| Age | Commit message (Collapse) | Author |
|---|
|
Currently, NodeBuilder takes a single template argument: An integer
that determines the expected number of arguments. This argument is used
to determine the size of the d_inlineNvChildSpace array. This array is
used to construct nodes inline. The advantage of this is that we don't
have to allocate a NodeValue on the heap for the node under
construction until we are sure that the node is new. While templating
the array size may save some stack space (or avoid a heap allocation if
we statically know that we a fixed number of children and that number is
greater than 10), it complicates the code and leads to longer compile
times. Thus, this commit removes the template argument and moves some of
the NodeBuilder code to a source file for faster compilation.
CPU build time before change (debug build): 2429.68s
CPU build time after change (debug build): 2228.44s
Signed-off-by: Andres Noetzli noetzli@amazon.com
|
|
|
|
|
|
This ensures we do not return WITNESS terms in response to ppRewrite. This makes standard utility methods in SkolemLemma to help make this easy.
It also removes TheorySetsPrivate::expandDefinitions, which was returning WITNESS in response to expandDefinitions, which is no longer permitted.
|
|
This is required for proofs. The internal calculus no longer uses witness forms for reasoning, and hence we cannot return witness terms in ppRewrite. This is required to fix a debug failure, as well as making life easier on external proof conversions.
As a result of this PR, for example, given (P a) as input to ppRewrite, previous we returned:
(P (witness ((x T)) (A x)))
now we return:
(P k), with skolem lemma ( (A k), k )
Followup PRs will remove the use of WITNESS in ppRewrite (e.g. in sets and strings); this PR modifies arithmetic to not return WITNESS in response to ppRewrite.
|
|
This PR does some more cleanup of the includes.
|
|
|
|
Theory preprocessing now theory-preprocesses lemmas until fixed point. This eliminates the old code for rewriting them only, which is no longer necessary as theory-preprocessing subsumes rewriting.
|
|
This simplifies our algorithm for theory preprocessing. The motivation is two-fold:
(1) Proofs are currently incorrect in rare cases due to incorrectly tracking pre vs post rewrites.
(2) We can avoid traversing Boolean connectives with term formula removal. Note that this PR relies on the observation that term formula removal leaves Boolean structure unchanged and can apply locally to theory atoms.
Furthermore, notice that our handling of lemmas generated by term formula removal is now more uniform. In the second algorithm, term formula removal is not applied to fixed point anymore. In other words, we do not remove skolems from lemmas generated by term formula removal recursively. Instead, since lemmas are theory-preprocessed recursively, we use the fixed point that runs outside the above algorithm instead of relying on the inner fixed point in term formula removal.
This PR resolves several open issues with proof production on proof-new.
I will performance test this change on SMT-LIB.
|
|
generator (#5833)
This is work towards resolving two kinds of failures on proof-new:
(1) Functional issues with proofs from the rewriter, in particular when a term pre-rewrites and post-rewrites to different things,
(2) Conversion issues in theory-preprocessing, where all steps are assumed to be post-rewrites but some are in fact pre-rewrites. This leads to the term conversion proof generator proving something different than what is expected.
A followup PR will simplify and fix proofs for theory-preprocessing.
|
|
This PR makes it so that theory-preprocessing is always called on lemmas. It simplifies the proof production in the theory preprocessor accordingly.
Additionally, it ensures that our theory-preprocessor is run on lemmas that are generated from term formula removal. Previously, this was not the case and in fact certain lemmas (e.g. literals within witness terms that are not in preprocessed form) would escape and be asserted to TheoryEngine. This was uncovered by a unit test failure, the corresponding regression is added in this PR.
It adds a new interface removeItes to PropEngine which is required for the (deprecated) preprocessing pass removeItes.
This PR now makes the lemma propery PREPROCESS obsolete. Further simplification is possible after this PR in non-linear arithmetic and quantifiers, where it is not necessary to distinguish 2 caches for preprocessed vs. non-preprocessed lemmas.
|
|
This is in preparation for lazy lemmas for term formula removal.
|
|
Some theories rewrite equalities during ppRewrite. An example is arithmetic with the option arith-rewrite-eq, which rewrites (= x y) to (and (>= x y) (<= x y)) during theory preprocessing.
This PR makes it so that ppRewrite is only called on equalities in preprocessing, during a new preprocessing pass "TheoryRewriteEq". On the other hand, ppRewrite is never called on new equalities generated in lemmas from TheoryEngine.
In detail, the motivation for this change:
(1) Rewriting equalities during ppRewrite is dangerous since it may break invariants wrt theory combination. In particular, equalities in splitting lemmas originating from theory combination must not be theory-preprocessed, or else we may be non-terminating or solution unsound. This can happen if a theory requests a split on (= x y) but is not notified of this atom when another theory rewrites (= x y) during ppRewrite.
(2) After this PR, we can simplify our policy for all lemmas generated, in particular, we can say that all lemmas must be theory preprocessed before their literals are asserted to TheoryEngine. This is now possible as the invariant cannot be broken (theoryRewriteEq is relegated to the preprocessor, which is only applied once). This will make LemmaProperty::PREPROCESS obsolete, which in turn will simplify several lemma caches for nonlinear and quantifiers. It will also significantly simplify proof production for the theory preprocessor (which maintains two stacks of utilities for preprocessed vs non-preprocessed lemmas).
(3) Simplifications to the above policy will make it significantly easier to implement theory-preprocessing apply when literals are asserted. It is currently not possible to implement this in a coherent way without tracking which literals were a part of lemmas marked as "do not theory-preprocess".
|
|
This makes all recursion (applying term formula removal on lemmas introduced by term formula removal) optionally happen at the top level call.
This is in preparation for making term formula removal lazier, in which case we will only apply one step of term formula removal at a time.
One QF_UFNIA regression times out due to changing the search, an option is changed for this benchmark.
|
|
This is work towards refactoring ITE removal (more generally, term formula removal) so that it happens at a configurable time, preferably post-CNF conversion.
This moves the TermFormulaRemover to the TheoryPreprocessor and changes several interfaces as a consequence of this move.
The next step will move the TheoryPreprocessor inside prop::TheoryProxy.
There are no behavior changes to solving in this PR. One aspect of CheckModels is simplified.
|
|
Previously, there was a block of code in TheoryEngine::lemma for processing the proofs from theory preprocessing.
It is much cleaner if this block was self contained in TheoryPreprocessor.
This is required for the planned move of TheoryPreprocessor from TheoryEngine -> PropEngine.
|
|
|
|
This fixes a rare case of theory preprocessing where rewriting a quantified formula would introduce a term that would not be preprocessed. This led to solution unsoundness on a branch where the relationship between expand definitions and preprocessing is modified.
This is required for updating to the new policy for expand definitions.
|
|
Now that extended arithmetic symbols are not eliminated during expandDefinitions, this allowed for a chain of events that would not apply theory preprocessing on certain terms.
In particular, a term t would not have theory preprocessing applied to it if it was a strict subterm of ITE-term s that occurred in a term position in a quantified formula body, and s did not have free variables. In this case, term formula removal would add a lemma corresponding to the ITE skolem definition, whose subterms did not have theory preprocessing applied. This meant that a (div a d) term was not being preprocessed in #5482, leading to solution unsoundness.
This changes the policy in term formula removal to be in sync with theory preprocessing: term formula removal and theory preprocessing only apply to terms that are not beneath quantifiers. In particular, this means that ground terms in quantifier bodies are left untouched until they are introduced e.g. by instantiation.
This fixes the solution soundness issue (fixes #5482).
|
|
This fixes 2 issues related to eliminating arithmetic operators:
(1) counterexample lemmas in CEGQI were not being preprocessed
(2) ensureLiteral was not doing term formula removal.
This corrects these two issues. Now ensureLiteral does full theory preprocessing on the term we are ensuring literal for, meaning this may trigger lemmas if the term contains extended arithmetic operators like div.
Fixes #5469, fixes #5470, fixes #5471.
This adds --sygus-inst to 2 of these benchmarks which moreover makes them solvable.
This also improves our assertions and trace messages.
|
|
Previously, theory preprocessing cache was manually cleared whenever the theory preprocess pass was run. However, proofs for theory preprocessing required to be alive for the remainder of the user context. This PR changes theory preprocessing so that both the cache and proofs in theory preprocessing are user-context dependent.
This PR also makes the theory preprocess pass proof producing.
|
|
Ensuring closed proofs should not be enabled by default, it is actually not used very often as a whole. Moreover, the "trust id" argument is the most useful argument and hence should come as the 3rd argument. This updates all uses of addLazyStep for the change in interface, also changes term conversion generator which had a similar issue with default arguments.
Notice that some calls to addLazyStep were checking closed but without providing a debug string, these I've left alone (they no longer check closed).
|
|
files. (#5241)
|
|
This fixes several subtle issues with theory preprocessing.
The main fix is to ensure proofs are correct for cases where the theory preprocessor applies either with or without theory preprocessing (calling Theory::ppRewrite on subterms) enabled, see argument doTheoryPreprocess of preprocess. If disabled, it applies term formula removal and rewriting only. This is required for processing lemmas that are marked as "do not preprocess", which is an optimization, but is also necessary since theory combination may e.g. split on equality that was solved during ppRewrite. The solution is to use 2 separate term conversion sequences for these 2 cases.
It also fixes an issue where preprocessing is term-context-sensitive: terms underneath quantifiers are treated differently. This is now handled by a new TermContext callback "InQuantTermContext".
|
|
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 makes the theory preprocessor proof producing in the new infrastructure, which involves updating its interfaces to TrustNode in a few places.
|
|
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 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.
|
|
|
|
This PR splits two classes from TheoryEngine (EngineOutputChannel and TheoryPreprocess) that will be undergoing further refactoring for proofs.
This PR does not change their behavior and is code-move only modulo a few cosmetic changes.
|
