This change was a lot bigger than I originally anticipated; among
other things it requires us storing more information in the CFG to
record what block-level expressions need to be evaluated as lvalues.
The big change is that CFGBlocks no longer contain Stmt*'s by
CFGElements. Currently CFGElements just wrap Stmt*, but they also
store a bit indicating whether the block-level expression should be
evalauted as an lvalue. DeclStmts involving the initialization of a
reference require us treating the initialization expression as an
lvalue, even though that information isn't recorded in the AST.
Conceptually this change isn't that complicated, but it required
bubbling up the data through the CFGBuilder, to GRCoreEngine, and
eventually to GRExprEngine.
The addition of CFGElement is also useful for when we want to handle
more control-flow constructs or other data we want to keep in the CFG
that isn't represented well with just a block of statements.
In GRExprEngine, this patch introduces logic for evaluating the
lvalues of references, which currently retrieves the internal "pointer
value" that the reference represents. EvalLoad does a two stage load
to catch null dereferences involving an invalid reference (although
this could possibly be caught earlier during the initialization of a
reference).
Symbols are currently symbolicated using the reference type, instead
of a pointer type, and special handling is required creating
ElementRegions that layer on SymbolicRegions (see the changes to
RegionStoreManager).
Along the way, the DeadStoresChecker also silences warnings involving
dead stores to references. This was the original change I introduced
(which I wrote test cases for) that I realized caused GRExprEngine to
crash.
llvm-svn: 91501
Remove isPod() from DenseMapInfo, splitting it out to its own
isPodLike type trait. This is a generally useful type trait for
more than just DenseMap, and we really care about whether something
acts like a pod, not whether it really is a pod.
llvm-svn: 91422
now, don't construct CFGs that contain C++ try/catch statements, and
have GRExprEngine abort a path if it encounters a C++ construct it
doesn't understand (which is mostly everything at this point).
llvm-svn: 91389
Otherwise, even when real evaluation occurs, the previous fake auto
transitions would still be in the destination set, causing fake state
bifurcation.
llvm-svn: 90967
by the test case in PR 5627. Essentially we shouldn't clear the
ExplodedNodeSet where we deposit newly constructed nodes if that set
is the 'Dst' set passed in. It is not okay to clear that set because
it may already contain nodes.
llvm-svn: 90931
- Refactor the MemRegion hierarchy to distinguish between different StackSpaceRegions for locals and parameters.
- VarRegions for "captured" variables now have the BlockDataRegion as their super region (except those passed by reference)
- Add transfer function support to GRExprEngine for BlockDeclRefExprs.
This change also supports analyzing blocks as an analysis entry point
(top-of-the-stack), which required pushing more context-sensitivity
around in the MemRegion hierarchy via the use of LocationContext
objects. Functionally almost everything is the same, except we track
LocationContexts in a few more areas and StackSpaceRegions now refer
to a StackFrameContext object. In the future we will need to modify
MemRegionManager to allow multiple StackSpaceRegions in flight at once
(for the analysis of multiple stack frames).
llvm-svn: 90809
handler to this interface.
GRExprEngine::CheckerEvalCall() will return true if one of the checkers has
processed the node. In the future this might return void when we have some
default checker.
llvm-svn: 90755
we don't need to use the DoneEvaluation hack when check for
ObjCMessageExpr.
PreVisitObjCMessageExpr() only checks for undefined receiver or arguments.
Add checker interface EvalNilReceiver(). This is a 'once-and-done' interface.
llvm-svn: 90296
the set of variables "captured" by a block. Until the analysis gets
more sophisticated, for now we stop the retain count tracking of any
objects (transitively) referenced by these variables.
llvm-svn: 89929
This required two changes:
1) Added 'getReferencedgetReferencedBlockVars()' to AnalysisContext so
that clients can iterate over the "captured" variables in a block.
2) Modified LiveVariables to take an AnalysisContext& in its
constructor and to call getReferencedgetReferencedBlockVars() when it
processes a BlockExpr*.
llvm-svn: 89924
'BlockDataRegion' to distinguish between the code associated with a
block (which is represented by 'BlockTextRegion') and an instance of a
block, which includes both code and data. 'BlockDataRegion' has an
associated LocationContext, which can be used to eventually model the
lifetime of a block object once LocationContexts can represent scopes
(and iterations around a loop, etc.).
llvm-svn: 89900
only stop processing the checkers after all the nodes for a current
check have been processed. This (I believe) handles the case where
PredSet (the input nodes) contains more than one node due to state
bifurcation. Zhongxing: can you review this?
llvm-svn: 89882
initial transition of the nil-receiver checker to the Checker
interface as done in r89745. Some important changes include:
1) We consolidate the BugType object used for nil receiver bug
reports, and don't include the type of the returned value in the
BugType (which would be wrong if a nil receiver bug was reported more
than once)
2) Added a new (temporary) flag to CheckerContext: DoneEvauating.
This is used by GRExprEngine when evaluating message expressions to
not continue evaluating the message expression if this flag is set.
This flag is currently set by the nil receiver checker. This is an
intermediate solution to allow the nil-receiver checker to properly
work as a plug-in outside of GRExprEngine. Basically, this flag
indicates that the entire message expression has been evaluated, not
just a precondition (which is what the nil-receiver checker does).
This flag *should not* be repurposed for general use, but just to pull
more things out of GRExprEngine that already in there as we devise a
better interface in the Checker class.
3) Cleaned up the logic in the nil-receiver checker, making the
control-flow a lot easier to read.
llvm-svn: 89804
along the way. Important changes:
1) To generate a sink node, use GenerateSink(); GenerateNode() is for
generating regular transitions. This makes the API clearer and also
allows us to use the 'bool' option to GenerateNode() for a different
purpose.
2) GenerateNode() now automatically adds the generated node to the
destination ExplodedNodeSet (autotransition) unless the client
specifies otherwise with a bool flag. Several checkers did not call
'addTransition()' after calling 'GenerateNode()', causing the
simulation path to be prematurely culled when a non-fail stop bug was
encountered.
3) Add variants of GenerateNode()/GenerateSink() that take neither a
Stmt* or a GRState*; most callers of GenerateNode() just pass in the
same Stmt* as provided when the CheckerContext object is created; we
can just use that the majority of the time. This cleanup also allows
us to potentially coelesce the APIs for evaluating branches and
end-of-paths (which currently directly use builders).
4) addTransition() no longer needs to be called except for a few
cases. We now have a variant of addTransition() that takes a
GRState*; this allows one to propagate the updated state without
caring about generating a new node explicitly. This nicely cleaned up
a bunch of cases that called autoTransition() with a bunch of
conditional logic surround the call (that common logic has now been
swallowed up by addTransition() itself).
llvm-svn: 89707
sugared types. The basic problem is that our qualifier accessors
(getQualifiers, getCVRQualifiers, isConstQualified, etc.) only look at
the current QualType and not at any qualifiers that come from sugared
types, meaning that we won't see these qualifiers through, e.g.,
typedefs:
typedef const int CInt;
typedef CInt Self;
Self.isConstQualified() currently returns false!
Various bugs (e.g., PR5383) have cropped up all over the front end due
to such problems. I'm addressing this problem by splitting each
qualifier accessor into two versions:
- the "local" version only returns qualifiers on this particular
QualType instance
- the "normal" version that will eventually combine qualifiers from this
QualType instance with the qualifiers on the canonical type to
produce the full set of qualifiers.
This commit adds the local versions and switches a few callers from
the "normal" version (e.g., isConstQualified) over to the "local"
version (e.g., isLocalConstQualified) when that is the right thing to
do, e.g., because we're printing or serializing the qualifiers. Also,
switch a bunch of
Context.getCanonicalType(T1).getUnqualifiedType() == Context.getCanonicalType(T2).getQualifiedType()
expressions over to
Context.hasSameUnqualifiedType(T1, T2)
llvm-svn: 88969
* Add a load type to GRExprEngine::EvalLoad().
* When retrieve from 'theValue' of OSAtomic funcitions, use the type of the
region instead of the argument expression as the load type.
* Then we can convert CastRetrievedSVal to a pure assertion. In the future
we can let all Retrieve() methods simply return SVal.
llvm-svn: 88888
- Eliminates many calls to std::string.c_str()
- Fixes an invalid read in ReturnStackAddressChecker due to an unsafe call to
StringRef.data() which doesn't guarantee null-termination.
llvm-svn: 88779
the old builder API. This percolated a bunch of changes up to the
Checker class (where CheckLocation has been renamed VisitLocation) and
GRExprEngine. ProgramPoint now has the notion of a "LocationCheck"
point (with PreLoad and PreStore respectively), and a bunch of the old
ProgramPoints that are no longer used have been removed.
llvm-svn: 86798
This is reasonable because people know what they are doing when they
intentionally dereference the pointer.
So now we only emit warning when a pointer variable is use literally.
llvm-svn: 86673
ArrayType>()) does not instantiate. Update all callers that used this
unsafe feature to use the appropriate ASTContext::getAs*ArrayType method.
llvm-svn: 86596
value into their own respective subclasses of Checker (and put them in .cpp files where their
implementation details are hidden from GRExprEngine).
llvm-svn: 86215
an "assign expression", representing the expressions where the value
binding occurs and the assignment takes place respectively. These are
largely syntactic clues for better error reporting.
llvm-svn: 86084
Split it to two checkers, one for undefined size,
the other for zero size, so that we don't need to query the size
when emitting the bug report.
llvm-svn: 85895
by Zhongxing Xu. RemoveDeadBindings() would falsely prune
SymbolicRegions from the store that wrapped derived symbols whose
liveness could only be determined after scanning the store.
llvm-svn: 85484
qualified reference to a declaration that is not a non-static data
member or non-static member function, e.g.,
namespace N { int i; }
int j = N::i;
Instead, extend DeclRefExpr to optionally store the qualifier. Most
clients won't see or care about the difference (since
QualifierDeclRefExpr inherited DeclRefExpr). However, this reduces the
number of top-level expression types that clients need to cope with,
brings the implementation of DeclRefExpr into line with MemberExpr,
and simplifies and unifies our handling of declaration references.
Extended DeclRefExpr to (optionally) store explicitly-specified
template arguments. This occurs when naming a declaration via a
template-id (which will be stored in a TemplateIdRefExpr) that,
following template argument deduction and (possibly) overload
resolution, is replaced with a DeclRefExpr that refers to a template
specialization but maintains the template arguments as written.
llvm-svn: 84962
AnalysisManager periodically cleanup its AnalysisContextManager and LocationContextManager objects,
as they don't need to forever retain all the CFGs ever created when analyzing a file.
llvm-svn: 84684
RegionStoreManager::Retrieve() that was intended to handle conflated uses of pointers as integers.
It turns out this isn't needed, and resulted in inconsistent behavior when creating symbolic values on the following test case in 'tests/Analysis/misc-ps.m':
typedef struct _BStruct { void *grue; } BStruct;
void testB_aux(void *ptr);
void testB(BStruct *b) {
{
int *__gruep__ = ((int *)&((b)->grue));
int __gruev__ = *__gruep__;
testB_aux(__gruep__);
}
{
int *__gruep__ = ((int *)&((b)->grue));
int __gruev__ = *__gruep__;
if (~0 != __gruev__) {}
}
}
When the code was analyzed with '-arch x86_64', the value assigned to '__gruev__' be would be a
symbolic integer, but for '-arch i386' the value assigned to '__gruev__' would be a symbolic region
(a blob of memory). With this change the value created is always a symbolic integer.
Since the code being removed was added to support analysis of code calling
OSAtomicCompareAndSwapXXX(), I also modified 'test/Analysis/NSString.m' to analyze the code in both
'-arch i386' and '-arch x86_64', and also added some complementary test cases to test the presence
of leaks when using OSAtomicCompareAndSwap32Barrier()/OSAtomicCompareAndSwap64Barrier() instead of
just their absence. This code change reveals that previously both RegionStore and BasicStore were
handling these cases wrong, and would never cause the analyzer to emit a leak in these cases (false
negatives). Now RegionStore gets it right, but BasicStore still gets it wrong (and hence it has been
disabled temporarily for this test case).
llvm-svn: 84163
'CVPixelBufferCreateWithPlanarBytes()' and
'CVPixelBufferCreateWithBytes' (Core Video API) can indirectly release
a pixel buffer object via a callback.
This fixes <rdar://problem/7283567>.
llvm-svn: 84064
Speedup: when doing 'clang-cc -analyze -dump-cfg' (without actual printing, just
CFG building) on the amalgamated SQLite source (all of SQLite in one source
file), runtime reduced by 9%.
This fixes: <rdar://problem/7250745>
llvm-svn: 83899
default binding for regions. This allows us to simply a lot of code. A
further simplification could be done is that many methods of
regionstore can only work on Store instead of GRState.
llvm-svn: 83762
adding assert
This fix required a few changes:
SimpleSValuator:
- Eagerly replace a symbolic value with its constant value in EvalBinOpNN
when it is constrained to a constant. This allows us to better constant fold
values along a path.
- Handle trivial case of '<', '>' comparison of pointers when the two pointers
are exactly the same.
RegionStoreManager:
llvm-svn: 83358
concrete types. Use unqualified desugaring for getAs<> and sundry.
Fix a few users to either not desugar or use qualified desugar, as seemed
appropriate. Removed Type's qualified desugar method, as it was easy
to accidentally use instead of QualType's.
llvm-svn: 83116
identified with a false positive reported by Thomas Clement. This
involved doing another rewrite of
RegionStoreManager::RemoveDeadBindings(), which phrases the entire
problem of scanning for dead regions as a graph exploration problem.
It is more methodic than the previous implementation.
llvm-svn: 83053
are only specially treated by RegionStore::InvalidateRegion() when
their super region is also invalidated. When this isn't the case,
conjure a new symbol for a FieldRegion. Thanks to Zhongxing Xu and
Daniel Dunbar for pointing out this issue.
llvm-svn: 83043
<rdar://problem/6914474> checker doesn't realize that variable might
have been assigned if a pointer to that variable was passed to another
function via a structure
The problem here was the RegionStoreManager::InvalidateRegion didn't
invalidate the bindings of invalidated regions. This required a
rewrite of this method using a worklist.
As part of this fix, changed ValueManager::getConjuredSymbolVal() to
require a 'void*' SymbolTag argument. This tag is used to
differentiate two different symbols created at the same location.
llvm-svn: 82920
value-dependent. Audit (and fixed) all calls to
Expr::isNullPointerConstant() to provide the correct behavior with
value-dependent expressions. Fixes PR5041 and a crash in libstdc++
<locale>.
In the same vein, properly compute value- and type-dependence for
ChooseExpr. Fixes PR4996.
llvm-svn: 82748
have the iterators and operator[] handle the traversal of statements, as they
are stored in reverse order. Tests show this has no real performance impact, but
it does simply the CFG construction logic and will make it slightly easier to
change the allocation strategy for CFGBlocks (as we have fewer copies).
llvm-svn: 82702
The issue was a discrepancy between how RegionStoreManager::Bind() and
RegionStoreManager::Retrieve() derived the "key" for the first element
of a symbolic region.
llvm-svn: 82680
Several of the existing methods were identical to their respective
specializations, and so have been removed entirely. Several more 'leaf'
optimizations were introduced.
The getAsFoo() methods which imposed extra conditions, like
getAsObjCInterfacePointerType(), have been left in place.
llvm-svn: 82501
integer pointer. For now just invalidate the fields of the struct.
This addresses: <rdar://problem/7185607> [RegionStore] support invalidation of bit fields using integer assignment
llvm-svn: 82492
when running the analyzer on real projects. We'll keep the change to
AnalysisManager.cpp in r82198 so that -fobjc-gc analyzes code
correctly in both GC and non-GC modes, although this may emit two
diagnostics for each bug in some cases (a better solution will come
later).
llvm-svn: 82201
pruning of diagnostics that may be emitted multiple times. This is
accomplished by adding FoldingSet profiling support to PathDiagnostic,
and then having BugReporter record what diagnostics have been issued.
This was motived to a serious bug introduced by moving the
'divide-by-zero' checking outside of GRExprEngine into a separate
'Checker' class. When analyzing code using the '-fobjc-gc' option, a
given function would be analyzed twice, but the second time various
"internal checks" would be disabled to avoid emitting multiple
diagnostics (e.g., "null dereference") for the same issue. The
problem is that such checks also effect path pruning and don't just
emit diagnostics. This resulted in an assertion failure involving a
real divide-by-zero in some analyzed code where we would get an
assertion failure in APInt because the 'DivZero' check was disabled
and didn't prune the logic that resulted in the divide-by-zero in the
analyzer.
The implemented solution is somewhat of a hack, and may not perform
extremely well. This will need to be cleaned up over time.
As a regression test, 'misc-ps.m' has been modified so that its tests
are run using -fobjc-gc to test this diagnostic pruning behavior.
llvm-svn: 82198
__builtin_offsetof in the static analyzer that __builtin_offsetof is
not guaranteed to return an integer constant. We will need to shore
this up later, but now at least we have correct support for when this
*is* an integer constant.
llvm-svn: 81830
Ted Kremenek