mirror of
https://github.com/Z3Prover/z3
synced 2025-08-08 12:11:23 +00:00
Integrate new regex solver (#4602)
* std::cout debugging statements * comment out std::cout debugging as this is now a shared fork * convert std::cout to TRACE statements for seq_rewriter and seq_regex * add cases to min_length and max_length for regexes * bug fix * update min_length and max_length functions for REs * initial pass on simplifying derivative normal forms by eliminating redundant predicates locally * add seq_regex_brief trace statements * working on debugging ref count issue * fix ref count bug and convert trace statements to seq_regex_brief * add compact tracing for cache hits/misses * seq_regex fix cache hit/miss tracing and wrapper around is_nullable * minor * label and disable more experimental changes for testing * minor documentation / tracing * a few more @EXP annotations * dead state elimination skeleton code * progress on dead state elimination * more progress on dead state elimination * refactor dead state class to separate self-contained state_graph class * finish factoring state_graph to only work with unsigned values, and implement separate functionality for expr* logic * implement get_all_derivatives, add debug tracing * trace statements for debugging is_nullable loop bug * fix is_nullable loop bug * comment out local nullable change and mark experimental * pretty printing for state_graph * rewrite state graph to remove the fragile assumption that all edges from a state are added at a time * start of general cycle detection check + fix some comments * implement full cycle detection procedure * normalize derivative conditions to form 'ele <= a' * order derivative conditions by character code * fix confusing names m_to and m_from * assign increasing state IDs from 1 instead of using get_id on AST node * remove elim_condition call in get_dall_derivatives * use u_map instead of uint_map to avoid memory leak * remove unnecessary call to is_ground * debugging * small improvements to seq_regex_brief tracing * fix bug on evil2 example * save work * new propagate code * work in progress on using same seq sort for deriv calls * avoid re-computing derivatives: use same head var for every derivative call * use min_length on regexes to prune search * simple implementation of can_be_in_cycle using rank function idea * add a disabled experimental change * minor cleanup comments, etc. * seq_rewriter cleanup for PR * typo noticed by Nikolaj * move state graph to util/state_graph * re-add accidentally removed line * clean up seq_regex code removing obsolete functions and comments * a few more cleanup items * remove experimental functionality for integration * fix compilation * remove some tracing and TODOs * remove old comment * update copyright dates to 2020 * feedback from Nikolaj * use [] for map access * make state_graph methods constant * avoid recursion in mark_dead_recursive and mark_live_recursive * a possible bug fix in propagate_nonempty * write down list of invariants in state_graph * implement partial invariant check and insert CASSERT statements * expand on invariant check and tracing * finish state graph invariant check * minor tweaks * regex propagation: convert first two axioms to propagations * remove obsolete regex solver functionality Co-authored-by: calebstanford-msr <t-casta@microsoft.com>
This commit is contained in:
Caleb Stanford
GitHub
parent
293b0b8cc2
commit
976e4c91b0
8 changed files with 922 additions and 257 deletions
182
src/util/state_graph.h
Normal file
182
src/util/state_graph.h
Normal file
|
|
@ -0,0 +1,182 @@
|
|||
/*++
|
||||
Copyright (c) 2020 Microsoft Corporation
|
||||
|
||||
Module Name:
|
||||
|
||||
state_graph.h
|
||||
|
||||
Abstract:
|
||||
|
||||
Data structure for incrementally tracking "live" and "dead" states in an
|
||||
abstract transition system.
|
||||
|
||||
Author:
|
||||
|
||||
Caleb Stanford (calebstanford-msr / cdstanford) 2020-7
|
||||
|
||||
--*/
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "util/map.h"
|
||||
#include "util/uint_set.h"
|
||||
#include "util/union_find.h"
|
||||
#include "util/vector.h"
|
||||
|
||||
/*
|
||||
state_graph
|
||||
|
||||
Data structure which is capable of incrementally tracking
|
||||
live states and dead states.
|
||||
|
||||
"States" are integers. States and edges are added to the data
|
||||
structure incrementally.
|
||||
- States can be marked as "live" or "done".
|
||||
"Done" signals that (1) no more outgoing edges will be
|
||||
added and (2) the state will not be marked as live. The data
|
||||
structure then tracks
|
||||
which other states are live (can reach a live state), dead
|
||||
(can't reach a live state), or neither.
|
||||
- Some edges are labeled as not contained in a cycle. This is to
|
||||
optimize search if it is known by the user of the structure
|
||||
that no cycle will ever contain this edge.
|
||||
|
||||
Internally, we use union_find to identify states within an SCC,
|
||||
and incrementally update SCCs, while propagating backwards
|
||||
live and dead SCCs.
|
||||
*/
|
||||
class state_graph {
|
||||
public:
|
||||
typedef unsigned state;
|
||||
typedef uint_set state_set;
|
||||
typedef u_map<state_set> edge_rel;
|
||||
typedef basic_union_find state_ufind;
|
||||
|
||||
private:
|
||||
/*
|
||||
All states are internally exactly one of:
|
||||
- live: known to reach a live state
|
||||
- dead: known to never reach a live state
|
||||
- unknown: all outgoing edges have been added, but the
|
||||
state is not known to be live or dead
|
||||
- unexplored: not all outgoing edges have been added
|
||||
|
||||
As SCCs are merged, some states become aliases, and a
|
||||
union find data structure collapses a now obsolete
|
||||
state to its current representative. m_seen keeps track
|
||||
of states we have seen, including obsolete states.
|
||||
*/
|
||||
state_set m_live;
|
||||
state_set m_dead;
|
||||
state_set m_unknown;
|
||||
state_set m_unexplored;
|
||||
|
||||
state_set m_seen;
|
||||
state_ufind m_state_ufind;
|
||||
|
||||
/*
|
||||
Edges are saved in both from and to maps.
|
||||
A subset of edges are also marked as possibly being
|
||||
part of a cycle by being stored in m_sources_maybecycle.
|
||||
*/
|
||||
edge_rel m_sources;
|
||||
edge_rel m_targets;
|
||||
edge_rel m_sources_maybecycle;
|
||||
|
||||
/*
|
||||
CLASS INVARIANTS
|
||||
|
||||
*** To enable checking invariants, run z3 with -dbg:state_graph
|
||||
(must also be in debug mode) ***
|
||||
|
||||
State invariants:
|
||||
- live, dead, unknown, and unexplored form a partition of
|
||||
the set of roots in m_state_ufind
|
||||
- all of these are subsets of m_seen
|
||||
- everything in m_seen is an integer less than the number of variables
|
||||
in m_state_ufind
|
||||
|
||||
Edge invariants:
|
||||
- all edges are between roots of m_state_ufind
|
||||
- m_sources and m_targets are converses of each other
|
||||
- no self-loops
|
||||
- m_sources_maybecycle is a subrelation of m_sources
|
||||
|
||||
Relationship between states and edges:
|
||||
- every state with a live target is live
|
||||
- every state with a dead source is dead
|
||||
- every state with only dead targets is dead
|
||||
- there are no cycles of unknown states on maybecycle edges
|
||||
*/
|
||||
#ifdef Z3DEBUG
|
||||
bool is_subset(state_set set1, state_set set2) const;
|
||||
bool is_disjoint(state_set set1, state_set set2) const;
|
||||
bool check_invariant() const;
|
||||
#endif
|
||||
|
||||
/*
|
||||
'Core' functions that modify the plain graph, without
|
||||
updating SCCs or propagating live/dead state information.
|
||||
These are for internal use only.
|
||||
*/
|
||||
void add_state_core(state s); // unexplored + seen
|
||||
void remove_state_core(state s); // unknown + seen -> seen
|
||||
void mark_unknown_core(state s); // unexplored -> unknown
|
||||
void mark_live_core(state s); // unknown -> live
|
||||
void mark_dead_core(state s); // unknown -> dead
|
||||
|
||||
void add_edge_core(state s1, state s2, bool maybecycle);
|
||||
void remove_edge_core(state s1, state s2);
|
||||
void rename_edge_core(state old1, state old2, state new1, state new2);
|
||||
|
||||
state merge_states(state s1, state s2);
|
||||
state merge_states(state_set& s_set);
|
||||
|
||||
/*
|
||||
Algorithmic search routines
|
||||
- live state propagation
|
||||
- dead state propagation
|
||||
- cycle / strongly-connected component detection
|
||||
*/
|
||||
void mark_live_recursive(state s);
|
||||
bool all_targets_dead(state s);
|
||||
void mark_dead_recursive(state s);
|
||||
state merge_all_cycles(state s);
|
||||
|
||||
public:
|
||||
state_graph():
|
||||
m_live(), m_dead(), m_unknown(), m_unexplored(), m_seen(),
|
||||
m_state_ufind(), m_sources(), m_targets(), m_sources_maybecycle()
|
||||
{
|
||||
CASSERT("state_graph", check_invariant());
|
||||
}
|
||||
|
||||
/*
|
||||
Exposed methods
|
||||
|
||||
These methods may be called in any order, as long as:
|
||||
- states are added before edges are added between them
|
||||
- outgoing edges are not added from a done state
|
||||
- a done state is not marked as live
|
||||
- edges are not added creating a cycle containing an edge with
|
||||
maybecycle = false (this is not necessary for soundness, but
|
||||
prevents completeness for successfully detecting dead states)
|
||||
*/
|
||||
void add_state(state s);
|
||||
void add_edge(state s1, state s2, bool maybecycle);
|
||||
void mark_live(state s);
|
||||
void mark_done(state s);
|
||||
|
||||
bool is_seen(state s) const;
|
||||
bool is_live(state s) const;
|
||||
bool is_dead(state s) const;
|
||||
bool is_done(state s) const;
|
||||
|
||||
unsigned get_size() const;
|
||||
|
||||
/*
|
||||
Pretty printing
|
||||
*/
|
||||
std::ostream& display(std::ostream& o) const;
|
||||
|
||||
};
|
||||
Loading…
Add table
Add a link
Reference in a new issue