| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
Currently, SEXPR applications are given a parametric type SEXPR_TYPE applied to the types of its arguments. This means that SEXPR that are type checked consume roughly double the memory. This issue arises in practice when printing proofs in the internal calculus.
There is no need to have SEXPR_TYPE as a parametric type, this PR makes SEXPR simply typed.
Also moves some implementation of TypeNode methods to type_node.cpp.
|
|
This commit refactors the run_regression.py script and adds options for enabling/disabling checking of proofs and unsat cores. Both options are enabled by default and disabled for each corresponding CI build.
|
|
calculus (#6051)
This makes a simplification to the internal proof calculus. In particular, purification skolems are no longer are special case of witness skolems. They are now independent concepts. The concept of "witness form" is replaced in most places by "original form".
This is required for fixing two issues:
(1) variable shadowing issues in skolemization.
(2) bookkeeping issues for bound variables introduced to construct witness terms. This made the LFSC proof conversion extremely cumbersome for e.g. string reductions.
In this PR, the main changes:
The internals of SkolemManager are changed to use original form vs witness form when necessary. This eliminates the need to do variable renaming in SkolemManager::skolemize.
the rule WITNESS_INTRO is replaced by SKOLEM_INTRO
MACRO_SR_* rules use original form
Proof post processing is simplified
These changes imply that ppRewrite should not return WITNESS terms. Followup PRs will modify arithmetic preprocessing so that its ppRewrite method returns skolems instead.
|
|
This is in preparation for enabling CI / proofs on master.
This does not throw an option exception when proofs are enabled, it also makes a fix that was missing on master and needed for regressions to pass on master.
This is partially taken from #5980.
|
|
This PR does some more cleanup of the includes.
|
|
|
|
This deletes variable flags from NodeManager::mkVar and moves ExprManager sort flags to NodeManager.
These flags are used for determining when a variable or sort should be printed via the old dump infrastructure. The old dump infrastructure is simplified in this PR accordingly.
This PR should preserve behavior of the previous dumping with a minor exception that the internal trace "declarations" will also included symbols introduced from define-fun. This will be further refactored later. This is in preparation for removing the includes expr.h/expr_manager.h from node_manager.h.
|
|
|
|
|
|
Ensures (witness ((x Int)) (= x (+ 1 a)) is rewritten to (+ 1 a), instead of e.g. (witness ((x Int)) (= a (- x 1)).
This is required for supporting purification skolems for arithmetic terms in the new proof architecture.
|
|
This ensures mkGroundTerm is deterministic using attributes.
This was uncovered by failures in the proof checker for datatypes.
|
|
This PR fixes proofs in the rewriter so that we use attributes for marking whether a node has been rewritten with proofs instead of consulting the provided TConvProofGenerator for hasRewriteStep. This additionally marks TRUST_REWRITE steps if a rewriter happens to be nondeterministic (e.g. quantifiers). It furthermore decouples rewriteWithProof from reconstruction for theory rewrite steps.
The proof postprocessor is additionally updated so that extended equality REWRITE steps are converted to THEORY_REWRITE steps with identifier RW_REWRITE_EQ_EXT for consistency, since eliminating REWRITE should result in THEORY_REWRITE only. This required generalizing the argument to THEORY_REWRITE to be a MethodId instead of a Boolean.
|
|
Mostly in cardinality extension, which was cleaned in the previous PR.
|
|
Trust substitution is a data structure that is used throughout preprocessing for proofs.
This adds its implementation.
|
|
This updates the SMT proof postprocessor with additional stats. It also adds the feature to handle conjunctions as substitution, e.g. a child proof concluding (and (= x t) (= y s)) is interpreted as a substitution for x, y, whereas previously we insisted they must be provided separately.
|
|
Some of the interfaces in the proof rule checker are unnecessary now and can be deleted.
This also updates STRING_TRUST to have pedantic level 2.
|
|
This includes several subtle issues encountered in the past month based on our regressions.
It also improves the expansion forms of MACRO_ rules to use EQ_RESOLVE instead of a TRUE_INTRO/TRUE_ELIM scheme. This is both more compact and avoids cyclic proofs for some corner cases of proofs with Boolean constant equalites.
|
|
This PR updates the update-copyright.pl script to also update/add copyright headers to CMake specific files. It further fixes a small typo in the header.
|
|
This includes support for some trusted rules (whose use is to come). It also modifies THEORY_REWRITE so that it is a "trusted" rule, that is, it is not re-checked. The reason for this is that TheoryRewriter is not deterministic. An example of non-determinism in TheoryRewriter are rules that introduce bound variables (e.g. quantifiers rewriter) since these bound variables are fresh each time it is invoked. Non-deterministic theory rewriters cannot be rechecked, which was leading to failures on proof-new at proof check time. The other way to fix this would be to cache TheoryRewriter responses, but then the checker would only be verifying that the caching was being done properly. This is not worthwhile.
|
|
This makes the theory preprocessor proof producing in the new infrastructure, which involves updating its interfaces to TrustNode in a few places.
|
|
This class is responsible for the connection between terms and their witness form in the final proof.
|
|
This adds proof support in the term formula removal pass.
It also refactors this class heavily so that its interface is more intuitive and does not depend on AssertionPipeline.
|
|
This commit changes UninterpretedConstant to use TypeNode instead of
Type.
|
|
This adds support for skeleton proofs in the rewriter (REWRITE -> THEORY_REWRITE).
It adds "extended equality rewrite" as a new method of the rewriter/theory rewriters.
The unit test of this feature should be added on a followup PR.
Co-authored-by: Andres Noetzli <andres.noetzli@gmail.com>
|
|
This commit changes ArrayStoreAll to use Node/TypeNode instead of
Expr/Type.
|
|
This adds the proof generator for storing proofs of preprocessing. It stores assertions added via preprocessing passes and their rewrites. This utility will eventually live in SmtEngine.
It also adds 2 new proof rules, and fixes an issue in ProofNodeUpdater.
|
|
This includes 4 changes:
Theory constructor takes a ProofNodeManager,
Theory::explain returns a TrustNode (of kind PROP_EXP),
Theory::expandDefinitions returns a TrustNode (of kind REWRITE),
Theory::ppRewrite returns a TrustNode (of kind REWRITE).
These are all currently planned updates to the interface of Theory.
This PR also connects some of the existing proof rule checkers into the proof checker, if one is provided to the constructor. It updates TheoryEngine and other places to process TrustNode in trivial ways (converting them into Node). These calls will later be updated as needed for proof support.
This PR is also contingent on the performance tests for proof-new on SMT-LIB.
|
|
Fixes #4685.
A recent commit #4661 added assertions for checking whether a witness rewrite corresponded to a legal elimination. #4685 demonstrates that these assertions can be violated and hence should be checked to ensure the rewrite is sound.
|
|
Previously, the proof rule checkers were run on witness form for convenience. However, it is more flexible for the sake of the internal calculus to run on Skolem forms. The main reason is that the conversion to/from witness form does not preserve terms that contain witness. Our preprocessing steps rely on using witness terms. This design change additionally simplifies a lot of the code of the builtin proof rule checker.
Instead, witness forms can be queried on an as-needed basis, e.g. in MACRO_SR_PRED_TRANSFORM.
|
|
Proof checking failures revealed that we are not rewriting witness for Boolean variables (witness ((x Bool)) x) ---> true and (witness ((x Bool)) (not x)) ---> false.
Also adds 2 assertions that are required for elimination (witness ((x T)) (= x t)) ---> t. These assertions should always hold due to the witness terms we construct.
|
|
|
|
This makes the conversion between lambdas and arrays, done in the theory builtin rewriter and used in higher-order model construction, robust to function definitions in terms of AND/OR.
This also improves tracing and makes a few parts of the code adhere to code guidelines.
|
|
This is a common class for adding steps for theory-specific proof rules into a ProofStepBuffer.
|
|
This PR changes the design of ProofSkolemCache so that it has facilities for tracking proofs. This is required so that the term formula removal pass can be made proof-producing.
This makes it so that ProofSkolemCache is renamed to SkolemManager, which lives in NodeManager. Creating (most) skolems must be accompanied by a corresponding ProofGenerator that can provide the proof for the existential corresponding to their witness form.
Further updates will include refactoring the mkSkolemExists method of SkolemManager so that its contract wrt proofs is simpler.
Once all calls to mkSkolem go through the standard interface, then NodeManager::mkSkolem will be moved to SkolemManager::mkSkolemInternal.
|
|
This adds the proof checker for TheoryBuiltin, which handles the core proof rules (SCOPE and ASSUME) and all proof rules corresponding to generic operations on Node objects. This includes proof rules for rewriting and substitution, which are added in this PR.
|
|
When we eliminate a variable x -> v during simplification, it may be the case that v contains "unevaluated" operators like forall, choice, etc. Thus, we do not produce correct models for such inputs unless simplification is disabled.
This PR ensures we only eliminate variables when v contains only evaluated operators.
Additionally, the kinds registered as unevaluated were slightly modified so that when we are in a logic like QF_LIA, there are no registered unevaluated operators, hence the check above is unnecessary. This is to minimize the performance impact of this change.
Fixes #4500.
|
|
Renamed operator CHOICE to WITNESS, and removed it from the front end
|
|
This commit adds statistics for string rewrites. This is work towards proof
support in the string solver. At a high level, this commit adds a pointer to a
`SequenceStatistics` in the rewriters and modifies
`SequencesRewriter::returnRewrite()` to count the rewrites done. In practice,
to make this work requires a couple of changes, some of them temporary:
- We can't have a single `Rewriter` instance shared between different
`SmtEngine` instances anymore. Thus the `Rewriter` is now owned by the
`SmtEngine` and calling the rewriter retrieves the rewriter associated with
the current `SmtEngine`. This is a temporary workaround before we get rid of
singletons.
- Methods in the `SequencesRewriter` and the `StringsRewriter` are made
non-`static` because they need access to the statistics instance.
- `StringsEntail` now has non-`static` methods because it needs a reference to
the sequences rewriter that it can call.
- The interaction between the `StringsRewriter` and the `SequencesRewriter`
changed: the `StringsRewriter` is now a proper `TheoryRewriter` that inherits
from `SequencesRewriter` and calls its `postRewrite()` before applying its
own rewrites (this is essentially a reversal of roles from before: the
`SequencesRewriter` used to call `static` methods in the `StringsRewriter`).
- The theory rewriters are now owned by the individual theories. This design
mirrors the `EqualityEngine`s owned by the individual theories.
|
|
Fixes #4170.
|
|
Until now, the `Rewriter` was responsible for creating `TheoryRewriter`
instances. This commit adds a method `mkTheoryRewriter()` that theories
override to create an instance of their corresponding theory rewriter.
The advantage is that the theories can pass additional information to
their theory rewriter (e.g. a statistics object).
|
|
Added the operator choice to Smt2.g and Cvc.g.
Removed the unused parameter hasBoundVars from TheoryModel::getModelValue
|
|
This standardizes the interface for using SMT engines as subsolvers in various approaches. More refactoring is possible, but this is an initial cut at cleaning things up.
This will make it easy to accommodate new feature request for SyGuS (timeouts for calls to verification steps).
Notice this also required adding a missing function (mkGroundTerm/isWellFounded for functions) which was caught after standardizing due to an optimization (don't create SmtEngines to check satisfiability of constant Booleans).
|
|
The lambda rewriter was not robust to the case where the lambda of the
array representation contained a disequality, e.g. `not(x = 1))`. It
would process it as `ite(not(x = 1), true, false)` instead of `ite(x =
1, false, true)`, which meant that it wasn't able to turn it into an
array representation when checking const-ness. Additionally, the
rewriter had issues when the lambda was of the form `ite((= x c1), true,
(= y c2))` (after turning it into an array and then into a lambda)
because it is expecting the false branch of the `ite` to not contain `y`
variables, making it non-constant despite the array being constant. This
commit solves that issue by normalizing `ite(not(c), x, y) ---> ite(c,
y, x)`.
|
|
|
|
Fixes #3536. The type checker for the chain operator was calling the
rewriter. However, the floating-point rewriter was expecting
`TheoryFp::expandDefinition()` to be applied before rewriting. If the
chain operator had subterms that were supposed to be removed by
`TheoryFp::expandDefinition()`, the FP rewriter was throwing an
exception. This commit fixes the issue by not calling the full rewriter
in the type checker but by just expanding the chain operator. This is a
bit less efficient than before because the rewriter does not cache the
result of expanding the chain operator anymore but assuming that there
are no long chains, the performance impact should be negligible. It also
seemed like a reasonable assumption that the rewriter can expect to run
after `expandDefinition()` because otherwise the rewriter has to expand
definitions, which may be too restrictive.
|
|
This commit changes theory rewriters to be non-static. This refactoring
is needed as a stepping stone to making our rewriter configurable: If we
have multiple solver objects with different rewrite configurations, we
cannot use `static` variables for the rewriter table in the BV rewriter
for example. It is also in line with our goal of getting rid of
singletons in general. Note that the `Rewriter` class is still a
singleton, which will be changed in a future commit.
|
|
|
|
|
