llvm-project/clang/lib/Analysis/RegionStore.cpp

1847 lines
64 KiB
C++
Raw Normal View History

//== RegionStore.cpp - Field-sensitive store model --------------*- C++ -*--==//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines a basic region store model. In this model, we do have field
// sensitivity. But we assume nothing about the heap shape. So recursive data
// structures are largely ignored. Basically we do 1-limiting analysis.
// Parameter pointers are assumed with no aliasing. Pointee objects of
// parameters are created lazily.
//
//===----------------------------------------------------------------------===//
#include "clang/Analysis/PathSensitive/MemRegion.h"
#include "clang/Analysis/PathSensitive/AnalysisContext.h"
#include "clang/Analysis/PathSensitive/GRState.h"
#include "clang/Analysis/PathSensitive/GRStateTrait.h"
#include "clang/Analysis/Analyses/LiveVariables.h"
#include "clang/Analysis/Support/Optional.h"
#include "clang/Basic/TargetInfo.h"
#include "llvm/ADT/ImmutableMap.h"
#include "llvm/ADT/ImmutableList.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Support/Compiler.h"
using namespace clang;
#define HEAP_UNDEFINED 0
#define USE_EXPLICIT_COMPOUND 0
// Actual Store type.
typedef llvm::ImmutableMap<const MemRegion*, SVal> RegionBindings;
//===----------------------------------------------------------------------===//
// Fine-grained control of RegionStoreManager.
//===----------------------------------------------------------------------===//
namespace {
struct VISIBILITY_HIDDEN minimal_features_tag {};
struct VISIBILITY_HIDDEN maximal_features_tag {};
class VISIBILITY_HIDDEN RegionStoreFeatures {
bool SupportsFields;
bool SupportsRemaining;
public:
RegionStoreFeatures(minimal_features_tag) :
SupportsFields(false), SupportsRemaining(false) {}
RegionStoreFeatures(maximal_features_tag) :
SupportsFields(true), SupportsRemaining(false) {}
void enableFields(bool t) { SupportsFields = t; }
bool supportsFields() const { return SupportsFields; }
bool supportsRemaining() const { return SupportsRemaining; }
};
}
//===----------------------------------------------------------------------===//
// Region "Extents"
//===----------------------------------------------------------------------===//
//
// MemRegions represent chunks of memory with a size (their "extent"). This
// GDM entry tracks the extents for regions. Extents are in bytes.
//
namespace { class VISIBILITY_HIDDEN RegionExtents {}; }
static int RegionExtentsIndex = 0;
namespace clang {
template<> struct GRStateTrait<RegionExtents>
: public GRStatePartialTrait<llvm::ImmutableMap<const MemRegion*, SVal> > {
static void* GDMIndex() { return &RegionExtentsIndex; }
};
}
//===----------------------------------------------------------------------===//
// Regions with default values.
//===----------------------------------------------------------------------===//
//
// This GDM entry tracks what regions have a default value if they have no bound
// value and have not been killed.
//
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
namespace {
class VISIBILITY_HIDDEN RegionDefaultValue {
public:
typedef llvm::ImmutableMap<const MemRegion*, SVal> MapTy;
};
}
static int RegionDefaultValueIndex = 0;
namespace clang {
template<> struct GRStateTrait<RegionDefaultValue>
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
: public GRStatePartialTrait<RegionDefaultValue::MapTy> {
static void* GDMIndex() { return &RegionDefaultValueIndex; }
};
}
typedef RegionDefaultValue::MapTy RegionDefaultBindings;
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
//===----------------------------------------------------------------------===//
// Utility functions.
//===----------------------------------------------------------------------===//
static bool IsAnyPointerOrIntptr(QualType ty, ASTContext &Ctx) {
if (ty->isAnyPointerType())
return true;
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
return ty->isIntegerType() && ty->isScalarType() &&
Ctx.getTypeSize(ty) == Ctx.getTypeSize(Ctx.VoidPtrTy);
}
//===----------------------------------------------------------------------===//
// Main RegionStore logic.
//===----------------------------------------------------------------------===//
namespace {
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
class VISIBILITY_HIDDEN RegionStoreSubRegionMap : public SubRegionMap {
typedef llvm::ImmutableSet<const MemRegion*> SetTy;
typedef llvm::DenseMap<const MemRegion*, SetTy> Map;
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
SetTy::Factory F;
Map M;
public:
bool add(const MemRegion* Parent, const MemRegion* SubRegion) {
Map::iterator I = M.find(Parent);
if (I == M.end()) {
M.insert(std::make_pair(Parent, F.Add(F.GetEmptySet(), SubRegion)));
return true;
}
I->second = F.Add(I->second, SubRegion);
return false;
}
void process(llvm::SmallVectorImpl<const SubRegion*> &WL, const SubRegion *R);
~RegionStoreSubRegionMap() {}
bool iterSubRegions(const MemRegion* Parent, Visitor& V) const {
Map::iterator I = M.find(Parent);
if (I == M.end())
return true;
llvm::ImmutableSet<const MemRegion*> S = I->second;
for (llvm::ImmutableSet<const MemRegion*>::iterator SI=S.begin(),SE=S.end();
SI != SE; ++SI) {
if (!V.Visit(Parent, *SI))
return false;
}
return true;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
typedef SetTy::iterator iterator;
std::pair<iterator, iterator> begin_end(const MemRegion *R) {
Map::iterator I = M.find(R);
SetTy S = I == M.end() ? F.GetEmptySet() : I->second;
return std::make_pair(S.begin(), S.end());
}
};
class VISIBILITY_HIDDEN RegionStoreManager : public StoreManager {
const RegionStoreFeatures Features;
RegionBindings::Factory RBFactory;
typedef llvm::DenseMap<const GRState *, RegionStoreSubRegionMap*> SMCache;
SMCache SC;
public:
RegionStoreManager(GRStateManager& mgr, const RegionStoreFeatures &f)
: StoreManager(mgr),
Features(f),
RBFactory(mgr.getAllocator()) {}
virtual ~RegionStoreManager() {
for (SMCache::iterator I = SC.begin(), E = SC.end(); I != E; ++I)
delete (*I).second;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
SubRegionMap *getSubRegionMap(const GRState *state);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
RegionStoreSubRegionMap *getRegionStoreSubRegionMap(const GRState *state);
/// getDefaultBinding - Returns an SVal* representing an optional default
/// binding associated with a region and its subregions.
Optional<SVal> getDefaultBinding(const GRState *state, const MemRegion *R);
/// getLValueString - Returns an SVal representing the lvalue of a
/// StringLiteral. Within RegionStore a StringLiteral has an
/// associated StringRegion, and the lvalue of a StringLiteral is
/// the lvalue of that region.
SVal getLValueString(const GRState *state, const StringLiteral* S);
/// getLValueCompoundLiteral - Returns an SVal representing the
/// lvalue of a compound literal. Within RegionStore a compound
/// literal has an associated region, and the lvalue of the
/// compound literal is the lvalue of that region.
SVal getLValueCompoundLiteral(const GRState *state, const CompoundLiteralExpr*);
/// getLValueVar - Returns an SVal that represents the lvalue of a
/// variable. Within RegionStore a variable has an associated
/// VarRegion, and the lvalue of the variable is the lvalue of that region.
SVal getLValueVar(const GRState *ST, const VarDecl *VD,
const LocationContext *LC);
SVal getLValueIvar(const GRState *state, const ObjCIvarDecl* D, SVal Base);
SVal getLValueField(const GRState *state, SVal Base, const FieldDecl* D);
SVal getLValueFieldOrIvar(const GRState *state, SVal Base, const Decl* D);
SVal getLValueElement(const GRState *state, QualType elementType,
SVal Base, SVal Offset);
/// ArrayToPointer - Emulates the "decay" of an array to a pointer
/// type. 'Array' represents the lvalue of the array being decayed
/// to a pointer, and the returned SVal represents the decayed
/// version of that lvalue (i.e., a pointer to the first element of
/// the array). This is called by GRExprEngine when evaluating
/// casts from arrays to pointers.
2009-03-30 13:55:46 +08:00
SVal ArrayToPointer(Loc Array);
SVal EvalBinOp(const GRState *state, BinaryOperator::Opcode Op,Loc L,
NonLoc R, QualType resultTy);
Store getInitialStore(const LocationContext *InitLoc) {
return RBFactory.GetEmptyMap().getRoot();
}
//===-------------------------------------------------------------------===//
// Binding values to regions.
//===-------------------------------------------------------------------===//
const GRState *InvalidateRegion(const GRState *state, const MemRegion *R,
const Expr *E, unsigned Count);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
private:
void RemoveSubRegionBindings(RegionBindings &B,
RegionDefaultBindings &DVM,
RegionDefaultBindings::Factory &DVMFactory,
const MemRegion *R,
RegionStoreSubRegionMap &M);
public:
const GRState *Bind(const GRState *state, Loc LV, SVal V);
const GRState *BindCompoundLiteral(const GRState *state,
const CompoundLiteralExpr* CL, SVal V);
const GRState *BindDecl(const GRState *ST, const VarDecl *VD,
const LocationContext *LC, SVal InitVal);
const GRState *BindDeclWithNoInit(const GRState *state, const VarDecl*,
const LocationContext *) {
return state;
}
/// BindStruct - Bind a compound value to a structure.
const GRState *BindStruct(const GRState *, const TypedRegion* R, SVal V);
const GRState *BindArray(const GRState *state, const TypedRegion* R, SVal V);
/// KillStruct - Set the entire struct to unknown.
const GRState *KillStruct(const GRState *state, const TypedRegion* R);
const GRState *setDefaultValue(const GRState *state, const MemRegion* R, SVal V);
Store Remove(Store store, Loc LV);
//===------------------------------------------------------------------===//
// Loading values from regions.
//===------------------------------------------------------------------===//
/// The high level logic for this method is this:
/// Retrieve (L)
/// if L has binding
/// return L's binding
/// else if L is in killset
/// return unknown
/// else
/// if L is on stack or heap
/// return undefined
/// else
/// return symbolic
SValuator::CastResult Retrieve(const GRState *state, Loc L,
QualType T = QualType());
SVal RetrieveElement(const GRState *state, const ElementRegion *R);
SVal RetrieveField(const GRState *state, const FieldRegion *R);
SVal RetrieveObjCIvar(const GRState *state, const ObjCIvarRegion *R);
SVal RetrieveVar(const GRState *state, const VarRegion *R);
SVal RetrieveLazySymbol(const GRState *state, const TypedRegion *R);
SVal RetrieveFieldOrElementCommon(const GRState *state, const TypedRegion *R,
QualType Ty, const MemRegion *superR);
2008-12-04 09:12:41 +08:00
/// Retrieve the values in a struct and return a CompoundVal, used when doing
/// struct copy:
/// struct s x, y;
2008-12-04 09:12:41 +08:00
/// x = y;
/// y's value is retrieved by this method.
SVal RetrieveStruct(const GRState *St, const TypedRegion* R);
SVal RetrieveArray(const GRState *St, const TypedRegion* R);
std::pair<const GRState*, const MemRegion*>
GetLazyBinding(RegionBindings B, const MemRegion *R);
const GRState* CopyLazyBindings(nonloc::LazyCompoundVal V,
const GRState *state,
const TypedRegion *R);
2008-12-04 09:12:41 +08:00
const ElementRegion *GetElementZeroRegion(const SymbolicRegion *SR,
QualType T);
//===------------------------------------------------------------------===//
// State pruning.
//===------------------------------------------------------------------===//
/// RemoveDeadBindings - Scans the RegionStore of 'state' for dead values.
/// It returns a new Store with these values removed.
void RemoveDeadBindings(GRState &state, Stmt* Loc, SymbolReaper& SymReaper,
llvm::SmallVectorImpl<const MemRegion*>& RegionRoots);
//===------------------------------------------------------------------===//
// Region "extents".
//===------------------------------------------------------------------===//
const GRState *setExtent(const GRState *state, const MemRegion* R, SVal Extent);
SVal getSizeInElements(const GRState *state, const MemRegion* R);
//===------------------------------------------------------------------===//
// Utility methods.
//===------------------------------------------------------------------===//
static inline RegionBindings GetRegionBindings(Store store) {
return RegionBindings(static_cast<const RegionBindings::TreeTy*>(store));
}
void print(Store store, llvm::raw_ostream& Out, const char* nl,
const char *sep);
2009-05-08 09:33:18 +08:00
void iterBindings(Store store, BindingsHandler& f) {
// FIXME: Implement.
}
// FIXME: Remove.
BasicValueFactory& getBasicVals() {
return StateMgr.getBasicVals();
}
// FIXME: Remove.
ASTContext& getContext() { return StateMgr.getContext(); }
};
} // end anonymous namespace
//===----------------------------------------------------------------------===//
// RegionStore creation.
//===----------------------------------------------------------------------===//
StoreManager *clang::CreateRegionStoreManager(GRStateManager& StMgr) {
RegionStoreFeatures F = maximal_features_tag();
return new RegionStoreManager(StMgr, F);
}
StoreManager *clang::CreateFieldsOnlyRegionStoreManager(GRStateManager &StMgr) {
RegionStoreFeatures F = minimal_features_tag();
F.enableFields(true);
return new RegionStoreManager(StMgr, F);
}
void
RegionStoreSubRegionMap::process(llvm::SmallVectorImpl<const SubRegion*> &WL,
const SubRegion *R) {
const MemRegion *superR = R->getSuperRegion();
if (add(superR, R))
if (const SubRegion *sr = dyn_cast<SubRegion>(superR))
WL.push_back(sr);
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
RegionStoreSubRegionMap*
RegionStoreManager::getRegionStoreSubRegionMap(const GRState *state) {
RegionBindings B = GetRegionBindings(state->getStore());
RegionStoreSubRegionMap *M = new RegionStoreSubRegionMap();
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
llvm::SmallVector<const SubRegion*, 10> WL;
for (RegionBindings::iterator I=B.begin(), E=B.end(); I!=E; ++I)
if (const SubRegion *R = dyn_cast<SubRegion>(I.getKey()))
M->process(WL, R);
RegionDefaultBindings DVM = state->get<RegionDefaultValue>();
for (RegionDefaultBindings::iterator I = DVM.begin(), E = DVM.end();
I != E; ++I)
if (const SubRegion *R = dyn_cast<SubRegion>(I.getKey()))
M->process(WL, R);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// We also need to record in the subregion map "intermediate" regions that
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// don't have direct bindings but are super regions of those that do.
while (!WL.empty()) {
const SubRegion *R = WL.back();
WL.pop_back();
M->process(WL, R);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
}
return M;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
SubRegionMap *RegionStoreManager::getSubRegionMap(const GRState *state) {
return getRegionStoreSubRegionMap(state);
}
//===----------------------------------------------------------------------===//
// Binding invalidation.
//===----------------------------------------------------------------------===//
void
RegionStoreManager::RemoveSubRegionBindings(RegionBindings &B,
RegionDefaultBindings &DVM,
RegionDefaultBindings::Factory &DVMFactory,
const MemRegion *R,
RegionStoreSubRegionMap &M) {
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
RegionStoreSubRegionMap::iterator I, E;
for (llvm::tie(I, E) = M.begin_end(R); I != E; ++I)
RemoveSubRegionBindings(B, DVM, DVMFactory, *I, M);
B = RBFactory.Remove(B, R);
DVM = DVMFactory.Remove(DVM, R);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
}
const GRState *RegionStoreManager::InvalidateRegion(const GRState *state,
const MemRegion *R,
const Expr *Ex,
unsigned Count) {
ASTContext& Ctx = StateMgr.getContext();
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Strip away casts.
R = R->getBaseRegion();
// Get the mapping of regions -> subregions.
llvm::OwningPtr<RegionStoreSubRegionMap>
SubRegions(getRegionStoreSubRegionMap(state));
RegionBindings B = GetRegionBindings(state->getStore());
RegionDefaultBindings DVM = state->get<RegionDefaultValue>();
RegionDefaultBindings::Factory &DVMFactory =
state->get_context<RegionDefaultValue>();
llvm::DenseMap<const MemRegion *, unsigned> Visited;
llvm::SmallVector<const MemRegion *, 10> WorkList;
WorkList.push_back(R);
while (!WorkList.empty()) {
R = WorkList.back();
WorkList.pop_back();
// Have we visited this region before?
unsigned &visited = Visited[R];
if (visited)
continue;
visited = 1;
// Add subregions to work list.
RegionStoreSubRegionMap::iterator I, E;
for (llvm::tie(I, E) = SubRegions->begin_end(R); I!=E; ++I)
WorkList.push_back(*I);
// Handle region.
if (isa<AllocaRegion>(R) || isa<SymbolicRegion>(R) ||
isa<ObjCObjectRegion>(R)) {
// Invalidate the region by setting its default value to
// conjured symbol. The type of the symbol is irrelavant.
DefinedOrUnknownSVal V = ValMgr.getConjuredSymbolVal(R, Ex, Ctx.IntTy,
Count);
DVM = DVMFactory.Add(DVM, R, V);
continue;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (!R->isBoundable())
continue;
const TypedRegion *TR = cast<TypedRegion>(R);
QualType T = TR->getValueType(Ctx);
if (const RecordType *RT = T->getAsStructureType()) {
// FIXME: handle structs with default region value.
const RecordDecl *RD = RT->getDecl()->getDefinition(Ctx);
// No record definition. There is nothing we can do.
if (!RD)
continue;
// Invalidate the region by setting its default value to
// conjured symbol. The type of the symbol is irrelavant.
DefinedOrUnknownSVal V = ValMgr.getConjuredSymbolVal(R, Ex, Ctx.IntTy,
Count);
DVM = DVMFactory.Add(DVM, R, V);
continue;
}
if (const ArrayType *AT = Ctx.getAsArrayType(T)) {
// Set the default value of the array to conjured symbol.
DefinedOrUnknownSVal V =
ValMgr.getConjuredSymbolVal(R, Ex, AT->getElementType(), Count);
DVM = DVMFactory.Add(DVM, R, V);
continue;
}
// Get the old binding. Is it a region? If so, add it to the worklist.
if (const SVal *OldV = B.lookup(R)) {
if (const MemRegion *RV = OldV->getAsRegion())
WorkList.push_back(RV);
}
if ((isa<FieldRegion>(R)||isa<ElementRegion>(R)||isa<ObjCIvarRegion>(R))
&& Visited[cast<SubRegion>(R)->getSuperRegion()]) {
// For fields and elements whose super region has also been invalidated,
// only remove the old binding. The super region will get set with a
// default value from which we can lazily derive a new symbolic value.
B = RBFactory.Remove(B, R);
continue;
}
// Invalidate the binding.
DefinedOrUnknownSVal V = ValMgr.getConjuredSymbolVal(R, Ex, T, Count);
assert(SymbolManager::canSymbolicate(T) || V.isUnknown());
B = RBFactory.Add(B, R, V);
}
// Create a new state with the updated bindings.
return state->makeWithStore(B.getRoot())->set<RegionDefaultValue>(DVM);
}
//===----------------------------------------------------------------------===//
// getLValueXXX methods.
//===----------------------------------------------------------------------===//
/// getLValueString - Returns an SVal representing the lvalue of a
/// StringLiteral. Within RegionStore a StringLiteral has an
/// associated StringRegion, and the lvalue of a StringLiteral is the
/// lvalue of that region.
SVal RegionStoreManager::getLValueString(const GRState *St,
const StringLiteral* S) {
return loc::MemRegionVal(MRMgr.getStringRegion(S));
}
/// getLValueVar - Returns an SVal that represents the lvalue of a
/// variable. Within RegionStore a variable has an associated
/// VarRegion, and the lvalue of the variable is the lvalue of that region.
SVal RegionStoreManager::getLValueVar(const GRState *ST, const VarDecl *VD,
const LocationContext *LC) {
return loc::MemRegionVal(MRMgr.getVarRegion(VD, LC));
}
/// getLValueCompoundLiteral - Returns an SVal representing the lvalue
/// of a compound literal. Within RegionStore a compound literal
/// has an associated region, and the lvalue of the compound literal
/// is the lvalue of that region.
SVal
RegionStoreManager::getLValueCompoundLiteral(const GRState *St,
const CompoundLiteralExpr* CL) {
return loc::MemRegionVal(MRMgr.getCompoundLiteralRegion(CL));
}
SVal RegionStoreManager::getLValueIvar(const GRState *St, const ObjCIvarDecl* D,
SVal Base) {
return getLValueFieldOrIvar(St, Base, D);
}
SVal RegionStoreManager::getLValueField(const GRState *St, SVal Base,
const FieldDecl* D) {
return getLValueFieldOrIvar(St, Base, D);
}
SVal RegionStoreManager::getLValueFieldOrIvar(const GRState *St, SVal Base,
const Decl* D) {
if (Base.isUnknownOrUndef())
return Base;
Loc BaseL = cast<Loc>(Base);
const MemRegion* BaseR = 0;
switch (BaseL.getSubKind()) {
case loc::MemRegionKind:
BaseR = cast<loc::MemRegionVal>(BaseL).getRegion();
break;
case loc::GotoLabelKind:
// These are anormal cases. Flag an undefined value.
return UndefinedVal();
case loc::ConcreteIntKind:
// While these seem funny, this can happen through casts.
// FIXME: What we should return is the field offset. For example,
// add the field offset to the integer value. That way funny things
// like this work properly: &(((struct foo *) 0xa)->f)
return Base;
default:
assert(0 && "Unhandled Base.");
return Base;
}
// NOTE: We must have this check first because ObjCIvarDecl is a subclass
// of FieldDecl.
if (const ObjCIvarDecl *ID = dyn_cast<ObjCIvarDecl>(D))
return loc::MemRegionVal(MRMgr.getObjCIvarRegion(ID, BaseR));
return loc::MemRegionVal(MRMgr.getFieldRegion(cast<FieldDecl>(D), BaseR));
}
SVal RegionStoreManager::getLValueElement(const GRState *St,
QualType elementType,
SVal Base, SVal Offset) {
// If the base is an unknown or undefined value, just return it back.
// FIXME: For absolute pointer addresses, we just return that value back as
// well, although in reality we should return the offset added to that
// value.
if (Base.isUnknownOrUndef() || isa<loc::ConcreteInt>(Base))
return Base;
// Only handle integer offsets... for now.
if (!isa<nonloc::ConcreteInt>(Offset))
return UnknownVal();
const MemRegion* BaseRegion = cast<loc::MemRegionVal>(Base).getRegion();
// Pointer of any type can be cast and used as array base.
const ElementRegion *ElemR = dyn_cast<ElementRegion>(BaseRegion);
// Convert the offset to the appropriate size and signedness.
Offset = ValMgr.convertToArrayIndex(Offset);
if (!ElemR) {
//
// If the base region is not an ElementRegion, create one.
// This can happen in the following example:
//
// char *p = __builtin_alloc(10);
// p[1] = 8;
//
// Observe that 'p' binds to an AllocaRegion.
//
return loc::MemRegionVal(MRMgr.getElementRegion(elementType, Offset,
BaseRegion, getContext()));
}
SVal BaseIdx = ElemR->getIndex();
if (!isa<nonloc::ConcreteInt>(BaseIdx))
return UnknownVal();
const llvm::APSInt& BaseIdxI = cast<nonloc::ConcreteInt>(BaseIdx).getValue();
const llvm::APSInt& OffI = cast<nonloc::ConcreteInt>(Offset).getValue();
assert(BaseIdxI.isSigned());
// Compute the new index.
SVal NewIdx = nonloc::ConcreteInt(getBasicVals().getValue(BaseIdxI + OffI));
// Construct the new ElementRegion.
const MemRegion *ArrayR = ElemR->getSuperRegion();
return loc::MemRegionVal(MRMgr.getElementRegion(elementType, NewIdx, ArrayR,
getContext()));
}
//===----------------------------------------------------------------------===//
// Extents for regions.
//===----------------------------------------------------------------------===//
SVal RegionStoreManager::getSizeInElements(const GRState *state,
const MemRegion *R) {
switch (R->getKind()) {
case MemRegion::MemSpaceRegionKind:
assert(0 && "Cannot index into a MemSpace");
return UnknownVal();
case MemRegion::CodeTextRegionKind:
// Technically this can happen if people do funny things with casts.
return UnknownVal();
// Not yet handled.
case MemRegion::AllocaRegionKind:
case MemRegion::CompoundLiteralRegionKind:
case MemRegion::ElementRegionKind:
case MemRegion::FieldRegionKind:
case MemRegion::ObjCIvarRegionKind:
case MemRegion::ObjCObjectRegionKind:
case MemRegion::SymbolicRegionKind:
return UnknownVal();
case MemRegion::StringRegionKind: {
const StringLiteral* Str = cast<StringRegion>(R)->getStringLiteral();
// We intentionally made the size value signed because it participates in
// operations with signed indices.
return ValMgr.makeIntVal(Str->getByteLength()+1, false);
}
case MemRegion::VarRegionKind: {
const VarRegion* VR = cast<VarRegion>(R);
// Get the type of the variable.
QualType T = VR->getDesugaredValueType(getContext());
// FIXME: Handle variable-length arrays.
if (isa<VariableArrayType>(T))
return UnknownVal();
if (const ConstantArrayType* CAT = dyn_cast<ConstantArrayType>(T)) {
// return the size as signed integer.
return ValMgr.makeIntVal(CAT->getSize(), false);
}
// Clients can use ordinary variables as if they were arrays. These
// essentially are arrays of size 1.
return ValMgr.makeIntVal(1, false);
}
case MemRegion::BEG_DECL_REGIONS:
case MemRegion::END_DECL_REGIONS:
case MemRegion::BEG_TYPED_REGIONS:
case MemRegion::END_TYPED_REGIONS:
assert(0 && "Infeasible region");
return UnknownVal();
}
assert(0 && "Unreachable");
return UnknownVal();
}
const GRState *RegionStoreManager::setExtent(const GRState *state,
const MemRegion *region,
SVal extent) {
return state->set<RegionExtents>(region, extent);
}
//===----------------------------------------------------------------------===//
// Location and region casting.
//===----------------------------------------------------------------------===//
/// ArrayToPointer - Emulates the "decay" of an array to a pointer
/// type. 'Array' represents the lvalue of the array being decayed
/// to a pointer, and the returned SVal represents the decayed
/// version of that lvalue (i.e., a pointer to the first element of
/// the array). This is called by GRExprEngine when evaluating casts
/// from arrays to pointers.
2009-03-30 13:55:46 +08:00
SVal RegionStoreManager::ArrayToPointer(Loc Array) {
if (!isa<loc::MemRegionVal>(Array))
return UnknownVal();
const MemRegion* R = cast<loc::MemRegionVal>(&Array)->getRegion();
const TypedRegion* ArrayR = dyn_cast<TypedRegion>(R);
if (!ArrayR)
return UnknownVal();
// Strip off typedefs from the ArrayRegion's ValueType.
QualType T = ArrayR->getValueType(getContext())->getDesugaredType();
ArrayType *AT = cast<ArrayType>(T);
T = AT->getElementType();
SVal ZeroIdx = ValMgr.makeZeroArrayIndex();
ElementRegion* ER = MRMgr.getElementRegion(T, ZeroIdx, ArrayR, getContext());
return loc::MemRegionVal(ER);
}
//===----------------------------------------------------------------------===//
// Pointer arithmetic.
//===----------------------------------------------------------------------===//
SVal RegionStoreManager::EvalBinOp(const GRState *state,
BinaryOperator::Opcode Op, Loc L, NonLoc R,
QualType resultTy) {
// Assume the base location is MemRegionVal.
if (!isa<loc::MemRegionVal>(L))
return UnknownVal();
const MemRegion* MR = cast<loc::MemRegionVal>(L).getRegion();
const ElementRegion *ER = 0;
switch (MR->getKind()) {
case MemRegion::SymbolicRegionKind: {
const SymbolicRegion *SR = cast<SymbolicRegion>(MR);
SymbolRef Sym = SR->getSymbol();
QualType T = Sym->getType(getContext());
QualType EleTy;
if (const PointerType *PT = T->getAs<PointerType>())
EleTy = PT->getPointeeType();
else
EleTy = T->getAs<ObjCObjectPointerType>()->getPointeeType();
SVal ZeroIdx = ValMgr.makeZeroArrayIndex();
ER = MRMgr.getElementRegion(EleTy, ZeroIdx, SR, getContext());
break;
}
case MemRegion::AllocaRegionKind: {
const AllocaRegion *AR = cast<AllocaRegion>(MR);
QualType T = getContext().CharTy; // Create an ElementRegion of bytes.
QualType EleTy = T->getAs<PointerType>()->getPointeeType();
SVal ZeroIdx = ValMgr.makeZeroArrayIndex();
ER = MRMgr.getElementRegion(EleTy, ZeroIdx, AR, getContext());
break;
}
case MemRegion::ElementRegionKind: {
ER = cast<ElementRegion>(MR);
break;
}
// Not yet handled.
case MemRegion::VarRegionKind:
case MemRegion::StringRegionKind:
case MemRegion::CompoundLiteralRegionKind:
case MemRegion::FieldRegionKind:
case MemRegion::ObjCObjectRegionKind:
case MemRegion::ObjCIvarRegionKind:
return UnknownVal();
case MemRegion::CodeTextRegionKind:
// Technically this can happen if people do funny things with casts.
return UnknownVal();
case MemRegion::MemSpaceRegionKind:
assert(0 && "Cannot perform pointer arithmetic on a MemSpace");
return UnknownVal();
case MemRegion::BEG_DECL_REGIONS:
case MemRegion::END_DECL_REGIONS:
case MemRegion::BEG_TYPED_REGIONS:
case MemRegion::END_TYPED_REGIONS:
assert(0 && "Infeasible region");
return UnknownVal();
}
SVal Idx = ER->getIndex();
nonloc::ConcreteInt* Base = dyn_cast<nonloc::ConcreteInt>(&Idx);
nonloc::ConcreteInt* Offset = dyn_cast<nonloc::ConcreteInt>(&R);
// Only support concrete integer indexes for now.
if (Base && Offset) {
// FIXME: Should use SValuator here.
SVal NewIdx = Base->evalBinOp(ValMgr, Op,
cast<nonloc::ConcreteInt>(ValMgr.convertToArrayIndex(*Offset)));
const MemRegion* NewER =
MRMgr.getElementRegion(ER->getElementType(), NewIdx, ER->getSuperRegion(),
getContext());
return ValMgr.makeLoc(NewER);
}
return UnknownVal();
}
//===----------------------------------------------------------------------===//
// Loading values from regions.
//===----------------------------------------------------------------------===//
Optional<SVal> RegionStoreManager::getDefaultBinding(const GRState *state,
const MemRegion *R) {
if (R->isBoundable())
if (const TypedRegion *TR = dyn_cast<TypedRegion>(R))
if (TR->getValueType(getContext())->isUnionType())
return UnknownVal();
return Optional<SVal>::create(state->get<RegionDefaultValue>(R));
}
static bool IsReinterpreted(QualType RTy, QualType UsedTy, ASTContext &Ctx) {
RTy = Ctx.getCanonicalType(RTy);
UsedTy = Ctx.getCanonicalType(UsedTy);
if (RTy == UsedTy)
return false;
// Recursively check the types. We basically want to see if a pointer value
// is ever reinterpreted as a non-pointer, e.g. void** and intptr_t*
// represents a reinterpretation.
if (Loc::IsLocType(RTy) && Loc::IsLocType(UsedTy)) {
const PointerType *PRTy = RTy->getAs<PointerType>();
const PointerType *PUsedTy = UsedTy->getAs<PointerType>();
return PUsedTy && PRTy &&
IsReinterpreted(PRTy->getPointeeType(),
PUsedTy->getPointeeType(), Ctx);
}
return true;
}
const ElementRegion *
RegionStoreManager::GetElementZeroRegion(const SymbolicRegion *SR, QualType T) {
ASTContext &Ctx = getContext();
SVal idx = ValMgr.makeZeroArrayIndex();
assert(!T.isNull());
return MRMgr.getElementRegion(T, idx, SR, Ctx);
}
SValuator::CastResult
RegionStoreManager::Retrieve(const GRState *state, Loc L, QualType T) {
assert(!isa<UnknownVal>(L) && "location unknown");
assert(!isa<UndefinedVal>(L) && "location undefined");
// FIXME: Is this even possible? Shouldn't this be treated as a null
// dereference at a higher level?
if (isa<loc::ConcreteInt>(L))
return SValuator::CastResult(state, UndefinedVal());
const MemRegion *MR = cast<loc::MemRegionVal>(L).getRegion();
// FIXME: return symbolic value for these cases.
// Example:
// void f(int* p) { int x = *p; }
// char* p = alloca();
// read(p);
// c = *p;
if (isa<AllocaRegion>(MR))
return SValuator::CastResult(state, UnknownVal());
if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(MR))
MR = GetElementZeroRegion(SR, T);
if (isa<CodeTextRegion>(MR))
return SValuator::CastResult(state, UnknownVal());
// FIXME: Perhaps this method should just take a 'const MemRegion*' argument
// instead of 'Loc', and have the other Loc cases handled at a higher level.
const TypedRegion *R = cast<TypedRegion>(MR);
QualType RTy = R->getValueType(getContext());
// FIXME: We should eventually handle funny addressing. e.g.:
//
// int x = ...;
// int *p = &x;
// char *q = (char*) p;
// char c = *q; // returns the first byte of 'x'.
//
// Such funny addressing will occur due to layering of regions.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
#if 0
ASTContext &Ctx = getContext();
if (!T.isNull() && IsReinterpreted(RTy, T, Ctx)) {
SVal ZeroIdx = ValMgr.makeZeroArrayIndex();
R = MRMgr.getElementRegion(T, ZeroIdx, R, Ctx);
RTy = T;
2009-07-15 12:23:32 +08:00
assert(Ctx.getCanonicalType(RTy) ==
Ctx.getCanonicalType(R->getValueType(Ctx)));
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
#endif
if (RTy->isStructureType())
return SValuator::CastResult(state, RetrieveStruct(state, R));
// FIXME: Handle unions.
if (RTy->isUnionType())
return SValuator::CastResult(state, UnknownVal());
if (RTy->isArrayType())
return SValuator::CastResult(state, RetrieveArray(state, R));
// FIXME: handle Vector types.
if (RTy->isVectorType())
return SValuator::CastResult(state, UnknownVal());
if (const FieldRegion* FR = dyn_cast<FieldRegion>(R))
return CastRetrievedVal(RetrieveField(state, FR), state, FR, T);
if (const ElementRegion* ER = dyn_cast<ElementRegion>(R))
return CastRetrievedVal(RetrieveElement(state, ER), state, ER, T);
if (const ObjCIvarRegion *IVR = dyn_cast<ObjCIvarRegion>(R))
return CastRetrievedVal(RetrieveObjCIvar(state, IVR), state, IVR, T);
if (const VarRegion *VR = dyn_cast<VarRegion>(R))
return CastRetrievedVal(RetrieveVar(state, VR), state, VR, T);
RegionBindings B = GetRegionBindings(state->getStore());
RegionBindings::data_type* V = B.lookup(R);
// Check if the region has a binding.
if (V)
return SValuator::CastResult(state, *V);
// The location does not have a bound value. This means that it has
// the value it had upon its creation and/or entry to the analyzed
// function/method. These are either symbolic values or 'undefined'.
#if HEAP_UNDEFINED
if (R->hasHeapOrStackStorage()) {
#else
if (R->hasStackStorage()) {
#endif
// All stack variables are considered to have undefined values
// upon creation. All heap allocated blocks are considered to
// have undefined values as well unless they are explicitly bound
// to specific values.
return SValuator::CastResult(state, UndefinedVal());
}
// All other values are symbolic.
return SValuator::CastResult(state,
ValMgr.getRegionValueSymbolValOrUnknown(R, RTy));
}
std::pair<const GRState*, const MemRegion*>
RegionStoreManager::GetLazyBinding(RegionBindings B, const MemRegion *R) {
if (const nonloc::LazyCompoundVal *V =
dyn_cast_or_null<nonloc::LazyCompoundVal>(B.lookup(R)))
return std::make_pair(V->getState(), V->getRegion());
if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
const std::pair<const GRState *, const MemRegion *> &X =
GetLazyBinding(B, ER->getSuperRegion());
if (X.first)
return std::make_pair(X.first,
MRMgr.getElementRegionWithSuper(ER, X.second));
}
else if (const FieldRegion *FR = dyn_cast<FieldRegion>(R)) {
const std::pair<const GRState *, const MemRegion *> &X =
GetLazyBinding(B, FR->getSuperRegion());
if (X.first)
return std::make_pair(X.first,
MRMgr.getFieldRegionWithSuper(FR, X.second));
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
return std::make_pair((const GRState*) 0, (const MemRegion *) 0);
}
SVal RegionStoreManager::RetrieveElement(const GRState* state,
const ElementRegion* R) {
// Check if the region has a binding.
RegionBindings B = GetRegionBindings(state->getStore());
if (const SVal* V = B.lookup(R))
return *V;
const MemRegion* superR = R->getSuperRegion();
// Check if the region is an element region of a string literal.
if (const StringRegion *StrR=dyn_cast<StringRegion>(superR)) {
const StringLiteral *Str = StrR->getStringLiteral();
SVal Idx = R->getIndex();
if (nonloc::ConcreteInt *CI = dyn_cast<nonloc::ConcreteInt>(&Idx)) {
int64_t i = CI->getValue().getSExtValue();
int64_t byteLength = Str->getByteLength();
if (i > byteLength) {
// Buffer overflow checking in GRExprEngine should handle this case,
// but we shouldn't rely on it to not overflow here if that checking
// is disabled.
return UnknownVal();
}
char c = (i == byteLength) ? '\0' : Str->getStrData()[i];
return ValMgr.makeIntVal(c, getContext().CharTy);
}
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Special case: the current region represents a cast and it and the super
// region both have pointer types or intptr_t types. If so, perform the
// retrieve from the super region and appropriately "cast" the value.
// This is needed to support OSAtomicCompareAndSwap and friends or other
// loads that treat integers as pointers and vis versa.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (R->getIndex().isZeroConstant()) {
if (const TypedRegion *superTR = dyn_cast<TypedRegion>(superR)) {
ASTContext &Ctx = getContext();
if (IsAnyPointerOrIntptr(superTR->getValueType(Ctx), Ctx)) {
QualType valTy = R->getValueType(Ctx);
if (IsAnyPointerOrIntptr(valTy, Ctx)) {
// Retrieve the value from the super region. This will be casted to
// valTy when we return to 'Retrieve'.
const SValuator::CastResult &cr = Retrieve(state,
loc::MemRegionVal(superR),
valTy);
return cr.getSVal();
}
}
}
}
// Check if the immediate super region has a direct binding.
if (const SVal *V = B.lookup(superR)) {
if (SymbolRef parentSym = V->getAsSymbol())
return ValMgr.getDerivedRegionValueSymbolVal(parentSym, R);
if (V->isUnknownOrUndef())
return *V;
// Handle LazyCompoundVals for the immediate super region. Other cases
// are handled in 'RetrieveFieldOrElementCommon'.
if (const nonloc::LazyCompoundVal *LCV =
dyn_cast<nonloc::LazyCompoundVal>(V)) {
R = MRMgr.getElementRegionWithSuper(R, LCV->getRegion());
return RetrieveElement(LCV->getState(), R);
}
// Other cases: give up.
return UnknownVal();
}
return RetrieveFieldOrElementCommon(state, R, R->getElementType(), superR);
}
SVal RegionStoreManager::RetrieveField(const GRState* state,
const FieldRegion* R) {
// Check if the region has a binding.
RegionBindings B = GetRegionBindings(state->getStore());
if (const SVal* V = B.lookup(R))
return *V;
QualType Ty = R->getValueType(getContext());
return RetrieveFieldOrElementCommon(state, R, Ty, R->getSuperRegion());
}
SVal RegionStoreManager::RetrieveFieldOrElementCommon(const GRState *state,
const TypedRegion *R,
QualType Ty,
const MemRegion *superR) {
// At this point we have already checked in either RetrieveElement or
// RetrieveField if 'R' has a direct binding.
RegionBindings B = GetRegionBindings(state->getStore());
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
while (superR) {
if (const Optional<SVal> &D = getDefaultBinding(state, superR)) {
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (SymbolRef parentSym = D->getAsSymbol())
return ValMgr.getDerivedRegionValueSymbolVal(parentSym, R);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (D->isZeroConstant())
return ValMgr.makeZeroVal(Ty);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (D->isUnknown())
return *D;
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
assert(0 && "Unknown default value");
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// If our super region is a field or element itself, walk up the region
// hierarchy to see if there is a default value installed in an ancestor.
if (isa<FieldRegion>(superR) || isa<ElementRegion>(superR)) {
superR = cast<SubRegion>(superR)->getSuperRegion();
continue;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
break;
}
// Lazy binding?
const GRState *lazyBindingState = NULL;
const MemRegion *lazyBindingRegion = NULL;
llvm::tie(lazyBindingState, lazyBindingRegion) = GetLazyBinding(B, R);
if (lazyBindingState) {
assert(lazyBindingRegion && "Lazy-binding region not set");
if (isa<ElementRegion>(R))
return RetrieveElement(lazyBindingState,
cast<ElementRegion>(lazyBindingRegion));
return RetrieveField(lazyBindingState,
cast<FieldRegion>(lazyBindingRegion));
}
if (R->hasStackStorage() && !R->hasParametersStorage()) {
if (isa<ElementRegion>(R)) {
// Currently we don't reason specially about Clang-style vectors. Check
// if superR is a vector and if so return Unknown.
if (const TypedRegion *typedSuperR = dyn_cast<TypedRegion>(superR)) {
if (typedSuperR->getValueType(getContext())->isVectorType())
return UnknownVal();
}
}
return UndefinedVal();
}
// All other values are symbolic.
return ValMgr.getRegionValueSymbolValOrUnknown(R, Ty);
}
SVal RegionStoreManager::RetrieveObjCIvar(const GRState* state,
const ObjCIvarRegion* R) {
// Check if the region has a binding.
RegionBindings B = GetRegionBindings(state->getStore());
if (const SVal* V = B.lookup(R))
return *V;
const MemRegion *superR = R->getSuperRegion();
// Check if the super region has a binding.
if (const SVal *V = B.lookup(superR)) {
if (SymbolRef parentSym = V->getAsSymbol())
return ValMgr.getDerivedRegionValueSymbolVal(parentSym, R);
// Other cases: give up.
return UnknownVal();
}
return RetrieveLazySymbol(state, R);
}
SVal RegionStoreManager::RetrieveVar(const GRState *state,
const VarRegion *R) {
// Check if the region has a binding.
RegionBindings B = GetRegionBindings(state->getStore());
if (const SVal* V = B.lookup(R))
return *V;
// Lazily derive a value for the VarRegion.
const VarDecl *VD = R->getDecl();
if (R->hasGlobalsOrParametersStorage())
return ValMgr.getRegionValueSymbolValOrUnknown(R, VD->getType());
return UndefinedVal();
}
SVal RegionStoreManager::RetrieveLazySymbol(const GRState *state,
const TypedRegion *R) {
QualType valTy = R->getValueType(getContext());
// All other values are symbolic.
return ValMgr.getRegionValueSymbolValOrUnknown(R, valTy);
}
SVal RegionStoreManager::RetrieveStruct(const GRState *state,
const TypedRegion* R) {
QualType T = R->getValueType(getContext());
assert(T->isStructureType());
const RecordType* RT = T->getAsStructureType();
RecordDecl* RD = RT->getDecl();
assert(RD->isDefinition());
(void)RD;
#if USE_EXPLICIT_COMPOUND
llvm::ImmutableList<SVal> StructVal = getBasicVals().getEmptySValList();
// FIXME: We shouldn't use a std::vector. If RecordDecl doesn't have a
// reverse iterator, we should implement one.
std::vector<FieldDecl *> Fields(RD->field_begin(), RD->field_end());
for (std::vector<FieldDecl *>::reverse_iterator Field = Fields.rbegin(),
FieldEnd = Fields.rend();
Field != FieldEnd; ++Field) {
FieldRegion* FR = MRMgr.getFieldRegion(*Field, R);
QualType FTy = (*Field)->getType();
SVal FieldValue = Retrieve(state, loc::MemRegionVal(FR), FTy).getSVal();
StructVal = getBasicVals().consVals(FieldValue, StructVal);
}
return ValMgr.makeCompoundVal(T, StructVal);
#else
return ValMgr.makeLazyCompoundVal(state, R);
#endif
}
SVal RegionStoreManager::RetrieveArray(const GRState *state,
const TypedRegion * R) {
#if USE_EXPLICIT_COMPOUND
QualType T = R->getValueType(getContext());
ConstantArrayType* CAT = cast<ConstantArrayType>(T.getTypePtr());
llvm::ImmutableList<SVal> ArrayVal = getBasicVals().getEmptySValList();
uint64_t size = CAT->getSize().getZExtValue();
for (uint64_t i = 0; i < size; ++i) {
SVal Idx = ValMgr.makeArrayIndex(i);
ElementRegion* ER = MRMgr.getElementRegion(CAT->getElementType(), Idx, R,
getContext());
QualType ETy = ER->getElementType();
SVal ElementVal = Retrieve(state, loc::MemRegionVal(ER), ETy).getSVal();
ArrayVal = getBasicVals().consVals(ElementVal, ArrayVal);
}
return ValMgr.makeCompoundVal(T, ArrayVal);
#else
assert(isa<ConstantArrayType>(R->getValueType(getContext())));
return ValMgr.makeLazyCompoundVal(state, R);
#endif
}
//===----------------------------------------------------------------------===//
// Binding values to regions.
//===----------------------------------------------------------------------===//
Store RegionStoreManager::Remove(Store store, Loc L) {
const MemRegion* R = 0;
if (isa<loc::MemRegionVal>(L))
R = cast<loc::MemRegionVal>(L).getRegion();
if (R) {
RegionBindings B = GetRegionBindings(store);
return RBFactory.Remove(B, R).getRoot();
}
return store;
}
const GRState *RegionStoreManager::Bind(const GRState *state, Loc L, SVal V) {
if (isa<loc::ConcreteInt>(L))
return state;
// If we get here, the location should be a region.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
const MemRegion *R = cast<loc::MemRegionVal>(L).getRegion();
// Check if the region is a struct region.
if (const TypedRegion* TR = dyn_cast<TypedRegion>(R))
if (TR->getValueType(getContext())->isStructureType())
return BindStruct(state, TR, V);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Special case: the current region represents a cast and it and the super
// region both have pointer types or intptr_t types. If so, perform the
// bind to the super region.
// This is needed to support OSAtomicCompareAndSwap and friends or other
// loads that treat integers as pointers and vis versa.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (const ElementRegion *ER = dyn_cast<ElementRegion>(R)) {
if (ER->getIndex().isZeroConstant()) {
if (const TypedRegion *superR =
dyn_cast<TypedRegion>(ER->getSuperRegion())) {
ASTContext &Ctx = getContext();
QualType superTy = superR->getValueType(Ctx);
QualType erTy = ER->getValueType(Ctx);
if (IsAnyPointerOrIntptr(superTy, Ctx) &&
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
IsAnyPointerOrIntptr(erTy, Ctx)) {
SValuator::CastResult cr =
ValMgr.getSValuator().EvalCast(V, state, superTy, erTy);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
return Bind(cr.getState(), loc::MemRegionVal(superR), cr.getSVal());
}
// For now, just invalidate the fields of the struct/union/class.
// FIXME: Precisely handle the fields of the record.
if (superTy->isRecordType())
return InvalidateRegion(state, superR, NULL, 0);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
}
}
}
else if (const SymbolicRegion *SR = dyn_cast<SymbolicRegion>(R)) {
// Binding directly to a symbolic region should be treated as binding
// to element 0.
QualType T = SR->getSymbol()->getType(getContext());
T = T->getAs<PointerType>()->getPointeeType();
R = GetElementZeroRegion(SR, T);
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Perform the binding.
RegionBindings B = GetRegionBindings(state->getStore());
return state->makeWithStore(RBFactory.Add(B, R, V).getRoot());
}
const GRState *RegionStoreManager::BindDecl(const GRState *ST,
const VarDecl *VD,
const LocationContext *LC,
SVal InitVal) {
QualType T = VD->getType();
VarRegion* VR = MRMgr.getVarRegion(VD, LC);
if (T->isArrayType())
return BindArray(ST, VR, InitVal);
if (T->isStructureType())
return BindStruct(ST, VR, InitVal);
return Bind(ST, ValMgr.makeLoc(VR), InitVal);
}
// FIXME: this method should be merged into Bind().
const GRState *
RegionStoreManager::BindCompoundLiteral(const GRState *state,
const CompoundLiteralExpr* CL,
SVal V) {
CompoundLiteralRegion* R = MRMgr.getCompoundLiteralRegion(CL);
return Bind(state, loc::MemRegionVal(R), V);
}
const GRState *RegionStoreManager::BindArray(const GRState *state,
const TypedRegion* R,
SVal Init) {
QualType T = R->getValueType(getContext());
ConstantArrayType* CAT = cast<ConstantArrayType>(T.getTypePtr());
QualType ElementTy = CAT->getElementType();
uint64_t size = CAT->getSize().getZExtValue();
// Check if the init expr is a StringLiteral.
if (isa<loc::MemRegionVal>(Init)) {
const MemRegion* InitR = cast<loc::MemRegionVal>(Init).getRegion();
const StringLiteral* S = cast<StringRegion>(InitR)->getStringLiteral();
const char* str = S->getStrData();
unsigned len = S->getByteLength();
unsigned j = 0;
// Copy bytes from the string literal into the target array. Trailing bytes
// in the array that are not covered by the string literal are initialized
// to zero.
for (uint64_t i = 0; i < size; ++i, ++j) {
if (j >= len)
break;
SVal Idx = ValMgr.makeArrayIndex(i);
ElementRegion* ER = MRMgr.getElementRegion(ElementTy, Idx, R,
getContext());
SVal V = ValMgr.makeIntVal(str[j], sizeof(char)*8, true);
state = Bind(state, loc::MemRegionVal(ER), V);
}
return state;
}
// Handle lazy compound values.
if (nonloc::LazyCompoundVal *LCV = dyn_cast<nonloc::LazyCompoundVal>(&Init))
return CopyLazyBindings(*LCV, state, R);
// Remaining case: explicit compound values.
nonloc::CompoundVal& CV = cast<nonloc::CompoundVal>(Init);
nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
uint64_t i = 0;
for (; i < size; ++i, ++VI) {
// The init list might be shorter than the array length.
if (VI == VE)
break;
SVal Idx = ValMgr.makeArrayIndex(i);
ElementRegion* ER = MRMgr.getElementRegion(ElementTy, Idx, R, getContext());
if (CAT->getElementType()->isStructureType())
state = BindStruct(state, ER, *VI);
else
// FIXME: Do we need special handling of nested arrays?
state = Bind(state, ValMgr.makeLoc(ER), *VI);
}
// If the init list is shorter than the array length, set the array default
// value.
if (i < size) {
if (ElementTy->isIntegerType()) {
SVal V = ValMgr.makeZeroVal(ElementTy);
state = setDefaultValue(state, R, V);
}
}
return state;
}
const GRState *
RegionStoreManager::BindStruct(const GRState *state, const TypedRegion* R,
SVal V) {
if (!Features.supportsFields())
return state;
QualType T = R->getValueType(getContext());
assert(T->isStructureType());
const RecordType* RT = T->getAs<RecordType>();
RecordDecl* RD = RT->getDecl();
if (!RD->isDefinition())
return state;
// Handle lazy compound values.
if (const nonloc::LazyCompoundVal *LCV = dyn_cast<nonloc::LazyCompoundVal>(&V))
return CopyLazyBindings(*LCV, state, R);
// We may get non-CompoundVal accidentally due to imprecise cast logic.
// Ignore them and kill the field values.
if (V.isUnknown() || !isa<nonloc::CompoundVal>(V))
return KillStruct(state, R);
nonloc::CompoundVal& CV = cast<nonloc::CompoundVal>(V);
nonloc::CompoundVal::iterator VI = CV.begin(), VE = CV.end();
RecordDecl::field_iterator FI, FE;
for (FI = RD->field_begin(), FE = RD->field_end(); FI != FE; ++FI, ++VI) {
if (VI == VE)
break;
QualType FTy = (*FI)->getType();
const FieldRegion* FR = MRMgr.getFieldRegion(*FI, R);
if (FTy->isArrayType())
state = BindArray(state, FR, *VI);
else if (FTy->isStructureType())
state = BindStruct(state, FR, *VI);
else
state = Bind(state, ValMgr.makeLoc(FR), *VI);
}
// There may be fewer values in the initialize list than the fields of struct.
if (FI != FE)
state = setDefaultValue(state, R, ValMgr.makeIntVal(0, false));
return state;
}
const GRState *RegionStoreManager::KillStruct(const GRState *state,
const TypedRegion* R){
// Set the default value of the struct region to "unknown".
state = state->set<RegionDefaultValue>(R, UnknownVal());
// Remove all bindings for the subregions of the struct.
Store store = state->getStore();
RegionBindings B = GetRegionBindings(store);
for (RegionBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) {
const MemRegion* R = I.getKey();
if (const SubRegion* subRegion = dyn_cast<SubRegion>(R))
if (subRegion->isSubRegionOf(R))
store = Remove(store, ValMgr.makeLoc(subRegion));
}
return state->makeWithStore(store);
}
const GRState *RegionStoreManager::setDefaultValue(const GRState *state,
const MemRegion* R, SVal V) {
return state->set<RegionDefaultValue>(R, V);
}
const GRState*
RegionStoreManager::CopyLazyBindings(nonloc::LazyCompoundVal V,
const GRState *state,
const TypedRegion *R) {
// Nuke the old bindings stemming from R.
RegionBindings B = GetRegionBindings(state->getStore());
RegionDefaultBindings DVM = state->get<RegionDefaultValue>();
RegionDefaultBindings::Factory &DVMFactory =
state->get_context<RegionDefaultValue>();
llvm::OwningPtr<RegionStoreSubRegionMap>
SubRegions(getRegionStoreSubRegionMap(state));
// B and DVM are updated after the call to RemoveSubRegionBindings.
RemoveSubRegionBindings(B, DVM, DVMFactory, R, *SubRegions.get());
// Now copy the bindings. This amounts to just binding 'V' to 'R'. This
// results in a zero-copy algorithm.
return state->makeWithStore(RBFactory.Add(B, R, V).getRoot());
}
//===----------------------------------------------------------------------===//
// State pruning.
//===----------------------------------------------------------------------===//
namespace {
class VISIBILITY_HIDDEN RBDNode
: public std::pair<const GRState*, const MemRegion *> {
public:
RBDNode(const GRState *st, const MemRegion *r)
: std::pair<const GRState*, const MemRegion*>(st, r) {}
const GRState *getState() const { return first; }
const MemRegion *getRegion() const { return second; }
};
enum VisitFlag { NotVisited = 0, VisitedFromSubRegion, VisitedFromSuperRegion };
class RBDItem : public RBDNode {
private:
const VisitFlag VF;
public:
RBDItem(const GRState *st, const MemRegion *r, VisitFlag vf)
: RBDNode(st, r), VF(vf) {}
VisitFlag getVisitFlag() const { return VF; }
};
} // end anonymous namespace
void RegionStoreManager::RemoveDeadBindings(GRState &state, Stmt* Loc,
SymbolReaper& SymReaper,
llvm::SmallVectorImpl<const MemRegion*>& RegionRoots)
{
Store store = state.getStore();
RegionBindings B = GetRegionBindings(store);
RegionDefaultBindings DVM = state.get<RegionDefaultValue>();
// The backmap from regions to subregions.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
llvm::OwningPtr<RegionStoreSubRegionMap>
SubRegions(getRegionStoreSubRegionMap(&state));
// Do a pass over the regions in the store. For VarRegions we check if
// the variable is still live and if so add it to the list of live roots.
// For other regions we populate our region backmap.
llvm::SmallVector<const MemRegion*, 10> IntermediateRoots;
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Scan the direct bindings for "intermediate" roots.
for (RegionBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) {
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
const MemRegion *R = I.getKey();
IntermediateRoots.push_back(R);
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Scan the default bindings for "intermediate" roots.
for (RegionDefaultBindings::iterator I = DVM.begin(), E = DVM.end();
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
I != E; ++I) {
const MemRegion *R = I.getKey();
IntermediateRoots.push_back(R);
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
// Process the "intermediate" roots to find if they are referenced by
// real roots.
llvm::SmallVector<RBDItem, 10> WorkList;
llvm::DenseMap<const MemRegion*,unsigned> IntermediateVisited;
while (!IntermediateRoots.empty()) {
const MemRegion* R = IntermediateRoots.back();
IntermediateRoots.pop_back();
unsigned &visited = IntermediateVisited[R];
if (visited)
continue;
visited = 1;
if (const VarRegion* VR = dyn_cast<VarRegion>(R)) {
if (SymReaper.isLive(Loc, VR->getDecl()))
WorkList.push_back(RBDItem(&state, VR, VisitedFromSuperRegion));
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
continue;
}
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (const SymbolicRegion* SR = dyn_cast<SymbolicRegion>(R)) {
if (SymReaper.isLive(SR->getSymbol()))
WorkList.push_back(RBDItem(&state, SR, VisitedFromSuperRegion));
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
continue;
}
// Add the super region for R to the worklist if it is a subregion.
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
if (const SubRegion* superR =
dyn_cast<SubRegion>(cast<SubRegion>(R)->getSuperRegion()))
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
IntermediateRoots.push_back(superR);
}
// Enqueue the RegionRoots onto WorkList.
for (llvm::SmallVectorImpl<const MemRegion*>::iterator I=RegionRoots.begin(),
E=RegionRoots.end(); I!=E; ++I) {
WorkList.push_back(RBDItem(&state, *I, VisitedFromSuperRegion));
}
RegionRoots.clear();
// Process the worklist.
typedef llvm::DenseMap<std::pair<const GRState*, const MemRegion*>, VisitFlag>
VisitMap;
VisitMap Visited;
while (!WorkList.empty()) {
RBDItem N = WorkList.back();
WorkList.pop_back();
// Have we visited this node before?
VisitFlag &VF = Visited[N];
if (VF >= N.getVisitFlag())
continue;
const MemRegion *R = N.getRegion();
const GRState *state_N = N.getState();
// Enqueue subregions?
if (N.getVisitFlag() == VisitedFromSuperRegion) {
RegionStoreSubRegionMap *M;
if (&state == state_N)
M = SubRegions.get();
else {
RegionStoreSubRegionMap *& SM = SC[state_N];
if (!SM)
SM = getRegionStoreSubRegionMap(state_N);
M = SM;
}
RegionStoreSubRegionMap::iterator I, E;
for (llvm::tie(I, E) = M->begin_end(R); I != E; ++I)
WorkList.push_back(RBDItem(state_N, *I, VisitedFromSuperRegion));
}
// At this point, if we have already visited this region before, we are
// done.
if (VF != NotVisited) {
VF = N.getVisitFlag();
continue;
}
VF = N.getVisitFlag();
// Enqueue the super region.
if (const SubRegion *SR = dyn_cast<SubRegion>(R)) {
const MemRegion *superR = SR->getSuperRegion();
if (!isa<MemSpaceRegion>(superR)) {
// If 'R' is a field or an element, we want to keep the bindings
// for the other fields and elements around. The reason is that
// pointer arithmetic can get us to the other fields or elements.
VisitFlag NewVisit =
isa<FieldRegion>(R) || isa<ElementRegion>(R) || isa<ObjCIvarRegion>(R)
? VisitedFromSuperRegion : VisitedFromSubRegion;
WorkList.push_back(RBDItem(state_N, superR, NewVisit));
}
}
// Mark the symbol for any live SymbolicRegion as "live". This means we
// should continue to track that symbol.
if (const SymbolicRegion* SymR = dyn_cast<SymbolicRegion>(R))
SymReaper.markLive(SymR->getSymbol());
Store store_N = state_N->getStore();
RegionBindings B_N = GetRegionBindings(store_N);
// Get the data binding for R (if any).
const SVal* Xptr = B_N.lookup(R);
// Check for lazy bindings.
if (const nonloc::LazyCompoundVal *V =
dyn_cast_or_null<nonloc::LazyCompoundVal>(Xptr)) {
const LazyCompoundValData *D = V->getCVData();
WorkList.push_back(RBDItem(D->getState(), D->getRegion(),
VisitedFromSuperRegion));
}
else {
// No direct binding? Get the default binding for R (if any).
if (!Xptr) {
RegionDefaultBindings DVM_N = &state == state_N ? DVM
: state_N->get<RegionDefaultValue>();
Xptr = DVM_N.lookup(R);
}
// Direct or default binding?
if (Xptr) {
SVal X = *Xptr;
// Update the set of live symbols.
for (SVal::symbol_iterator SI=X.symbol_begin(), SE=X.symbol_end();
SI!=SE;++SI)
SymReaper.markLive(*SI);
// If X is a region, then add it to the worklist.
if (const MemRegion *RX = X.getAsRegion())
WorkList.push_back(RBDItem(state_N, RX, VisitedFromSuperRegion));
}
}
}
// We have now scanned the store, marking reachable regions and symbols
// as live. We now remove all the regions that are dead from the store
// as well as update DSymbols with the set symbols that are now dead.
for (RegionBindings::iterator I = B.begin(), E = B.end(); I != E; ++I) {
const MemRegion* R = I.getKey();
// If this region live? Is so, none of its symbols are dead.
if (Visited.find(std::make_pair(&state, R)) != Visited.end())
continue;
// Remove this dead region from the store.
store = Remove(store, ValMgr.makeLoc(R));
// Mark all non-live symbols that this region references as dead.
if (const SymbolicRegion* SymR = dyn_cast<SymbolicRegion>(R))
SymReaper.maybeDead(SymR->getSymbol());
SVal X = I.getData();
SVal::symbol_iterator SI = X.symbol_begin(), SE = X.symbol_end();
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
for (; SI != SE; ++SI)
SymReaper.maybeDead(*SI);
}
// Remove dead 'default' bindings.
RegionDefaultBindings NewDVM = DVM;
RegionDefaultBindings::Factory &DVMFactory =
state.get_context<RegionDefaultValue>();
for (RegionDefaultBindings::iterator I = DVM.begin(), E = DVM.end();
I != E; ++I) {
const MemRegion *R = I.getKey();
// If this region live? Is so, none of its symbols are dead.
if (Visited.find(std::make_pair(&state, R)) != Visited.end())
continue;
// Remove this dead region.
NewDVM = DVMFactory.Remove(NewDVM, R);
// Mark all non-live symbols that this region references as dead.
if (const SymbolicRegion* SymR = dyn_cast<SymbolicRegion>(R))
SymReaper.maybeDead(SymR->getSymbol());
SVal X = I.getData();
SVal::symbol_iterator SI = X.symbol_begin(), SE = X.symbol_end();
for (; SI != SE; ++SI)
SymReaper.maybeDead(*SI);
}
// Write the store back.
state.setStore(store);
// Write the updated default bindings back.
// FIXME: Right now this involves a fetching of a persistent state.
// We can do better.
if (DVM != NewDVM)
state.setGDM(state.set<RegionDefaultValue>(NewDVM)->getGDM());
}
//===----------------------------------------------------------------------===//
// Utility methods.
//===----------------------------------------------------------------------===//
void RegionStoreManager::print(Store store, llvm::raw_ostream& OS,
const char* nl, const char *sep) {
RegionBindings B = GetRegionBindings(store);
This is a fairly large patch, which resulted from a cascade of changes made to RegionStore (and related classes) in order to handle some analyzer failures involving casts and manipulation of symbolic memory. The root of the change is in StoreManager::CastRegion(). Instead of using ad hoc heuristics to decide when to layer an ElementRegion on a casted MemRegion, we now always layer an ElementRegion when the cast type is different than the original type of the region. This carries the current cast information associated with a region around without resorting to the error prone recording of "casted types" in GRState. Along with this new policy of layering ElementRegions, I added a new algorithm to strip away existing ElementRegions when they simply represented casts of a base memory object. This algorithm computes the raw "byte offset" that an ElementRegion represents from the base region, and allows the new ElementRegion to be based off that offset. The added benefit is that this naturally handles a series of casts of a MemRegion without building up a set of redundant ElementRegions (thus canonicalizing the region view). Other related changes that cascaded from this one (as tests were failing in RegionStore): - Revamped RegionStoreManager::InvalidateRegion() to completely remove all bindings and default values from a region and all subregions. Now invalidated fields are not bound directly to new symbolic values; instead the base region has a "default" symbol value from which "derived symbols" can be created. The main advantage of this approach is that it allows us to invalidate a region hierarchy and then lazily instantiate new values no matter how deep the hierarchy went (i.e., regardless of the number of field accesses, e.g. x->f->y->z->...). The previous approach did not do this. - Slightly reworked RegionStoreManager::RemoveDeadBindings() to also incorporate live symbols and live regions that do not have direct bindings but also have "default values" used for lazy instantiation. The changes to 'InvalidateRegion' revealed that these were necessary in order to achieve lazy instantiation of values in the region store with those bindings being removed too early. - The changes to InvalidateRegion() and RemoveDeadBindings() revealed a serious bug in 'getSubRegionMap()' where not all region -> subregion relationships involved in actually bindings (explicit and implicit) were being recorded. This has been fixed by using a worklist algorithm to iteratively fill in the region map. - Added special support to RegionStoreManager::Bind()/Retrieve() to handle OSAtomicCompareAndSwap in light of the new 'CastRegion' changes and the layering of ElementRegions. - Fixed a bug in SymbolReaper::isLive() where derived symbols were not being marked live if the symbol they were derived from was also live. This fix was critical for getting lazy instantiation in RegionStore to work. - Tidied up the implementation of ValueManager::getXXXSymbolVal() methods to use SymbolManager::canSymbolicate() to decide whether or not a symbol should be symbolicated. - 'test/Analysis/misc-ps-xfail.m' now passes; that test case has been moved to 'test/Analysis/misc-ps.m'. - Tweaked some pretty-printing of MemRegions, and implemented 'ElementRegion::getRawOffset()' for use with the CastRegion changes. llvm-svn: 77782
2009-08-01 14:17:29 +08:00
OS << "Store (direct bindings):" << nl;
for (RegionBindings::iterator I = B.begin(), E = B.end(); I != E; ++I)
OS << ' ' << I.getKey() << " : " << I.getData() << nl;
}