This file contains a summary of important user-visible changes.
- Added support for
SymbolManager::getNamedTerms()
to retrieve the set of terms that have been given names by the SMT-LIB attribute:named
.
-
The option
--safe-options
now disables experimental theories and their extensions in cvc5. This includes the theory of bags, the theory of finite fields, the theory of separation logic, higher-order extensions to the theory of equality, as well as extensions of the theory of arithmetic for transcendental functions, integer-and, and power functions. These theories are still enabled by default and further more can be used in combination with safe options by the options e.g.--arith-exp
prior to setting--safe-options
. -
Renamed the flag
--sets-ext
to--sets-exp
, which enables non-standard extensions of the sets theory.
-
New C API, implemented as a thin wrapper around the C++ API.
-
Exposed creation and maintenance of Skolem functions to the API. Skolem functions are symbols that are internally introduced by cvc5 during solving. They are formalized via an identifier (see https://cvc5.github.io/docs-ci/docs-main/skolem-ids.html for details) as well as a (possibly empty) set of indices.
- The API methods
Term::isSkolem()
,Term::getSkolemId()
, andTerm::getSkolemIndices()
may now be used to identify terms corresponding to skolems.
- The API methods
-
We now export kinds
BITVECTOR_FROM_BOOLS
,BITVECTOR_BIT
,DIVISION_TOTAL
,INTS_DIVISION_TOTAL
,INTS_MODULUS_TOTAL
which may appear in terms resulting from simplification or terms appearing in proofs. -
Proof rules corresponding to rewrite rules are now exported in the API via the enum
ProofRewriteRule
. -
A new
Plugin
class is available in our API. This class allows a user to be notified when SAT clauses or theory lemmas are learned internally by cvc5. The plugin class also contains a callback for the user to introduce new learned clauses into the state of cvc5 during solving. -
Added a new strategy
--mbqi-fast-sygus
(disabled by default) for quantifier instantiation which uses SyGuS enumeration to augment instantiations from model-based quantifier instantiation.
-
We now require CMake >= 3.16.
-
API All APIs have been refactored to expose a TermManager to the API. A term manager manages creation and maintenance of all terms and sorts (across potentially several solver instances within a thread). Corresponding functions that were previously associated with a solver instance and are now associated with a term manager are now deprecated and will be removed in a future release.
- Python:
- Constructor
SymbolManager(Solver)
is now deprecated and replaced bySymbolManager(TermManager)
.
- Constructor
- Pythonic:
- Unsat cores and proofs are now available via the
Solver
methodsunsat_core()
andproof()
, respectively.
- Unsat cores and proofs are now available via the
- Python:
-
The default proof format of cvc5 has been renamed to the Cooperating Proof Calculus (CPC) proof format. This proof format coincides with proof objects in our API.
- API
- The option
--proof-format=cpc
prints proofs in the CPC format. This option is now enabled by default.
- The option
- The Ethos proof checker is available, which can check proofs in this
format. In particular, the 0.1.0 release of Ethos can check proofs generated
by this version of cvc5. This checker is the second generation of the
AletheLF checker (
alfc
). Ethos inherits the code base of alfc and is based on a logical framework called Eunoia (formerly called AletheLF). See https://github.com/cvc5/ethos for more details. Note that the development version of Ethos is available for download via the script./contrib/get-ethos-checker
, which will be kept in sync with further development versions of cvc5. - The rules of this format have been formalized in Eunoia and are available
in the cvc5 repository under the directory
./proofs/eo/cpc/
.
- API
-
The semantics of
SQRT
was changed to assume the result is positive. -
Fixed a bug involving how sequence terms are printed in model values.
-
Added support for nullable sorts and lift operator to the theory of datatypes.
-
Added a new strategy
--sub-cbqi
(disabled by default) for quantifier instantiation which uses subsolvers to compute unsat cores of instantiations that are used in proofs of unsatisfiability. -
Add support for the kind
BITVECTOR_NEGO
corresponding to bitvector negation overflow detection.
- SAT clauses are no longer marked as removable in MiniSat. This change improves performance overall on quantifier-free logics with arithmetic and strings.
- API
- Functions
kindToString(Kind)
andsortKindToString(SortKind)
are now deprecated and replaced bystd::to_string(Kind)
andstd::to_string(SortKind)
. They will be removed in the next minor release.
- Functions
- Minor changes to the output format for proofs. Proofs in the AletheLF
proof format generated by cvc5 now correspond directly to their representation
in proof objects obtained in via the API (the
Proof
class). Moreover, proofs now use SMT-LIB compliant syntax for quantified formulas and unary arithmetic negation. - The option
--safe-options
now disallows cases where more than one regular option is used. - Fixes the parsing of
define-fun-rec
anddefine-funs-rec
in interactive mode. - Renamed
bag.duplicate_removal
tobag.setof
. - Improvements to handling of constant production rules (
Constant
) in SyGuS grammars.
- Added support for forward declarations (feature
:fwd-decls
) in SyGuS inputs. This allows functions-to-synthesize to include previous functions-to-synthesize in their grammars. This feature is enabled by default for all SyGuS inputs.
- Fixed a bug when piping input from stdin, which caused cvc5 to have degraded performance. This bug could also cause cvc5 to throw spurious parser errors.
-
API
- The signature of functions
Solver::mkFiniteFieldSort(const std::string&)
andSolver::mkFiniteFieldElem(const std::string&, const Sort&)
is now extended with an additional (optional) parameter toSolver::mkFiniteFieldSort(const std::string& size, uint32_t base)
andSolver::mkFiniteFieldElem(const string& value, const Sort& sort, uint32_t base)
to configure the base of the string representation of the given string parameters. - A new API for proofs is available. The new
Proof
class represents a node of the proof tree. FunctionSolver::getProof(modes::ProofComponent c = modes::ProofComponent::FULL)
returns the root proof nodes of a proof component as a vector. FunctionSolver::proofToString(std::vector<Proof> proof, modes::ProofFormat format, modes::ProofComponent component)
can be used to print proof components to a string with a specified proof format. - Added support for parsing inputs from file, string, or input stream. We
provide an
InputParser
,SymbolManager
, andCommand
class for reading inputs (seeinclude/cvc5_parser.h
). Example use cases of these classes are available at https://cvc5.github.io/docs/cvc5-1.1.1/examples/parser.html and https://cvc5.github.io/docs/cvc5-1.1.1/examples/parser_sym_manager.html. These interfaces are also available in the Java and Python APIs. - Added a variant of timeout cores that accepts a set of assumptions. This
is available via the API method
Solver::getTimeoutCoreAssuming
or the SMT-LIB commandget-timeout-core-assuming
, which accept a list of formulas to assume, while all current assertions are implicitly included in the core. - Added new method
Solver::getUnsatCoreLemmas
which returns the set of theory lemmas that were relevant to showing the last query was unsatisfiable. This is also avialable via the SMT-LIB commandget-unsat-core-lemmas
.
- The signature of functions
-
Support for the AletheLF (ALF) proof format. This format combines the strengths of the Alethe and LFSC proof formats, namely it borrows much of the syntax of Alethe, while being based on a logical framework like LFSC.
- API
- The option
--proof-format=alf
prints proofs in the AletheLF format. This option is enabled by default.
- The option
- The ALF proof checker (alfc) is available for download via the script
./contrib/get-alf-checker
.
- API
-
CaDiCaL is now integrated via the IPASIR-UP interface as CDCL(T) SAT solver. The CDCL(T) SAT solver can be configured via option
--sat-solver
. Currently, MiniSat is still default. Note that CaDiCaL cannot be used as the CDCL(T) SAT engine when proof production is enabled. In that case, option--sat-solver
will default back to MiniSat.
- Various bug fixes.
Note: This is a pre-release version for the upcoming version 1.1.0.
- API
- Added the ability to query the logic that has been set in the solver via
Solver::isLogicSet
andSolver::getLogic
.
- Added the ability to query the logic that has been set in the solver via
- SMT-LIB
- The syntax for 0-ary tuples has been changed for the purposes of
disambiguation. The new syntax for 0-ary tuple sort is
UnitTuple
whose 0-ary constructor istuple.unit
(the previous syntax had overloadedTuple
andtuple
, with no arguments).
- The syntax for 0-ary tuples has been changed for the purposes of
disambiguation. The new syntax for 0-ary tuple sort is
Note: This is a pre-release version for the upcoming version 1.1.0.
- API
- C++ enums are now enum classes
- Added argument
fresh
toSolver::declareFun
which distinguishes whether the solver should construct a new term or (when applicable) return a term constructed with the same name and sort. An analogous flag is added toSolver::declareSort
. The option--fresh-declarations
determines whether the parser constructs fresh terms and sorts for each declaration (true by default, which matches the previous behavior).
- Various bug fixes
-
API
- New API function
Solver::mkFloatingPoint(const Term& sign, const Term& exp, const Term& sig)
, returns a floating-point value from the three IEEE-754 bit-vector value. - Added support for querying the values of real algebraic numbers in the API.
In particular, the methods
Term::isRealAlgebraicNumber()
,Term::getRealAlgebraicNumberDefiningPolynomial()
,Term::getRealAlgebraicNumberLowerBound()
, andTerm::getRealAlgebraicNumberUpperBound()
may now be used to query the contents of terms corresponding to real algebraic numbers.
- New API function
-
Support for timeout cores
- API
- New API function
Solver::getTimeoutCore()
when applicable returns a subset of the current assertions that cause the solver to timeout without a provided timeout (option--timeout-core-timeout
).
- New API function
- SMT-LIB
- New command
(get-timeout-core)
which invokes the above method.
- New command
- API
-
Support for new interfaces to the SyGuS solver.
- API
- New API function
Solver::findSynth
which takes an identifier specifying a target term to synthesize and (optionally) a grammar. This method can be used to directly enumerate terms in a provided grammar (FindSynthTarget::ENUM
), or as a way of finding other terms of interest, e.g,. a rewrite rule that is applicable to the current set of assertions (FindSynthTarget::REWRITE_INPUT
). - New API function
Solver::findSynthNext
which gets the next term in the enumeration.
- New API function
- SMT-LIB
- New commands
find-synth
andfind-synth-next
which invoke the above methods.
- New commands
- API
- Removed support for the ANTLR parser and parsing for the TPTP language. components.
- API
- Simplified the
Solver::mkTuple
method. The sorts of the elements no longer need to be provided. - Option
--print-unsat-cores-full
has been renamed to--print-cores-full
. Setting this option to true will print all assertions in the unsat core, regardless of whether they are named. This option also impacts how timeout cores are printed. - Removed API support for the deprecated SyGuS 2.0 command
Solver::synthInv
. This method is equivalent toSolver::synthFun
with a Boolean range sort.
- Simplified the
- A new (hand-written) parser is available and enabled by default.
- Note: the previous parser can be enabled using command line options
--no-flex-parser --no-stdin-input-per-line
. These options will be available until version 1.1 is released.
- Note: the previous parser can be enabled using command line options
- Support for the theory of (prime-order) finite fields:
- Sorts are created with
- C++:
Solver::makeFiniteFieldSort
- SMT-LIB:
(_ FiniteField P)
for prime orderP
- C++:
- Constants are created with
- C++:
Solver::makeFiniteFieldElem
- SMT-LIB:
(as ffN F)
for integerN
and field sortF
- C++:
- The operators are multiplication, addition and negation
- C++: kinds
FF_MUL
,FF_ADD
, andFF_NEG
- SMT-LIB: operators
ff.mul
,ff.add
, andff.neg
- C++: kinds
- The only predicate is equality
- Sorts are created with
- API
- New API function
Solver::getVersion()
, returns a string representation of the solver version.
- New API function
- Support for bit-vector overflow detection operators:
BITVECTOR_UADDO
unsigned addition overflow detectionBITVECTOR_SADDO
signed addition overflow detectionBITVECTOR_UMULO
unsigned multiplication overflow detectionBITVECTOR_SMULO
signed multiplication overflow detectionBITVECTOR_USUBO
unsigned subtraction overflow detectionBITVECTOR_SSUBO
signed subtraction overflow detectionBITVECTOR_SDIVO
signed division overflow detection
- Support for Web Assembly compilation using Emscripten.
- The (non-standard) operators
BITVECTOR_TO_NAT
andINT_TO_BITVECTOR
now belong to the UF theory. A logic that includes UF is required to use them. - The sort for (non-standard) bit-vector operators
BITVECTOR_REDAND
andBITVECTOR_REDOR
is now(_ BitVec 1)
(was Boolean), following the definition of reduction operators in Verilog (their origin). - Reenable functionality that allows
(get-model)
commands after answeringunknown
when:produce-models
is set totrue
. Note that there is no guarantee that building a model succeeds in this case.
- API
- Previously, it was not possible to share Sort, Term, Op, Grammar and datatype objects between Solver instances. This is now allowed for solvers that belong to the same thread.
- Support for cross-compiling an ARM binary of cvc5 on x86 macOS.
- Support for declaring oracle functions in the API via the method
declareOracleFun
. This allows users to declare functions whose semantics are associated with a provided executable implementation.
- Removed support for non-standard
declare-funs
,declare-consts
, anddeclare-sorts
commands. - The kind for integer constants is now
CONST_INTEGER
, while real constants continue to have kindCONST_RATIONAL
. - Type rules for (dis)equality and if-then-else no longer allow mixing
integers and reals. Type rules for other operators like
APPLY_UF
now require their arguments to match the type of the function being applied, and do not assume integer/real subtyping. - The API method
mkTuple
no longer supports casting integers to reals when constructing tuples.
cvc5: A Versatile and Industrial-Strength SMT Solver.
Barbosa H., Barrett C., Brain M., Kremer G., Lachnitt H., Mann M., Mohamed A.,
Mohamed M., Niemetz A., Nötzli A., Ozdemir A., Preiner M., Reynolds A., Sheng
Y., Tinelli C., and Zohar Y., TACAS 2022.
http://dx.doi.org/10.1007/978-3-030-99524-9_24
Documentation: https://cvc5.github.io/docs/latest/api/cpp/cpp.html
- Base module: Feature complete with C++ API
- Pythonic module: A pythonic wrapper around the base module
- Documentation: https://cvc5.github.io/docs/latest/api/python/python.html
Documentation: https://cvc5.github.io/docs/latest/api/java/java.html
A new theory of bags, which are collections that allow duplicates. It supports basic operators like union disjoint, union max, intersection, difference subtract, difference remove, duplicate removal, and multiplicity of an element in a bag.
- A new parametric theory of sequences whose syntax is compatible with the syntax for sequences used by Z3.
- A new array-inspired procedure (option
--seq-array
).
Nonlinear real arithmetic is now solved using a new solver based on
cylindrical algebraic coverings. Includes full support for non-rational
models and a number of options --nl-cov-*
for, e.g., different projection
operators, Lazard's lifting or variable elimination.
The new solver requires the libpoly library, and Lazard's lifting
additionally requires CoCoALib.
Added support for an eqrange
predicate. (eqrange a b i j)
is true
if arrays a
and b
are equal on all indices within indices i
and j
.
- New bit-vector solver with CaDiCaL as default SAT back-end.
- New approach for solving bit-vectors as integers (option
--solve-bv-as-int
).
Support for generic datatype updaters ((_ update s) t u)
which replaces
the field specified by selector s
of t
by the value u
.
- Support for an integer operator
(_ iand n)
that returns the bitwiseand
of two integers, seen as integers modulo n. - Support for an integer operator
int.pow2
, used as(int.pow2 x)
which represents 2 to the power of x.
- Improved support for enumerative instantiation (option
--enum-inst
). - SyGuS-based quantifier instantiation (option
--sygus-inst
).
- Improved performance using new techniques, including model-based reductions, eager context-dependent simplification, and equality-based conflict finding.
- Support for
str.indexof_re(s, r, n)
, which returns the index of the first occurrence of a regular expressionr
in a strings
after indexn
or -1 ifr
does not match a substring aftern
.
- Documentation available at: https://cvc5.github.io/docs/latest/proofs/proofs.html
- When used after an unsatisfiable response to
checkSat
,getProof
returns a representation of the refutation proof for the current set of assertions (get-proof in SMT-LIB). - Preliminary support for translations to external proof formats LFSC and Alethe.
The API method getDifficulty
returns a map from asserted formulas to
integers which estimates how much that formula contributed to the
difficulty of the overall problem (get-difficulty
in SMT-LIB).
The API method getLearnedLiterals
gets a list of literals that are
entailed by the current set of assertions that were learned during solving
(get-learned-literals
in SMT-LIB).
- The API method
getAbduct
can be used to find an abduct for the current set of assertions and provided goal (get-abduct in SMT-LIB). Optionally, a SyGuS grammar can be provided to restrict the shape of possible abducts. If using the text inferace, the grammar may be provided using the same syntax for grammars as in SyGuS IF format. - The API method
getInterpolant
can be used to find an interpolant for the current set of assertions and provided goal (get-interpolant in SMT-LIB). Like abduction, grammars may be provided to restrict the shape of interpolants.
- The core SyGuS solver now supports getting multiple solutions for a
synthesis conjecture via the API. The method
checkSynthNext
finds the next SyGuS solution to the current set of SyGuS constraints (check-synth-next
in SyGuS IF). getAbductNext
finds the next abduct for the current set of assertions and provided goal (get-abduct-next
in SMT-LIB).getInterpolantNext
finds the next interpolant for the current set of assertions and provided goal (get-interpolant-next
in SMT-LIB).
- The API method
declarePool
declares symbol sets of terms called pools (declare-pool
in SMT-LIB). - Pools can be used in annotations of quantified formulas for fine grained
control over quantifier instantiations (
:inst-pool
,:inst-add-to-pool
,:skolem-add-to-pool
in SMT-LIB).
The option -o TAG
can be used to enable a class of useful diagnostic
information, such as the state of the input formula before and after
pre-preprocessing, quantifier instantiations, literals learned
during solving, among others. For SyGuS problems, it can be used to
show candidate solutions and grammars that the solver has generated.
- Production modes based on the new proof infrastructure
(
--unsat-cores-mode=sat-proof
) and on the solving-under-assumption feature of Minisat (--unsat-cores-mode=assumptions
). The mode based on SAT assumptions + preprocessing proofs is the new default. - Computing minimal unsat cores (option
--minimal-unsat-cores
).
- The API method
blockModels
can be used to block the current model using various policies for how to exclude the values of terms (block-model
in SMT-LIB). - The API method
blockModelValues
can be used to block the current model for a provided set of terms (block-model-values
in SMT-LIB).
- The API method
isModelCoreSymbol
can be used to query whether the value for a symbol was critical to whether the model satisfies the current set of assertions. - Models can be limited to show only model core symbols (option
--model-cores
).
- CaDiCaL and SymFPU are now required dependencies. CaDiCaL 1.4.1 is now the version used by default.
- Options have been extensively refactored, please refer to the cvc5 documentation for further information.
- Removed support for the CVC language.
- SyGuS: Removed support for SyGuS-IF 1.0.
- Removed support for the (non-standard)
define
command. - Removed the legacy API along with the Java and Python bindings for it.
- Interactive shell: the GPL-licensed Readline library has been replaced the
BSD-licensed Editline. Compiling with
--best
now enables Editline, instead of Readline. Without selecting optional GPL components, Editline-enabled cvc5 builds will be BSD licensed. - The
competition
build type includes the dependencies used for SMT-COMP by default. Note that this makes this build type produce GPL-licensed binaries. - Bit-vector operator
bvxnor
was previously mistakenly marked as left-assoicative in SMT-LIB. This has recently been corrected in SMT-LIB. We now restrictbvxnor
to only allow two operands in order to avoid confusion about the semantics, since the behavior of n-ary operands tobvxnor
is now undefined in SMT-LIB. - SMT-LIB output for
get-model
command now conforms with the standard, and does not begin with the keywordmodel
. The output is the same as before, only with this word removed from the beginning. - Building with Python 2 is now deprecated.
- The SMT-LIB syntax for some extensions has been changed. Notably, set
operators are now prefixed by
set.
, and relations byrel.
. For example,union
is now writtenset.union
. - Removed support for redundant logics
ALL_SUPPORTED
andQF_ALL_SUPPORTED
, useALL
andQF_ALL
instead.