420 // it is a simple rewrite to justify this
421 if (d_epg
!= nullptr)
423 return d_epg
->mkTrustNode(ret
, PfRule::MACRO_SR_PRED_INTRO
, {}, {ret
});
425 return TrustNode::mkTrustLemma(ret
, nullptr);
428 void TermRegistry::registerTermAtomic(Node n
, LengthStatus s
)
430 if (d_lengthLemmaTermsCache
.find(n
) != d_lengthLemmaTermsCache
.end())
434 d_lengthLemmaTermsCache
.insert(n
);
436 if (s
== LENGTH_IGNORE
)
441 std::map
<Node
, bool> reqPhase
;
442 TrustNode lenLem
= getRegisterTermAtomicLemma(n
, s
, reqPhase
);
443 if (!lenLem
.isNull())
445 Trace("strings-lemma") << "Strings::Lemma REGISTER-TERM-ATOMIC : " << lenLem
447 Trace("strings-assert")
448 << "(assert " << lenLem
.getNode() << ")" << std::endl
;
449 d_im
->trustedLemma(lenLem
, InferenceId::STRINGS_REGISTER_TERM_ATOMIC
);
451 for (const std::pair
<const Node
, bool>& rp
: reqPhase
)
453 d_im
->requirePhase(rp
.first
, rp
.second
);
457 SkolemCache
* TermRegistry::getSkolemCache() { return &d_skCache
; }
459 const context::CDList
<TNode
>& TermRegistry::getFunctionTerms() const
461 return d_functionsTerms
;
464 const context::CDHashSet
<Node
>& TermRegistry::getInputVars() const
469 bool TermRegistry::hasStringCode() const { return d_hasStrCode
; }
471 TrustNode
TermRegistry::getRegisterTermAtomicLemma(
472 Node n
, LengthStatus s
, std::map
<Node
, bool>& reqPhase
)
476 // No need to send length for constant terms. This case may be triggered
477 // for cases where the skolem cache automatically replaces a skolem by
479 return TrustNode::null();
481 Assert(n
.getType().isStringLike());
482 NodeManager
* nm
= NodeManager::currentNM();
483 Node n_len
= nm
->mkNode(kind::STRING_LENGTH
, n
);
484 Node emp
= Word::mkEmptyWord(n
.getType());
485 if (s
== LENGTH_GEQ_ONE
)
487 Node neq_empty
= n
.eqNode(emp
).negate();
488 Node len_n_gt_z
= nm
->mkNode(GT
, n_len
, d_zero
);
489 Node len_geq_one
= nm
->mkNode(AND
, neq_empty
, len_n_gt_z
);
490 Trace("strings-lemma") << "Strings::Lemma SK-GEQ-ONE : " << len_geq_one
492 Trace("strings-assert") << "(assert " << len_geq_one
<< ")" << std::endl
;
493 return TrustNode::mkTrustLemma(len_geq_one
, nullptr);
498 Node len_one
= n_len
.eqNode(d_one
);
499 Trace("strings-lemma") << "Strings::Lemma SK-ONE : " << len_one
501 Trace("strings-assert") << "(assert " << len_one
<< ")" << std::endl
;
502 return TrustNode::mkTrustLemma(len_one
, nullptr);
504 Assert(s
== LENGTH_SPLIT
);
506 // get the positive length lemma
507 Node lenLemma
= lengthPositive(n
);
508 // split whether the string is empty
509 Node n_len_eq_z
= n_len
.eqNode(d_zero
);
510 Node n_len_eq_z_2
= n
.eqNode(emp
);
511 Node case_empty
= nm
->mkNode(AND
, n_len_eq_z
, n_len_eq_z_2
);
512 Node case_emptyr
= Rewriter::rewrite(case_empty
);
513 if (!case_emptyr
.isConst())
515 // prefer trying the empty case first
516 // notice that requirePhase must only be called on rewritten literals that
517 // occur in the CNF stream.
518 n_len_eq_z
= Rewriter::rewrite(n_len_eq_z
);
519 Assert(!n_len_eq_z
.isConst());
520 reqPhase
[n_len_eq_z
] = true;
521 n_len_eq_z_2
= Rewriter::rewrite(n_len_eq_z_2
);
522 Assert(!n_len_eq_z_2
.isConst());
523 reqPhase
[n_len_eq_z_2
] = true;
527 // If n = "" ---> true or len( n ) = 0 ----> true, then we expect that
528 // n ---> "". Since this method is only called on non-constants n, it must
529 // be that n = "" ^ len( n ) = 0 does not rewrite to true.
530 Assert(!case_emptyr
.getConst
<bool>());
533 if (d_epg
!= nullptr)
535 return d_epg
->mkTrustNode(lenLemma
, PfRule::STRING_LENGTH_POS
, {}, {n
});
537 return TrustNode::mkTrustLemma(lenLemma
, nullptr);
540 Node
TermRegistry::getSymbolicDefinition(Node n
, std::vector
<Node
>& exp
) const
542 if (n
.getNumChildren() == 0)
544 Node pn
= getProxyVariableFor(n
);
549 Node eq
= n
.eqNode(pn
);
550 eq
= Rewriter::rewrite(eq
);
551 if (std::find(exp
.begin(), exp
.end(), eq
) == exp
.end())
557 std::vector
<Node
> children
;
558 if (n
.getMetaKind() == metakind::PARAMETERIZED
)
560 children
.push_back(n
.getOperator());
562 for (const Node
& nc
: n
)
564 if (n
.getType().isRegExp())
566 children
.push_back(nc
);
570 Node ns
= getSymbolicDefinition(nc
, exp
);
577 children
.push_back(ns
);
581 return NodeManager::currentNM()->mkNode(n
.getKind(), children
);
584 Node
TermRegistry::getProxyVariableFor(Node n
) const
586 NodeNodeMap::const_iterator it
= d_proxyVar
.find(n
);
587 if (it
!= d_proxyVar
.end())
594 Node
TermRegistry::ensureProxyVariableFor(Node n
)
596 Node proxy
= getProxyVariableFor(n
);
600 proxy
= getProxyVariableFor(n
);
602 Assert(!proxy
.isNull());
606 void TermRegistry::removeProxyEqs(Node n
, std::vector
<Node
>& unproc
) const
608 if (n
.getKind() == AND
)
610 for (const Node
& nc
: n
)
612 removeProxyEqs(nc
, unproc
);
616 Trace("strings-subs-proxy") << "Input : " << n
<< std::endl
;
617 Node ns
= Rewriter::rewrite(n
);
618 if (ns
.getKind() == EQUAL
)
620 for (size_t i
= 0; i
< 2; i
++)
622 // determine whether this side has a proxy variable
623 if (ns
[i
].getAttribute(StringsProxyVarAttribute()))
625 if (getProxyVariableFor(ns
[1 - i
]) == ns
[i
])
627 Trace("strings-subs-proxy")
628 << "...trivial definition via " << ns
[i
] << std::endl
;
629 // it is a trivial equality, e.g. between a proxy variable
630 // and its definition
636 if (!n
.isConst() || !n
.getConst
<bool>())
638 Trace("strings-subs-proxy") << "...unprocessed" << std::endl
;
643 } // namespace strings
644 } // namespace theory