Summary:
When processing a call to a function, which got passed less arguments than it
expects, the analyzer would crash.
I've also added a test for that and a analyzer warning which detects these
cases.
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D994
llvm-svn: 184288
This silences warnings that could occur when one is swapping partially initialized structs. We suppress
not only the assignments of uninitialized members, but any values inside swap because swap could
potentially be used as a subroutine to swap class members.
This silences a warning from std::try::function::swap() on partially initialized objects.
llvm-svn: 184256
The untemplated implementation of getParents() doesn't need to be in a
header file.
RecursiveASTVisitor.h is full of repeated macro expansion. Moving this
include to ASTContext.cpp speeds up compilation of
LambdaMangleContext.cpp, a small C++ file with few includes, from 3.7s
to 2.8s for me locally. I haven't measured a full build, but it can't
hurt.
I had to fix a few static analyzer files that were depending on
transitive includes of C++ AST headers.
Reviewers: rsmith, klimek
Differential Revision: http://llvm-reviews.chandlerc.com/D982
llvm-svn: 184075
Introduce CXXStdInitializerListExpr node, representing the implicit
construction of a std::initializer_list<T> object from its underlying array.
The AST representation of such an expression goes from an InitListExpr with a
flag set, to a CXXStdInitializerListExpr containing a MaterializeTemporaryExpr
containing an InitListExpr (possibly wrapped in a CXXBindTemporaryExpr).
This more detailed representation has several advantages, the most important of
which is that the new MaterializeTemporaryExpr allows us to directly model
lifetime extension of the underlying temporary array. Using that, this patch
*drastically* simplifies the IR generation of this construct, provides IR
generation support for nested global initializer_list objects, fixes several
bugs where the destructors for the underlying array would accidentally not get
invoked, and provides constant expression evaluation support for
std::initializer_list objects.
llvm-svn: 183872
Summary:
"register" functions for the checker were caching the checker objects in a
static variable. This caused problems when the function is called with a
different CheckerManager.
Reviewers: klimek
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D955
llvm-svn: 183823
We drew the diagnostic edges to wrong statements in cases the note was on a macro.
The fix is simple, but seems to work just fine for a whole bunch of test cases (plist-macros.cpp).
Also, removes an unnecessary edge in edges-new.mm, when function signature starts with a macro.
llvm-svn: 183599
The function in which we were doing it used to be conditionalized. Add a new unconditional
cleanup step.
This fixes PR16227 (radar://14073870) - a crash when generating html output for one of the test files.
llvm-svn: 183451
Previously our edges were completely broken here; now, the final result
is a very simple set of edges in most cases: one up to the "for" keyword
for context, and one into the body of the loop. This matches the behavior
for ObjC for-in loops.
In the AST, however, CXXForRangeStmts are handled very differently from
ObjCForCollectionStmts. Since they are specified in terms of equivalent
statements in the C++ standard, we actually have implicit AST nodes for
all of the semantic statements. This makes evaluation very easy, but
diagnostic locations a bit trickier. Fortunately, the problem can be
generally defined away by marking all of the implicit statements as
part of the top-level for-range statement.
One of the implicit statements in a for-range statement is the declaration
of implicit iterators __begin and __end. The CFG synthesizes two
separate DeclStmts to match each of these decls, but until now these
synthetic DeclStmts weren't in the function's ParentMap. Now, the CFG
keeps track of its synthetic statements, and the AnalysisDeclContext will
make sure to add them to the ParentMap.
<rdar://problem/14038483>
llvm-svn: 183449
When processing ArrayToPointerDecay, we expect the array to be a location, not a LazyCompoundVal.
Special case the rvalue arrays by using a location to represent them. This case is handled similarly
elsewhere in the code.
Fixes PR16206.
llvm-svn: 183359
We previously asserted that there was a top-level function entry edge, but
if the function decl's location is invalid (or within a macro) this edge
might not exist. Change the assertion to an actual check, and don't drop
the first path piece if it doesn't match.
<rdar://problem/14070304>
llvm-svn: 183358
The edge optimizer needs to see edges for, say, implicit casts (which have
the same source location as their operand) to uniformly simplify the
entire path. However, we still don't want to produce edges from a statement
to /itself/, which could occur when two nodes in a row have the same
statement location.
This necessitated moving the check for redundant notes to after edge
optimization, since the check relies on notes being adjacent in the path.
<rdar://problem/14061675>
llvm-svn: 183357
...but don't yet migrate over the existing plist tests. Some of these
would be trivial to migrate; others could use a bit of inspection first.
In any case, though, the new edge algorithm seems to have proven itself,
and we'd like more coverage (and more usage) of it going forwards.
llvm-svn: 183165
A.1 -> A -> B
becomes
A.1 -> B
This only applies if there's an edge from a subexpression to its parent
expression, and that is immediately followed by another edge from the
parent expression to a subsequent expression. Normally this is useful for
bringing the edges back to the left side of the code, but when the
subexpression is on a different line the backedge ends up looking strange,
and may even obscure code. In these cases, it's better to just continue
to the next top-level statement.
llvm-svn: 183164
Specifically, if the line is over 80 characters, or if the top-level
statement spans mulitple lines, we should preserve sub-expression edges
even if they form a simple cycle as described in the last commit, because
it's harder to infer what's going on than it is for shorter lines.
llvm-svn: 183163
Generating context arrows can result in quite a few arrows surrounding a
relatively simple expression, often containing only a single path note.
|
1 +--2---+
v/ v
auto m = new m // 3 (the path note)
|\ |
5 +--4---+
v
Note also that 5 and 1 are two ends of the "same" arrow, i.e. they go from
event to event. 3 is not an arrow but the path note itself.
Now, if we see a pair of edges like 2 and 4---where 4 is the reverse of 2
and there is optionally a single path note between them---we will
eliminate /both/ edges. Anything more complicated will be left as is
(more edges involved, an inlined call, etc).
The next commit will refine this to preserve the arrows in a larger
expression, so that we don't lose all context.
llvm-svn: 183162
The old edge builder didn't have a notion of nested statement contexts,
so there was no special treatment of a logical operator inside an if
(or inside another logical operator). The new edge builder always tries
to establish the full context up to the top-level statement, so it's
important to know how much context has been established already rather
than just checking the innermost context.
This restores some of the old behavior for the old edge generation:
the context of a logical operator's non-controlling expression is the
subexpression in the old edge algorithm, but the entire operator
expression in the new algorithm.
llvm-svn: 183160
The current edge-generation algorithm sometimes creates edges from a
top-level statement A to a sub-expression B.1 that's not at the start of B.
This creates a "swoosh" effect where the arrow is drawn on top of the
text at the start of B. In these cases, the results are clearer if we see
an edge from A to B, then another one from B to B.1.
Admittedly, this does create a /lot/ of arrows, some of which merely hop
into a subexpression and then out again for a single note. The next commit
will eliminate these if the subexpression is simple enough.
This updates and reuses some of the infrastructure from the old edge-
generation algorithm to find the "enclosing statement" context for a
given expression. One change in particular marks the context of the
LHS or RHS of a logical binary operator (&&, ||) as the entire operator
expression, rather than the subexpression itself. This matches our behavior
for ?:, and allows us to handle nested context information.
<rdar://problem/13902816>
llvm-svn: 183159
Although we don't want to show a function entry edge for a top-level path,
having it makes optimizing edges a little more uniform.
This does not affect any edges now, but will affect context edge generation
(next commit).
llvm-svn: 183158
Jordan has pointed out that it is valuable to warn in cases when the arguments to init escape.
For example, NSData initWithBytes id not going to free the memory.
llvm-svn: 183062
In many cases, the edge from the "if" to the condition, followed by an edge from the branch condition to the target code, is uninteresting.
In such cases, we should fold the two edges into one from the "if" to the target.
This also applies to loops.
Implements <rdar://problem/14034763>.
llvm-svn: 183018
...and make this work correctly in the current codebase.
After living on this for a while, it turns out to look very strange for
inlined functions that have only a single statement, and somewhat strange
for inlined functions in general (since they are still conceptually in the
middle of the path, and there is a function-entry path note).
It's worth noting that this only affects inlined functions; in the new
arrow generation algorithm, the top-level function still starts at the
first real statement in the function body, not the enclosing CompoundStmt.
This reverts r182078 / dbfa950abe0e55b173286a306ee620eff5f72ea.
llvm-svn: 182963
It is okay to declare a block without an argument list: ^ {} or ^void {}.
In these cases, the BlockDecl's signature-as-written will just contain
the return type, rather than the entire function type. It is unclear if
this is intentional, but the analyzer shouldn't crash because of it.
<rdar://problem/14018351>
llvm-svn: 182948
Most loop notes (like "entering loop body") are attached to the condition
expression guarding a loop or its equivalent. For loops may not have a
condition expression, though. Rather than crashing, just use the entire
ForStmt as the location. This is probably the best we can do.
<rdar://problem/14016063>
llvm-svn: 182904
In C, 'void' is treated like any other incomplete type, and though it is
never completed, you can cast the address of a void-typed variable to do
something useful. (In C++ it's illegal to declare a variable with void type.)
Previously we asserted on this code; now we just treat it like any other
incomplete type.
And speaking of incomplete types, we don't know their extent. Actually
check that in TypedValueRegion::getExtent, though that's not being used
by any checkers that are on by default.
llvm-svn: 182880
This gives slightly better precision, specifically, in cases where a non-typed region represents the array
or when the type is a non-array type, which can happen when an array is a result of a reinterpret_cast.
llvm-svn: 182810
It’s important for us to reason about the cast as it is used in std::addressof. The reason we did not
handle the cast previously was a crash on a test case (see commit r157478). The crash was in
processing array to pointer decay when the region type was not an array. Address the issue, by
just returning an unknown in that case.
llvm-svn: 182808
In addition to enabling more code reuse, this suppresses some false positives by allowing us to
compare an element region to its base. See the ptr-arith.cpp test cases for an example.
llvm-svn: 182780
When generating path notes, implicit function bodies are shown at the call
site, so that, say, copying a POD type in C++ doesn't jump you to a header
file. This is especially important when the synthesized function itself
calls another function (or block), in which case we should try to jump the
user around as little as possible.
By checking whether a called function has a body in the AST, we can tell
if the analyzer synthesized the body, and if we should therefore collapse
the call down to the call site like a true implicitly-defined function.
<rdar://problem/13978414>
llvm-svn: 182677
The new edge algorithm would keep track of the previous location in each
location context, so that it could draw arrows coming in and out of each
inlined call. However, it tried to access the location of the call before
it was actually set (at the CallEnter node). This only affected
unterminated calls at the end of a path; calls with visible exit nodes
already had a valid location.
This patch ditches the location context map, since we're processing the
nodes in order anyway, and just unconditionally updates the PrevLoc
variable after popping out of an inlined call.
<rdar://problem/13983470>
llvm-svn: 182676
Currently, blocks instantiated in templates lose their "signature as
written"; it's not clear if this is intentional. Change the analyzer's
use of BlockDecl::getSignatureAsWritten to check whether or not the
signature is actually there.
<rdar://problem/13954714>
llvm-svn: 182497
The crash is triggered by the newly added option (-analyzer-config report-in-main-source-file=true) introduced in r182058.
Note, ideally, we’d like to report the issue within the main source file here as well.
For now, just do not crash.
llvm-svn: 182445
Ted and I spent a long time discussing this today and found out that neither
the existing code nor the new code was doing what either of us thought it
was, which is never good. The good news is we found a much simpler way to
fix the motivating test case (an ObjCSubscriptExpr).
This reverts r182083, but pieces of it will come back in subsequent commits.
llvm-svn: 182185
This optimizes some spurious edges resulting from PseudoObjectExprs.
This required far more changes than I anticipated. The current
ParentMap does not record any hierarchy information between
a PseudoObjectExpr and its *semantic* expressions that may be
wrapped in OpaqueValueExprs, which are the expressions actually
laid out in the CFG. This means the arrow pruning logic could
not map from an expression to its containing PseudoObjectExprs.
To solve this, this patch adds a variant of ParentMap that
returns the "semantic" parentage of expressions (essentially
as they are viewed by the CFG). This alternate ParentMap is then
used by the arrow reducing logic to identify edges into pseudo
object expressions, and then eliminate them.
llvm-svn: 182083
The analyzer can't see the reference count for shared_ptr, so it doesn't
know whether a given destruction is going to delete the referenced object.
This leads to spurious leak and use-after-free warnings.
For now, just ban destructors named '~shared_ptr', which catches
std::shared_ptr, std::tr1::shared_ptr, and boost::shared_ptr.
PR15987
llvm-svn: 182071
Previously, we’ve used the last location of the analyzer issue path as the location of the
report. This might not provide the best user experience, when one analyzer a source
file and the issue appears in the header. Introduce an option to use the last location
of the path that is in the main source file as the report location.
New option can be enabled with -analyzer-config report-in-main-source-file=true.
llvm-svn: 182058
This class is a StmtVisitor that distinguishes between block-level and
non-block-level statements in a CFG. However, it does so using a hard-coded
idea of which statements might be block-level, which probably isn't accurate
anymore. The only implementer of the CFGStmtVisitor hierarchy was the
analyzer's DeadStoresChecker, and the analyzer creates a linearized CFG
anyway (every non-trivial statement is a block-level statement).
This also allows us to remove the block-expr map ("BlkExprMap"), which
mapped statements to positions in the CFG. Apart from having a helper type
that really should have just been Optional<unsigned>, it was only being
used to ask /if/ a particular expression was block-level, for traversal
purposes in CFGStmtVisitor.
llvm-svn: 181945
ASTDumper was already trying to do this & instead got an implicit bool
conversion by surprise (thus printing out 0 or 1 instead of the name of
the declaration). To avoid that issue & simplify call sites, simply make
it the normal/expected operator<<(raw_ostream&, ...) overload & simplify
all the existing call sites. (bonus: this function doesn't need to be a
member or friend, it's just using public API in DeclarationName)
llvm-svn: 181832
This patch renames getLinkage to getLinkageInternal. Only code that
needs to handle UniqueExternalLinkage specially should call this.
Linkage, as defined in the c++ standard, is provided by
getFormalLinkage. It maps UniqueExternalLinkage to ExternalLinkage.
Most places in the compiler actually want isExternallyVisible, which
handles UniqueExternalLinkage as internal.
llvm-svn: 181677
In most cases it is, by just looking at the name. Also, this check prevents the heuristic from working in strange user settings.
radar://13839692
llvm-svn: 181615
Consider this example:
char *p = malloc(sizeof(char));
systemFunction(&p);
free(p);
In this case, when we call systemFunction, we know (because it's a system
function) that it won't free 'p'. However, we /don't/ know whether or not
it will /change/ 'p', so the analyzer is forced to invalidate 'p', wiping
out any bindings it contains. But now the malloc'd region looks like a
leak, since there are no more bindings pointing to it, and we'll get a
spurious leak warning.
The fix for this is to notice when something is becoming inaccessible due
to invalidation (i.e. an imperfect model, as opposed to being explicitly
overwritten) and stop tracking it at that point. Currently, the best way
to determine this for a call is the "indirect escape" pointer-escape kind.
In practice, all the patch does is take the "system functions don't free
memory" special case and limit it to direct parameters, i.e. just the
arguments to a call and not other regions accessible to them. This is a
conservative change that should only cause us to escape regions more
eagerly, which means fewer leak warnings.
This isn't perfect for several reasons, the main one being that this
example is treated the same as the one above:
char **p = malloc(sizeof(char *));
systemFunction(p + 1);
// leak
Currently, "addresses accessible by offsets of the starting region" and
"addresses accessible through bindings of the starting region" are both
considered "indirect" regions, hence this uniform treatment.
Another issue is our longstanding problem of not distinguishing const and
non-const bindings; if in the first example systemFunction's parameter were
a char * const *, we should know that the function will not overwrite 'p',
and thus we can safely report the leak.
<rdar://problem/13758386>
llvm-svn: 181607
The one user has been changed to use getLValue on the compound literal
expression and then use the normal bindLoc to assign a value. No need
to special case this in the StoreManager.
llvm-svn: 181214
This occurs because in C++11 the compound literal syntax can trigger a
constructor call via list-initialization. That is, "Point{x, y}" and
"(Point){x, y}" end up being equivalent. If this occurs, the inner
CXXConstructExpr will have already handled the object construction; the
CompoundLiteralExpr just needs to propagate that value forwards.
<rdar://problem/13804098>
llvm-svn: 181213
This change required some minor changes to LocationContextMap to have it map
from PathPieces to LocationContexts instead of PathDiagnosticCallPieces to
LocationContexts. These changes are in the other diagnostic
generation logic as well, but are functionally equivalent.
Interestingly, this optimize requires delaying "cleanUpLocation()" until
later; possibly after all edges have been optimized. This is because
we need PathDiagnosticLocations to refer to the semantic entity (e.g. a statement)
as long as possible. Raw source locations tell us nothing about
the semantic relationship between two locations in a path.
llvm-svn: 181084
FindLastStoreBRVisitor is responsible for finding where a particular region
gets its value; if the region is a VarRegion, it's possible that value was
assigned at initialization, i.e. at its DeclStmt. However, if a function is
called recursively, the same DeclStmt may be evaluated multiple times in
multiple stack frames. FindLastStoreBRVisitor was not taking this into
account and just picking the first one it saw.
<rdar://problem/13787723>
llvm-svn: 180997
There were actually two bugs here:
- if we decided to look for an interesting lvalue or call expression, we
wouldn't go find its node if we also knew we were at a (different) call.
- if we looked through one message send with a nil receiver, we thought we
were still looking at an argument to the original call.
Put together, this kept us from being able to track the right values, which
means sub-par diagnostics and worse false-positive suppression.
Noticed by inspection.
llvm-svn: 180996
BugReporter is used to process ALL bug reports. By using a shared map,
we are having mappings from different PathDiagnosticPieces to LocationContexts
well beyond the point where we are processing a given report. This
state is inherently error prone, and is analogous to using a global
variable. Instead, just create a temporary map, one per report,
and when we are done with it we throw it away. No extra state.
llvm-svn: 180974
...and don't consider '0' to be a null pointer constant if it's the
initializer for a float!
Apparently null pointer constant evaluation looks through both
MaterializeTemporaryExpr and ImplicitCastExpr, so we have to be more
careful about types in the callers. For RegionStore this just means giving
up a little more; for ExprEngine this means handling the
MaterializeTemporaryExpr case explicitly.
Follow-up to r180894.
llvm-svn: 180944
It is unfortunate that we have to mark these exceptions in multiple places.
This was already in CallEvent. I suppose it does let us be more precise
about saying /which/ arguments have their retain counts invalidated -- the
connection's is still valid even though the context object's isn't -- but
we're not tracking the retain count of XPC objects anyway.
<rdar://problem/13783514>
llvm-svn: 180904
Previously, this was scattered across Environment (literal expressions),
ExprEngine (default arguments), and RegionStore (global constants). The
former special-cased several kinds of simple constant expressions, while
the latter two deferred to the AST's constant evaluator.
Now, these are all unified as SValBuilder::getConstantVal(). To keep
Environment fast, the special cases for simple constant expressions have
been left in, but the main benefits are that (a) unusual constants like
ObjCStringLiterals now work as default arguments and global constant
initializers, and (b) we're not duplicating code between ExprEngine and
RegionStore.
This actually caught a bug in our test suite, which is awesome: we stop
tracking allocated memory if it's passed as an argument along with some
kind of callback, but not if the callback is 0. We were testing this in
a case where the callback parameter had a default value, but that value
was 0. After this change, the analyzer now (correctly) flags that as a
leak!
<rdar://problem/13773117>
llvm-svn: 180894
This goes with r178516, which instructed the analyzer not to inline the
constructors and destructors of C++ container classes. This goes a step
further and does the same thing for iterators, so that the analyzer won't
falsely decide we're trying to construct an iterator pointing to a
nonexistent element.
The heuristic for determining whether something is an iterator is the
presence of an 'iterator_category' member. This is controlled under the
same -analyzer-config option as container constructor/destructor inlining:
'c++-container-inlining'.
<rdar://problem/13770187>
llvm-svn: 180890
This doesn't appear to be the cause of the slowdown. I'll have to try a
manual bisect to see if there's really anything there, or if it's just
the bot itself taking on additional load. Meanwhile, this change helps
with correctness.
This changes an assertion and adds a test case, then re-applies r180638,
which was reverted in r180714.
<rdar://problem/13296133> and PR15863
llvm-svn: 180864
Much of this patch outside of PathDiagnostics.h are just minor
syntactic changes due to the return type for operator* and the like
changing for the iterator, so the real focus should be on
PathPieces itself.
This change is motivated so that we can do efficient insertion
and removal of individual pieces from within a PathPiece, just like
this was a kind of "IR" for static analyzer diagnostics. We
currently implement path transformations by iterating over an
entire PathPiece and making a copy. This isn't very natural for
some algorithms.
We use an ilist here instead of std::list because we want operations
to rip out/insert nodes in place, just like IR manipulation. This
isn't being used yet, but opens the door for more powerful
transformation algorithms on diagnostic paths.
llvm-svn: 180741
This seems to be causing quite a slowdown on our internal analyzer bot,
and I'm not sure why. Needs further investigation.
This reverts r180638 / 9e161ea981f22ae017b6af09d660bfc3ddf16a09.
llvm-svn: 180714
In an Objective-C for-in loop "for (id element in collection) {}", the loop
will run 0 times if the collection is nil. This is because the for-in loop
is implemented using a protocol method that returns 0 when there are no
elements to iterate, and messages to nil will result in a 0 return value.
At some point we may want to actually model this message send, but for now
we may as well get the nil case correct, and avoid the false positives that
would come with this case.
<rdar://problem/13744632>
llvm-svn: 180639
Casts to bool (and _Bool) are equivalent to checks against zero,
not truncations to 1 bit or 8 bits.
This improved reasoning does cause a change in the behavior of the alpha
BoolAssignment checker. Previously, this checker complained about statements
like "bool x = y" if 'y' was known not to be 0 or 1. Now it does not, since
that conversion is well-defined. It's hard to say what the "best" behavior
here is: this conversion is safe, but might be better written as an explicit
comparison against zero.
More usefully, besides improving our model of booleans, this fixes spurious
warnings when returning the address of a local variable cast to bool.
<rdar://problem/13296133>
llvm-svn: 180638
We get a CallEnter with a null expression, when processing a destructor. All other users of
CallEnter::getCallExpr work fine with null as return value.
(Addresses PR15832, Thanks to Jordan for reducing the test case!)
llvm-svn: 180234
- If only partial invalidators exist and there are no full invalidators in @implementation, report every ivar that has
not been invalidated. (Previously, we reported the first Ivar in the list, which could actually have been invalidated
by a partial invalidator. The code assumed you cannot have only partial invalidators.)
- Do not report missing invalidation method declaration if a partial invalidation method declaration exists.
llvm-svn: 180170
The uniqueing location is the location which is part of the hash used to determine if two reports are
the same. This is used by the CmpRuns.py script to compare two analyzer runs and determine which
warnings are new.
llvm-svn: 180166
The 2 functions were computing the same location using different logic (each one had edge case bugs that the other
one did not). Refactor them to rely on the same logic.
The location of the warning reported in text/command line output format will now match that of the plist file.
There is one change in the plist output as well. When reporting an error on a BinaryOperator, we use the location of the
operator instead of the beginning of the BinaryOperator expression. This matches our output on command line and
looks better in most cases.
llvm-svn: 180165
The analyzer represents all pointer-to-pointer bitcasts the same way, but
this can be problematic if an implicit base cast gets layered on top of a
manual base cast (performed with reinterpret_cast instead of static_cast).
Fix this (and avoid a valid assertion) by looking through cast regions.
Using reinterpret_cast this way is only valid if the base class is at the
same offset as the derived class; this is checked by -Wreinterpret-base-class.
In the interest of performance, the analyzer doesn't repeat this check
anywhere; it will just silently do the wrong thing (use the wrong offsets
for fields of the base class) if the user code is wrong.
PR15394
llvm-svn: 180052
Add a CXXDefaultInitExpr, analogous to CXXDefaultArgExpr, and use it both in
CXXCtorInitializers and in InitListExprs to represent a default initializer.
There's an additional complication here: because the default initializer can
refer to the initialized object via its 'this' pointer, we need to make sure
that 'this' points to the right thing within the evaluation.
llvm-svn: 179958
Introduce a new helper function, which computes the first symbolic region in
the base region chain. The corresponding symbol has been used for assuming that
a pointer is null. Now, it will also be used for checking if it is null.
This ensures that we are tracking a null pointer correctly in the BugReporter.
llvm-svn: 179916
The analyzer uses LazyCompoundVals to represent rvalues of aggregate types,
most importantly structs and arrays. This allows us to efficiently copy
around an entire struct, rather than doing a memberwise load every time a
struct rvalue is encountered. This can also keep memory usage down by
allowing several structs to "share" the same snapshotted bindings.
However, /lookup/ through LazyCompoundVals can be expensive, especially
since they can end up chaining back to the original value. While we try
to reuse LazyCompoundVals whenever it's safe, and cache information about
this transitivity, the fact is it's sometimes just not a good idea to
perpetuate LazyCompoundVals -- the tradeoffs just aren't worth it.
This commit changes RegionStore so that binding a LazyCompoundVal to struct
will do a memberwise copy if the struct is simple enough. Today's definition
of "simple enough" is "up to N scalar members" (see below), but that could
easily be changed in the future. This is enough to bring the test case in
PR15697 back down to a manageable analysis time (within 20% of its original
time, in an unfair test where the new analyzer is not compiled with LTO).
The actual value of "N" is controlled by a new -analyzer-config option,
'region-store-small-struct-limit'. It defaults to "2", meaning structs with
zero, one, or two scalar members will be considered "simple enough" for
this code path.
It's worth noting that a more straightforward implementation would do this
on load, not on bind, and make use of the structure we already have for this:
CompoundVal. A long time ago, this was actually how RegionStore modeled
aggregate-to-aggregate copies, but today it's only used for compound literals.
Unfortunately, it seems that we've special-cased LazyCompoundVal in certain
places (such as liveness checks) but failed to similarly special-case
CompoundVal in all of them. Until we're confident that CompoundVal is
handled properly everywhere, this solution is safer, since the entire
optimization is just an implementation detail of RegionStore.
<rdar://problem/13599304>
llvm-svn: 179767
A C++ overloaded operator may be implemented as an instance method, and
that instance method may be called on an rvalue object, which has no
associated region. The analyzer handles this by creating a temporary region
just for the evaluation of this call; however, it is possible that /by
creating the region/, the analyzer ends up in a previously-explored state.
In this case we don't need to continue along this path.
This doesn't actually show any behavioral change now, but it starts being
used with the next commit and prevents an assertion failure there.
llvm-svn: 179766
In the committed example, we now see a note that tells us when the pointer
was assumed to be null.
This is the only case in which getDerefExpr returned null (failed to get
the dereferenced expr) throughout our regression tests. (There were multiple
occurrences of this one.)
llvm-svn: 179736
We always register the visitor on a node in which the value we are tracking is live and constrained. However,
the visitation can restart at a node, later on the path, in which the value is under constrained because
it is no longer live. Previously, we just silently stopped tracking in that case.
llvm-svn: 179731
This was slightly tricky because BlockDecls don't currently store an
inferred return type. However, we can rely on the fact that blocks with
inferred return types will have return statements that match the inferred
type.
<rdar://problem/13665798>
llvm-svn: 179699
This is an opt-in tweak for leak diagnostics to reference the allocation
site if the diagnostic consumer only wants a pithy amount of information,
and not the entire path.
This is a strawman enhancement that I expect to see some experimentation
with over the next week, and can go away if we don't want it.
Currently it is only used by RetainCountChecker, but could be used
by MallocChecker if and when we decide this should stay in.
llvm-svn: 179634
during checker registration. There are no immediate clients of this,
but this provides a way for checkers to query the options table
at startup instead.
llvm-svn: 179626
When computing the value of ?: expression, we rely on the last expression in
the previous basic block to be the resulting value of the expression. This is
not the case for binary "?:" operator (GNU extension) in C++. As the last
basic block has the expression for the condition subexpression, which is an
R-value, whereas the true subexpression is the L-value.
Note the operator evaluation just happens to work in C since the true
subexpression is an R-value (like the condition subexpression). CFG is the
same in C and C++ case, but the AST nodes are different, which the LValue to
Rvalue conversion happening after the BinaryConditionalOperator evaluation.
Changed the logic to only use the last expression from the predecessor only
if it matches either true or false subexpression. Note, the logic needed
fortification anyway: L and R were passed but not even used by the function.
Also, change the conjureSymbolVal to correctly compute the type, when the
expression is an LG-value.
llvm-svn: 179574
While we don't do anything intelligent with pointers-to-members today,
it's perfectly legal to need a temporary of pointer-to-member type to, say,
pass by const reference. Tweak an assertion to allow this.
PR15742 and PR15747
llvm-svn: 179563
Pavel Labath