Age | Commit message (Collapse) | Author |
|
collectModelInfo with fullModel argument, most theory-specific implementation out of the model class, model printer relegated to printer classes. Also updates to finite mode finding, modifications to datatypes making them compatible with theory combination, support for theory-specific handling of user attributes, refactoring of uf models
|
|
|
|
no explicit setLogic(). This is important for the CVC language, where the parser doesn't ensure that setLogic() is called, and for API uses. setLogicInternal() should be called in order to properly set up heuristics, even if the logic is just ALL_SUPPORTED.
This means that the CVC language can now take advantage of statistics.
Also added the ability to set the logic from CVC presentation language via (e.g.)
OPTION "logic" "QF_UFLIA";
Disabled the justification decision heuristic for ALL_SUPPORTED, as it interferes with incrementality. Kshitij may have a fix (I warned him about this commit).
|
|
|
|
1. changes the way options are declared (see http://church.cims.nyu.edu/wiki/Options)
2. moves module-specific options enumerations (SimplificationMode, DecisionMode, ArithUnateLemmaMode, etc.) to their own header files, also they are no longer inside the Options:: class namespace.
3. includes many SMT-LIBv2 compliance fixes, especially to (set-option..) and (get-option..)
The biggest syntactical changes (outside of adding new options) you'll notice are in accessing and setting options:
* to access an option, write (e.g.) options::unconstrainedSimp() instead of Options::current()->unconstrainedSimp.
* to determine if an option value was set by the user, check (e.g.) options::unconstrainedSimp.wasSetByUser().
* ensure that you have the option available (you have to #include the right module's options.h file, e.g. #include "theory/uf/options.h" for UF options)
*** this point is important. If you access an option and it tells you the option doesn't exist, you aren't #including the appropriate options.h header file ***
Note that if you want an option to be directly set (i.e., other than via command-line parsing or SmtEngine::setOption()), you need to mark the option :read-write in its options file (otherwise it's read-only), and you then write (e.g.) options::unconstrainedSimp.set(true).
Adding new options is incredibly simple for primitive types (int, unsigned, bool, string, double). For option settings that you need to turn into a member of an enumerated type, you write a custom "handler" for the option---this is no additional work than it was before, and there are many examples to copy from (a good one is stringToSimplificationMode() in src/smt/options_handlers.h).
Benefits of the new options system include:
1. changes to options declarations don't require a full-source rebuild (you only have to rebuild those sources that depend on the set of options that changed).
2. lots of sanity checks (that the same option isn't declared twice, that option values are in range for their type, that all options are documented properly, etc.)
3. consistency: Boolean-valued option --foo gets a --no-foo automatically, documentation is generated consistently, the option-parsing matches the documented option name, etc.
4. setting options programmatically via SmtEngine::setOption() is enabled, and behaves the same as command-line equivalents (including checking the value is in range, etc.)
5. the notion of options being "set by the user" is now primitive; you can use (e.g.) options::unconstrainedSimp.wasSetByUser() instead of having to use (and maintain) a separate Boolean option for the purpose
I've taken lots of care not to break anything. Hopefully, I've succeeded in that.
|
|
|
|
got it in quickly for Andy.
A "fair" version forthcoming.
|
|
Basically, this involves creating a separate StatisticsRegistry for the
ExprManager and for the SmtEngine. Otherwise, theories register the
same statistic twice. This is a larger problem, though, for creating
multiple instances of theories, and that is unaddressed. Still,
separating out the expr statistics into a separate registry is probably
a good idea, since the expr package is somewhat separate anyway (and in
the short term it allows two SmtEngines to co-exist).
|
|
support yet for enumerating arrays, or for enumerating non-trivial datatypes.
|
|
Adds TheoryQuantifiers and TheoryRewriteRules, QuantifiersEngine, and other infrastructure.
Adds theory instantiators to many theories.
Adds the UF strong solver.
|
|
SmtEngine resolved.
ALL_SUPPORTED and QF_ALL_SUPPORTED logics now supported by SMT-LIB parsers. In SMT-LIBv2, if a (set-logic..) command is missing, ALL_SUPPORTED is assumed, and a warning is issued, as discussed on the cvc4-devel mailing list.
|
|
are still the AUFBV wrong results, but it seems better.
http://church.cims.nyu.edu/regress-results/compare_jobs.php?job_id=4382&reference_id=4359&p=5
|
|
|
|
QF_UF and maybe others
|
|
|
|
This should also fix bug 325.
|
|
Changes include
* fixed term visitor from the bvprop branch
* removed all the warnings from builds -- warnings are there to be noted *NOT* to be used as scribbles
* moved the LogicInfo into the theory constructor
|
|
src/theory/logic_info.{h,cpp} contains the CVC4::LogicInfo class, which keeps
track of which theories are active (which should remain constant throughout
the life of an SmtEngine) and other details (like integers, reals,
linear/nonlinear, etc.).
This class has a default constructor which is the most all-inclusive logic.
Alternatively, this class can be constructed from an SMT-LIB logic string
(the empty string gives the same as "QF_SAT"). Once constructed, theories
can be enabled or disabled, quantifiers flipped on and off, integers flipped
on and off, etc. At any point an SMT-LIB-like logic string can be extracted.
The SmtEngine keeps a LogicInfo for itself and shares with the TheoryEngine
(and, in turn, the theories) only a const reference to it. This ensures that
the logic info doesn't mutate over the course of the run.
As part of this commit, the TheoryEngine's old notion of "active theories"
has been completely removed. As a result, SMT benchmarks that are incorrectly
tagged with a logic will assert-fail or worse. (We should probably fail
more gracefully in this case.) One such example was bug303.smt2,
which used UF reasoning but was tagged QF_LIA. This has been fixed.
|
|
Below is a highlight of the changes:
- This introduces a new normal form to arithmetic.
-- Equalities and disequalities are in solved form.
Roughly speaking this means: (= x (+ y z)) is in normal form.
(See the comments in normal_form.h for what this formally requires.)
-- The normal form for inequality atoms always uses GEQ and GT instead of GEQ and LEQ.
Integer atoms always use GEQ.
- Constraint was added to TheoryArith.
-- A constraint is a triple of (k x v) where:
--- k is the type of the constraint (either LowerBound, UpperBound, Equality or Disequality),
--- x is an ArithVar, and
--- v is a DeltaRational value.
-- Constraints are always attached to a ConstraintDatabase.
-- A Constraint has its negation in the ConstraintDatabase [at least for now].
-- Every constraint belongs to a set of constraints for each ArithVar sorted by the delta rational values.
-- This set can be iterated over and provides efficient access to other constraints for this variable.
-- A literal may be attached to a constraint.
-- Constraints with attached literals may be marked as being asserted to the theory (sat context dependent).
-- Constraints can be propagated.
-- Every constraint has a proof (sat context dependent).
-- Proofs can be explained for either conflicts or propagations (if the node was propagated). (These proofs may be different.)
-- Equalities and disequalities can be marked as being split (user context dependent)
- This removes and replaces:
-- src/theory/arith/arith_prop_manager.*
-- src/theory/arith/atom_database.*
-- src/theory/arith/ordered_set.h
- Added isZero(), isOne() and isNegativeOne() to Rational and Integer.
- Added operator+ to CDList::const_iterator.
- Added const_iterator to CDQueue.
- Changes to regression tests.
|
|
Seems to be working better <http://church.cims.nyu.edu/regress-results/compare_jobs.php?job_id=3749&category=&p=5&reference_id=3739>, and should fix the failing cases in the regressions.
Removing one test case from the integer regress0.
|
|
* separated SatSolverInput interface class into two classes:
- TheoryProxy for the sat solver to communicate with the theories
- SatSolverInterface abstract class to communicate with the sat solver
* instead of using #ifdef typedef for SatClauses and SatLiterals, now there are CVC4 SatLiteral/SatClause types and mappings between them and the internal sat solver clause/literal representation
* added abstract classes for DPLLSatSolver and BVSatSolver different interfaces
Replaced TheoryBV with bitblasting implementation:
* all operators bitblasted
* only operator elimination rewrite rules so far
|
|
to request a decision on a literal. All these theory requests are kept in a
context-dependent queue and serviced in order when the SAT solver goes to make a
decision. Requests that don't have a SAT literal give an assert-fail. Requests
for literals that already have an assignment are silently ignored.
Since the queue is CD, requests can actually be serviced more than once (e.g., if
a request is made at DL 5, but not serviced until DL 10, and later, a conflict
backtracks to level 7, the request may be serviced again).
Performance impact: none to negligible for theories that don't use it
See http://church.cims.nyu.edu/regress-results/compare_jobs.php?job_id=3620&reference_id=3614&mode=&category=&p=0
|
|
|
|
* OutputChannel::lemma() now returns an unsigned int. This facility isn't functional yet, but the signature is there. For now, it always returns the current user level (which is "correct" from the interface point of view, but not what we want).
* Pseudobooleans disabled. This should fix some quantifier benchmarks Andy's been working with on the quantifiers2 branch.
* --limit / --time-limit options renamed --rlimit and --tlimit.
There may be slowdown from disabling pseudobooleans.
|
|
|
|
and model gen also.
I also expect this commit to fix bug #273.
No performance change is expected on regressions with this commit, see
http://goedel.cims.nyu.edu/regress-results/compare_jobs.php?job_id=2871&reference_id=2863
|
|
|
|
Dumping infrastructure. Can dump preprocessed queries and clauses. Can
also dump queries (for testing with another solver) to see if any conflicts
are missed, T-propagations are missed, all lemmas are T-valid, etc. For a
full list of options see --dump=help.
CUDD building much cleaner.
Documentation and assertion fixes.
Printer improvements, printing of commands in language-defined way, etc.
Typechecker stuff in expr package now autogenerated, no need to manually
edit the expr package when adding a new theory.
CVC3 compatibility layer (builds as libcompat).
SWIG detection and language binding support (infrastructure).
Support for some Z3 extended commands (like datatypes) in SMT-LIBv2 mode
(when not in compliance mode).
Copyright and file headers regenerated.
|
|
- This adds code for bounds refinement, and conflict weakening.
- This adds util/boolean_simplification.h.
- This adds a propagation manager to theory of arithmetic.
- Propagation is disabled by default.
- Propagation can be enabled by the command line flag "--enable-arithmetic-propagation"
- Propagation interacts *heavily* with rewriting equalities, and will work best if the command line flag "--rewrite-arithmetic-equalities" is enabled.
|
|
trunk commits)
|
|
|
|
Also, only build doxygen documentation on stuff in src/,
not test/ or contrib/ or anywhere else. Hopefully this
turns our 3000+ page user manual into something a little
more useful!
|
|
called. The OutputChannel is now untouched by TheoryArith during preregistration.
|
|
* Makes Options an "omnipresent thread-local global" (like the notion
of the "current NodeManager" was already). Options::current() accesses
this structure.
* Removes Options from constructors and data structures everywhere
(this cleans up a lot of things).
* No longer uses StatisticsRegistry statically. An instance of the
registry is created and linked to a NodeManager.
* StatisticsRegistry::current() is similar to Options::current(), but
the pointer is stowed in the NodeManager (rather than stored)
* The static functions of StatisticsRegistry have been left, for backward
compatibility; they now use the "current" statistics registry.
* SmtEngine::getStatisticsRegistry() is a public accessor for the
registry; this is needed by main() to reach in and get the registry,
for flushing statistics at the end.
|
|
are somewhat disparate but belonged on the same branch because they were
held back from trunk all for the same reason (to keep the trunk stable
for furious bitvector development). Dejan has now given me the go-ahead
for a merge.
=========================================
THIS COMMIT CHANGES THE THEORY INTERFACE!
=========================================
Theory constructors are expected to take an additional "Valuation*"
parameter that each Theory should send along to the base class
constructor. The base class Theory keeps the Valuation* in a
d_valuation field for use by it and by its derived classes.
Theory::getValue() no longer takes a Valuation* (it is expected
to use d_valuation instead). This allows other theory functions
to take advantage of getValue() for debugging or heuristic
purposes.
TODO BEFORE MERGE TO TRUNK:
****implement BitIterator find() in CDAttrHash<bool>.
Specifically:
* Added QF_BV support for SMT-LIB v2.
* Two adjustments to the theory interface as requested by Tim King:
1. As described above.
2. Theories now have const access to the fact queue through base
class functions facts_begin() and facts_end(); useful for
debugging.
* Added an "Asserted" attribute so that theories can check if something
has been asserted or not (and therefore not propagate it). However, this
has been disabled for now, pending more data on the overhead of it, and
pending discussion at the 3/25/2011 meeting.
* Do not define NDEBUG in MiniSat in assertion-enabled builds (so
that MiniSat asserts are evaluated).
* As a result of the new MiniSat assertions, some --incremental
regressions had to be disabled; also, some bitvectors ?!!
* Bug 71 is resolved by adding a specialization for CDAttrHash<> in the
attribute package.
* Fixes for some warnings flagged by clang.
* System tests have arrived! So far mainly infrastructure for having
system tests, but there is a system test aimed at improving code
coverage of the printer package.
* Minor other adjustments to documentation and coding to be more
conformant to CVC4 policy.
Tests have been performed to demonstrate that these changes have no or
negligible effect on performance. In particular, changing the
CDAttrHash<> doesn't have any real effect on performance or memory right
now, since there is only one context-dependent boolean flag (as soon
as another is added, the effect is noticeable but probably still slight).
|
|
dependences; now, if you touch theory_engine.h, only a few things in theory need be recompiled (TheoryEngine, SharedTermManager, .... but no theory implementations), along with the PropEngine and SmtEngine. If you touch a specific theory's .h file, only that theory must be recompiled (along with the TheoryEngine, since it uses traits, and SmtEngine, since it tells the TheoryEngine which theory implementations to use).
|
|
|
|
and not yet finalized but I need to put it in to work further with the theory writers. Please check the files that you 'own'. Any comments or discussion is welcome. Further details will be coming in a follow up email later.
|
|
* Theory::staticLearning() for statically adding new T-stuff before
normal preprocessing. UF's staticLearning() does transitivity of
equality/iff, solving the diamonds.
* more aggressive T-propagation for UF
* new KEEP_STATISTIC macro to hide Theories from having to
register/deregister statistics (and also has the advantage of
keeping the statistic type, field name, and the 'tag' used to output
the statistic in the same place---instead of scattered in the theory
definition and constructor initializer list. See documentation for
KEEP_STATISTIC in src/util/stats.h for more of an explanation).
* more statistics for UF
* restart notifications from SAT (through TheoryEngine) via
Theory::notifyRestart()
* StackingMap and UnionFind unit tests
* build fixes/adjustments
* code cleanup; minor other improvements
|
|
|
|
done by way of the intermediate branch arith-prop-tmp. Both arith-prop-opt and arith-prop-tmp will now be phased out.
|
|
|
|
util/options.h,cpp
|
|
* Adding options --lazy-type-checking and --eager-type-checking
to control type checking in NodeBuilder, which can now be enabled
in production mode and disabled in debug mode
* Option --no-checking implies --no-type-checking
* Adding constructor SmtEngine(ExprManager* em) that uses default options
|
|
|
|
notifies the theory engine through its output channel); some cleanup; add a regression for bug #216
|
|
supported; work on Result type (biggest noticeable change is that CVC4 now outputs lowercase "sat" and "unsat"), Options class moved to src/smt, to allow for future work on runtime configuration via (set-option) command
|
|
|
|
modified in this commit
|
|
engine code and unit test
|