3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2025-04-06 17:44:08 +00:00

Reorganizing the code

Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
This commit is contained in:
Leonardo de Moura 2012-10-20 22:28:22 -07:00
parent 492484c5aa
commit aa949693d4
13 changed files with 0 additions and 78 deletions

View file

@ -1,78 +0,0 @@
The assertion_set class is essentially a set of formulas.
The assertion_set copy() method is constant time. They use persistent
arrays for "sharing" common parts.
The goal is to provide functors (strategies) that apply transformations to these sets.
Examples of transformations are:
- simplification
- gaussian elimination
- CNF, NNF, OCNF transformation
- bit-blasting
- abstraction
- grobner basis computation
- completion
- term rewriting
- bound propagation
- contextal simplification
- if-then-else lifting
- quantifier elimination
- etc
A functor/strategy is essentially updating an assertion set. It may
also provide a model_converter to convert a model for the resultant
assertion_set into a model for the original assertion_set. See
gaussian_elim.cpp for an example of model_converter. A functor may
return a vector of assertion sets. It may also return a
proof_converter for converting proofs for the resultant assertion
set(s) into a proof for the original assertion_set. The
functor/strategy user is responsible for "plumbing" the
model_converter(s) and proof_converter(s). In the future we may have
tacticals for combining these functors/strategies. This is not needed
now.
The idea is to provide assertion_set_solvers. They are essentially end-game
strategies. The abstract API for them is defined at assertion_set_solver.h.
Currently, the only assertion_set_solver is based on smt_context.h.
The next one will be based on a pure SAT solver. I also have plans for
one based on linear programming.
The main goals of the new design are:
- increase modularity.
- expose assertion_set abstraction to external users.
- expose these functor/strategies to external users.
- allow us to combine problem specific solvers in a modular way.
- everything can be interrupted.
- some actions may be resumed.
- clean parallel_z3 implementation.
- support non-satisfiability preserving transformations.
the basic idea is:
* if the transformation is an over-approximation,
then the model_converter must throw an exception if it
can't convert the model into a model for the original assertion_set.
* if the transformation is an under-approximation,
then the proof_converter must throw an exception if it
can't convert the proof(s) into a proof for the original assertion_set.
I don't expect us to provide extensive support for proofs.
So, under-approximations will never really be used to show that
an assertion_set is unsatisfiable.
Another goal is to a solver object that is essentially invoking an
external solver (process). I expect the external solver (in most cases) to be
Z3 itself. The main advantages are: the main process is safe from crashes,
and we can even invoke solvers in remote machines.
The new functor/strategy design is not meant for incremental solving.
We may still have solvers that are incremental such as smt_context.h.
Additional remarks:
Failures and interruptions are reported using exceptions.
Each functor must provide a cancel() method that can be invoked to
interrupt it. A resume() method is optional. It is a must have if the
functor/strategy may be too time consuming. Each functor may have its
own set of parameters. See rewriter_params.h for an example. The
parameters may have options such as m_max_memory for blocking the
executiong of the functor. Each functor should have a cleanup()
method that must reclaim (almost) all memory consumed by the functor.