ai.h

Go to the documentation of this file.

/*******************************************************************\
 
Module: Abstract Interpretation
 
Author: Daniel Kroening, kroening@kroening.com
 
\*******************************************************************/
 
 
#ifndef CPROVER_ANALYSES_AI_H
#define CPROVER_ANALYSES_AI_H
 
#include <iosfwd>
#include <memory>
 
#include <util/deprecate.h>
#include <util/json.h>
#include <util/message.h>
#include <util/xml.h>
 
#include <goto-programs/goto_model.h>
 
#include "ai_domain.h"
#include "ai_history.h"
#include "ai_storage.h"
#include "is_threaded.h"
 
 
class ai_baset
{
public:
  typedef ai_domain_baset statet;
  typedef ai_storage_baset::cstate_ptrt cstate_ptrt;
  typedef ai_history_baset::trace_ptrt trace_ptrt;
  typedef ai_history_baset::trace_sett trace_sett;
  typedef ai_storage_baset::ctrace_set_ptrt ctrace_set_ptrt;
  typedef goto_programt::const_targett locationt;
 
  ai_baset(
    std::unique_ptr<ai_history_factory_baset> &&hf,
    std::unique_ptr<ai_domain_factory_baset> &&df,
    std::unique_ptr<ai_storage_baset> &&st,
    message_handlert &mh)
    : history_factory(std::move(hf)),
      domain_factory(std::move(df)),
      storage(std::move(st)),
      message_handler(mh)
  {
  }
 
  virtual ~ai_baset()
  {
  }
 
  void operator()(
    const irep_idt &function_id,
    const goto_programt &goto_program,
    const namespacet &ns)
  {
    goto_functionst goto_functions;
    initialize(function_id, goto_program);
    trace_ptrt p = entry_state(goto_program);
    if(p != nullptr)
      fixedpoint(p, function_id, goto_program, goto_functions, ns);
    finalize();
  }
 
  void operator()(
    const goto_functionst &goto_functions,
    const namespacet &ns)
  {
    initialize(goto_functions);
    trace_ptrt p = entry_state(goto_functions);
    if(p != nullptr)
      fixedpoint(p, goto_functions, ns);
    finalize();
  }
 
  void operator()(const abstract_goto_modelt &goto_model)
  {
    const namespacet ns(goto_model.get_symbol_table());
    initialize(goto_model.get_goto_functions());
    trace_ptrt p = entry_state(goto_model.get_goto_functions());
    if(p != nullptr)
      fixedpoint(p, goto_model.get_goto_functions(), ns);
    finalize();
  }
 
  void operator()(
    const irep_idt &function_id,
    const goto_functionst::goto_functiont &goto_function,
    const namespacet &ns)
  {
    goto_functionst goto_functions;
    initialize(function_id, goto_function);
    trace_ptrt p = entry_state(goto_function.body);
    if(p != nullptr)
      fixedpoint(p, function_id, goto_function.body, goto_functions, ns);
    finalize();
  }
 
  virtual ctrace_set_ptrt abstract_traces_before(locationt l) const
  {
    return storage->abstract_traces_before(l);
  }
 
  virtual ctrace_set_ptrt abstract_traces_after(locationt l) const
  {
    INVARIANT(!l->is_end_function(), "No state after the last instruction");
    return storage->abstract_traces_before(std::next(l));
  }
 
  virtual cstate_ptrt abstract_state_before(locationt l) const
  {
    return storage->abstract_state_before(l, *domain_factory);
  }
 
  virtual cstate_ptrt abstract_state_after(locationt l) const
  {
    INVARIANT(!l->is_end_function(), "No state after the last instruction");
    return abstract_state_before(std::next(l));
  }
 
  virtual cstate_ptrt abstract_state_before(const trace_ptrt &p) const
  {
    return storage->abstract_state_before(p, *domain_factory);
  }
 
  virtual cstate_ptrt abstract_state_after(const trace_ptrt &p) const
  {
    locationt l = p->current_location();
    INVARIANT(!l->is_end_function(), "No state after the last instruction");
 
    locationt n = std::next(l);
 
    auto step_return = p->step(
      n,
      *(storage->abstract_traces_before(n)),
      ai_history_baset::no_caller_history);
    // Caller history not needed as this is a local step
 
    return storage->abstract_state_before(step_return.second, *domain_factory);
  }
 
  virtual void clear()
  {
    storage->clear();
  }
 
  virtual void output(
    const namespacet &ns,
    const irep_idt &function_id,
    const goto_programt &goto_program,
    std::ostream &out) const;
 
  virtual jsont output_json(
    const namespacet &ns,
    const irep_idt &function_id,
    const goto_programt &goto_program) const;
 
  virtual xmlt output_xml(
    const namespacet &ns,
    const irep_idt &function_id,
    const goto_programt &goto_program) const;
 
  virtual void output(
    const namespacet &ns,
    const goto_functionst &goto_functions,
    std::ostream &out) const;
 
  void output(
    const goto_modelt &goto_model,
    std::ostream &out) const
  {
    const namespacet ns(goto_model.symbol_table);
    output(ns, goto_model.goto_functions, out);
  }
 
  void output(
    const namespacet &ns,
    const goto_functionst::goto_functiont &goto_function,
    std::ostream &out) const
  {
    output(ns, irep_idt(), goto_function.body, out);
  }
 
  virtual jsont output_json(
    const namespacet &ns,
    const goto_functionst &goto_functions) const;
 
  jsont output_json(
    const goto_modelt &goto_model) const
  {
    const namespacet ns(goto_model.symbol_table);
    return output_json(ns, goto_model.goto_functions);
  }
 
  jsont output_json(
    const namespacet &ns,
    const goto_programt &goto_program) const
  {
    return output_json(ns, irep_idt(), goto_program);
  }
 
  jsont output_json(
    const namespacet &ns,
    const goto_functionst::goto_functiont &goto_function) const
  {
    return output_json(ns, irep_idt(), goto_function.body);
  }
 
  virtual xmlt output_xml(
    const namespacet &ns,
    const goto_functionst &goto_functions) const;
 
  xmlt output_xml(
    const goto_modelt &goto_model) const
  {
    const namespacet ns(goto_model.symbol_table);
    return output_xml(ns, goto_model.goto_functions);
  }
 
  xmlt output_xml(
    const namespacet &ns,
    const goto_programt &goto_program) const
  {
    return output_xml(ns, irep_idt(), goto_program);
  }
 
  xmlt output_xml(
    const namespacet &ns,
    const goto_functionst::goto_functiont &goto_function) const
  {
    return output_xml(ns, irep_idt(), goto_function.body);
  }
 
protected:
  virtual void
  initialize(const irep_idt &function_id, const goto_programt &goto_program);
 
  virtual void initialize(
    const irep_idt &function_id,
    const goto_functionst::goto_functiont &goto_function);
 
  virtual void initialize(const goto_functionst &goto_functions);
 
  virtual void finalize();
 
  trace_ptrt entry_state(const goto_programt &goto_program);
 
  trace_ptrt entry_state(const goto_functionst &goto_functions);
 
  typedef trace_sett working_sett;
 
  trace_ptrt get_next(working_sett &working_set);
 
  void put_in_working_set(working_sett &working_set, trace_ptrt t)
  {
    working_set.insert(t);
  }
 
  virtual bool fixedpoint(
    trace_ptrt starting_trace,
    const irep_idt &function_id,
    const goto_programt &goto_program,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  virtual void fixedpoint(
    trace_ptrt starting_trace,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  virtual bool visit(
    const irep_idt &function_id,
    trace_ptrt p,
    working_sett &working_set,
    const goto_programt &goto_program,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  // function calls and return are special cases
  // different kinds of analysis handle these differently so these are virtual
  // visit_function_call handles which function(s) to call,
  // while visit_edge_function_call handles a single call
  virtual bool visit_function_call(
    const irep_idt &function_id,
    trace_ptrt p_call,
    working_sett &working_set,
    const goto_programt &goto_program,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  virtual bool visit_end_function(
    const irep_idt &function_id,
    trace_ptrt p,
    working_sett &working_set,
    const goto_programt &goto_program,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  // The most basic step, computing one edge / transformer application.
  bool visit_edge(
    const irep_idt &function_id,
    trace_ptrt p,
    const irep_idt &to_function_id,
    locationt to_l,
    trace_ptrt caller_history,
    const namespacet &ns,
    working_sett &working_set);
 
  virtual bool visit_edge_function_call(
    const irep_idt &calling_function_id,
    trace_ptrt p_call,
    locationt l_return,
    const irep_idt &callee_function_id,
    working_sett &working_set,
    const goto_programt &callee,
    const goto_functionst &goto_functions,
    const namespacet &ns);
 
  std::unique_ptr<ai_history_factory_baset> history_factory;
 
  std::unique_ptr<ai_domain_factory_baset> domain_factory;
 
  virtual bool merge(const statet &src, trace_ptrt from, trace_ptrt to)
  {
    statet &dest = get_state(to);
    return domain_factory->merge(dest, src, from, to);
  }
 
  virtual std::unique_ptr<statet> make_temporary_state(const statet &s)
  {
    return domain_factory->copy(s);
  }
 
  // Domain and history storage
  std::unique_ptr<ai_storage_baset> storage;
 
  virtual statet &get_state(trace_ptrt p)
  {
    return storage->get_state(p, *domain_factory);
  }
 
  // Logging
  message_handlert &message_handler;
};
 
// Perform interprocedural analysis by simply recursing in the interpreter
// This can lead to a call stack overflow if the domain has a large height
class ai_recursive_interproceduralt : public ai_baset
{
public:
  ai_recursive_interproceduralt(
    std::unique_ptr<ai_history_factory_baset> &&hf,
    std::unique_ptr<ai_domain_factory_baset> &&df,
    std::unique_ptr<ai_storage_baset> &&st,
    message_handlert &mh)
    : ai_baset(std::move(hf), std::move(df), std::move(st), mh)
  {
  }
 
protected:
  // Override the function that handles a single function call edge
  bool visit_edge_function_call(
    const irep_idt &calling_function_id,
    trace_ptrt p_call,
    locationt l_return,
    const irep_idt &callee_function_id,
    working_sett &working_set,
    const goto_programt &callee,
    const goto_functionst &goto_functions,
    const namespacet &ns) override;
};
 
template <typename domainT>
class ait : public ai_recursive_interproceduralt
{
public:
  // constructor
  ait()
    : ai_recursive_interproceduralt(
        std::make_unique<
          ai_history_factory_default_constructort<ahistoricalt>>(),
        std::make_unique<ai_domain_factory_default_constructort<domainT>>(),
        std::make_unique<location_sensitive_storaget>(),
        no_logging)
  {
  }
 
  explicit ait(std::unique_ptr<ai_domain_factory_baset> &&df)
    : ai_recursive_interproceduralt(
        std::make_unique<
          ai_history_factory_default_constructort<ahistoricalt>>(),
        std::move(df),
        std::make_unique<location_sensitive_storaget>(),
        no_logging)
  {
  }
 
  typedef goto_programt::const_targett locationt;
 
  // The older interface for non-modifying access
  // Not recommended as it will throw an exception if a location has not
  // been reached in an analysis and there is no (other) way of telling
  // if a location has been reached.
  DEPRECATED(SINCE(2019, 08, 01, "use abstract_state_{before,after} instead"))
  const domainT &operator[](locationt l) const
  {
    auto p = storage->abstract_state_before(l, *domain_factory);
 
    if(p.use_count() == 1)
    {
      // Would be unsafe to return the dereferenced object
      throw std::out_of_range("failed to find state");
    }
 
    return static_cast<const domainT &>(*p);
  }
 
protected:
  // Support the legacy get_state interface which is needed for a few domains
  // This is one of the few users of the legacy get_state(locationt) method
  // in location_sensitive_storaget.
  DEPRECATED(SINCE(2019, 08, 01, "use get_state(trace_ptrt p) instead"))
  virtual statet &get_state(locationt l)
  {
    auto &s = dynamic_cast<location_sensitive_storaget &>(*storage);
    return s.get_state(l, *domain_factory);
  }
 
  using ai_baset::get_state;
 
private:
  void dummy(const domainT &s) { const statet &x=s; (void)x; }
 
  // To keep the old constructor interface we disable logging
  null_message_handlert no_logging;
};
 
template<typename domainT>
class concurrency_aware_ait:public ait<domainT>
{
public:
  using statet = typename ait<domainT>::statet;
  using locationt = typename statet::locationt;
 
  // constructor
  concurrency_aware_ait():ait<domainT>()
  {
  }
  explicit concurrency_aware_ait(std::unique_ptr<ai_domain_factory_baset> &&df)
    : ait<domainT>(std::move(df))
  {
  }
 
  virtual bool merge_shared(
    const statet &src,
    locationt from,
    locationt to,
    const namespacet &ns)
  {
    statet &dest=this->get_state(to);
    return static_cast<domainT &>(dest).merge_shared(
      static_cast<const domainT &>(src), from, to, ns);
  }
 
protected:
  using working_sett = ai_baset::working_sett;
 
  void fixedpoint(
    ai_baset::trace_ptrt start_trace,
    const goto_functionst &goto_functions,
    const namespacet &ns) override
  {
    ai_baset::fixedpoint(start_trace, goto_functions, ns);
 
    is_threadedt is_threaded(goto_functions);
 
    // construct an initial shared state collecting the results of all
    // functions
    goto_programt tmp;
    tmp.add_instruction();
    goto_programt::const_targett sh_target = tmp.instructions.begin();
    ai_baset::trace_ptrt target_hist =
      ai_baset::history_factory->epoch(sh_target);
    statet &shared_state = ait<domainT>::get_state(sh_target);
 
    struct wl_entryt
    {
      wl_entryt(
        const irep_idt &_function_id,
        const goto_programt &_goto_program,
        locationt _location)
        : function_id(_function_id),
          goto_program(&_goto_program),
          location(_location)
      {
      }
 
      irep_idt function_id;
      const goto_programt *goto_program;
      locationt location;
    };
 
    typedef std::list<wl_entryt> thread_wlt;
    thread_wlt thread_wl;
 
    for(const auto &gf_entry : goto_functions.function_map)
    {
      forall_goto_program_instructions(t_it, gf_entry.second.body)
      {
        if(is_threaded(t_it))
        {
          thread_wl.push_back(
            wl_entryt(gf_entry.first, gf_entry.second.body, t_it));
 
          goto_programt::const_targett l_end =
            gf_entry.second.body.instructions.end();
          --l_end;
 
          merge_shared(shared_state, l_end, sh_target, ns);
        }
      }
    }
 
    // now feed in the shared state into all concurrently executing
    // functions, and iterate until the shared state stabilizes
    bool new_shared = true;
    while(new_shared)
    {
      new_shared = false;
 
      for(const auto &wl_entry : thread_wl)
      {
        working_sett working_set;
        ai_baset::trace_ptrt t(
          ai_baset::history_factory->epoch(wl_entry.location));
        ai_baset::put_in_working_set(working_set, t);
 
        statet &begin_state = ait<domainT>::get_state(wl_entry.location);
        ait<domainT>::merge(begin_state, target_hist, t);
 
        while(!working_set.empty())
        {
          ai_baset::trace_ptrt p = ai_baset::get_next(working_set);
          goto_programt::const_targett l = p->current_location();
 
          ai_baset::visit(
            wl_entry.function_id,
            p,
            working_set,
            *(wl_entry.goto_program),
            goto_functions,
            ns);
 
          // the underlying domain must make sure that the final state
          // carries all possible values; otherwise we would need to
          // merge over each and every state
          if(l->is_end_function())
            new_shared |= merge_shared(shared_state, l, sh_target, ns);
        }
      }
    }
  }
};
 
#endif // CPROVER_ANALYSES_AI_H