an initial baseline for enqueued blocks, but use a simple DFS stack
for propagating changes quickly up back edges.
This provides a 3.5% reduction in -fsyntax-only time on sqlite3.c.
llvm-svn: 168241
This code assigned the last created CFGBlock* to the variable 'Block',
which is a scratch variable which is null'ed out after a block is
completed. By assigning the last created block to 'Block', we start
editing a completed block, inserting CFGStmts that should be in
another block. This was the case with 'try'. The test case that
showed this had a while loop inside a 'try', and the logic before
the while loop was being included as part of the "condition block"
for the loop. This showed up as a bogus dead store, but could
have lots of implications.
Turns out this bug was replicated a few times within CFG.cpp, so
I went and fixed up those as well.
llvm-svn: 167788
Clang will now honor the FP_CONTRACT pragma and emit LLVM
fmuladd intrinsics for expressions of the form A * B + C (when they occur in a
single statement).
llvm-svn: 164989
- The size of the packed vector is often small, save mallocs using SmallBitVector.
- Copying SmallBitVectors is also cheap, remove a level of indirection.
llvm-svn: 164827
their implementations are unavailable. Start by simulating dispatch_sync().
This change is largely a bunch of plumbing around something very simple. We
use AnalysisDeclContext to conjure up a fake function body (using the
current ASTContext) when one does not exist. This is controlled
under the analyzer-config option "faux-bodies", which is off by default.
The plumbing in this patch is largely to pass the necessary machinery
around. CallEvent needs the AnalysisDeclContextManager to get
the function definition, as one may get conjured up lazily.
BugReporter and PathDiagnosticLocation needed to be relaxed to handle
invalid locations, as the conjured body has no real source locations.
We do some primitive recovery in diagnostic generation to generate
some reasonable locations (for arrows and events), but it can be
improved.
llvm-svn: 164339
Objective-C related to NSException.
Fixes <rdar://problem/12287498>
I debated whether or not this logic should be sunk into the CFG
itself. It's not clear if we should, as different analyses may
wish to have different policies. We can re-evaluate this in the
future.
llvm-svn: 163760
in NSException to a helper object in libAnalysis that can also
be used by Sema. Not sure if the predicate name 'isImplicitNoReturn'
is the best one, but we can massage that later.
No functionality change.
llvm-svn: 163759
for halting the propagation of uninitialized value tracking along
a path. Unlike __attribute__((noreturn)), this attribute (which
is used by clients of the static analyzer) can be used to annotate
functions that essentially never return, but in rare cares may be
allowed to return for (special) debugging purposes. This attribute
has been shown in reducing false positives in the static analyzer
by pruning false postives, and is equally applicable here.
Handling this attribute in the CFG itself is another option, but
this is not something all clients (e.g., possibly -Wunreachable-code)
would want to see.
Addresses <rdar://problem/12281583>.
llvm-svn: 163681
analysis that may give false positives because it is confused by aliasing, and
a less precise analysis that has fewer false positives, but may have false
negatives. The more precise warnings are enabled by -Wthread-safety-precise.
An additional note clarify the warnings in the precise case.
llvm-svn: 163537
While destructors will continue to not be inlined (unless the analyzer
config option 'c++-inlining' is set to 'destructors'), leaving them out
of the CFG is an incomplete model of the behavior of an object, and
can cause false positive warnings (like PR13751, now working).
Destructors for temporaries are still not on by default, since
(a) we haven't actually checked this code to be sure it's fully correct
(in particular, we probably need to be very careful with regard to
lifetime-extension when a temporary is bound to a reference,
C++11 [class.temporary]p5), and
(b) ExprEngine doesn't actually do anything when it sees a temporary
destructor in the CFG -- not even invalidate the object region.
To enable temporary destructors, set the 'cfg-temporary-dtors' analyzer
config option to '1'. The old -cfg-add-implicit-dtors cc1 option, which
controlled all implicit destructors, has been removed.
llvm-svn: 163264
A CXXDefaultArgExpr wraps an Expr owned by a ParmVarDecl belonging to the
called function. In general, ExprEngine and Environment ought to treat this
like a ParenExpr or other transparent wrapper expression, with the inside
expression evaluated first.
However, if we call the same function twice, we'd produce a CFG that contains
the same wrapped expression twice, and we're not set up to handle that. I've
added a FIXME to the CFG builder to come back to that, but meanwhile we can
at least handle expressions that don't need to be explicitly evaluated:
literals. This probably handles many common uses of default parameters:
true/false, null, etc.
Part of PR13385 / <rdar://problem/12156507>
llvm-svn: 162453
Also rename 'getCurrentBlockCounter()' to 'blockCount()'.
This ripples a bunch of code simplifications; mostly aesthetic,
but makes the code a bit tighter.
llvm-svn: 162349
This is useful for example for %n in printf, which expects
a pointer to int with the same logic for checking as %d
would have in scanf.
llvm-svn: 161407
Clang's -Wformat fix-its currently suggest using "%zu" for values of
type size_t (in C99 or C++11 mode). However, for a type such as
std::vector<T>::size_type, it does not notice that type is actually
typedeffed to size_t, and instead suggests a format for the underlying
type, such as "%lu" or "%u".
This commit makes the format string fix mechanism walk the typedef chain
so that it notices if the type is size_t, even if that isn't "at the
top".
llvm-svn: 160886
a defaulted special member function until the exception specification is needed
(using the same criteria used for the delayed instantiation of exception
specifications for function temploids).
EST_Delayed is now EST_Unevaluated (using 1330's terminology), and, like
EST_Uninstantiated, carries a pointer to the FunctionDecl which will be used to
resolve the exception specification.
This is enabled for all C++ modes: it's a little faster in the case where the
exception specification isn't used, allows our C++11-in-C++98 extensions to
work, and is still correct for C++98, since in that mode the computation of the
exception specification can't fail.
The diagnostics here aren't great (in particular, we should include implicit
evaluation of exception specifications for defaulted special members in the
template instantiation backtraces), but they're not much worse than before.
Our approach to the problem of cycles between in-class initializers and the
exception specification for a defaulted default constructor is modified a
little by this change -- we now reject any odr-use of a defaulted default
constructor if that constructor uses an in-class initializer and the use is in
an in-class initialzer which is declared lexically earlier. This is a closer
approximation to the current draft solution in core issue 1351, but isn't an
exact match (but the current draft wording isn't reasonable, so that's to be
expected).
llvm-svn: 160847
Our BugReporter knows how to deal with implicit statements: it looks in
the ParentMap until it finds a parent with a valid location. However, since
initializers are not in the body of a constructor, their sub-expressions are
not in the ParentMap. That was easy enough to fix in AnalysisDeclContext.
...and then even once THAT was fixed, there's still an extra funny case
of Objective-C object pointer fields under ARC, which are initialized with
a top-level ImplicitValueInitExpr. To catch these cases,
PathDiagnosticLocation will now fall back to the start of the current
function if it can't find any other valid SourceLocations. This isn't great,
but it's miles better than a crash.
(All of this is only relevant when constructors and destructors are being
inlined, i.e. under -cfg-add-initializers and -cfg-add-implicit-dtors.)
llvm-svn: 160810
short-circuiting when building the CFG. Also be sure to skip parens before
checking for the && / || special cases. Finally, fix some crashes in CFG
printing in the presence of calls to destructors for array of array of class
type.
llvm-svn: 160691
The CFG creates dummy DeclStmts with one Decl per statement, and it has
to do so from last to first in order to build the graph correctly.
llvm-svn: 160560
* Treat compound assignment as a use, at Jordy's request.
* Always add compound assignments into the CFG, so we can correctly diagnose the use in 'return x += 1;'
llvm-svn: 160334
use out of TransferFunctions, and compute it in advance rather than on-the-fly.
This allows us to handle compound assignments with DeclRefExprs on the RHS
correctly, and also makes it trivial to treat const& function parameters as not
initializing the argument. The patch also makes both of those changes.
llvm-svn: 160330
instead push the terminator for the branch down into the basic blocks of the subexpressions of '&&' and '||'
respectively. This eliminates some artifical control-flow from the CFG and results in a more
compact CFG.
Note that this patch only alters the branches 'while', 'if' and 'for'. This was complex enough for
one patch. The remaining branches (e.g., do...while) can be handled in a separate patch, but they
weren't immediately tackled because they were less important.
It is possible that this patch introduces some subtle bugs, particularly w.r.t. to destructor placement.
I've tried to audit these changes, but it is also known that the destructor logic needs some refinement
in the area of '||' and '&&' regardless (i.e., their are known bugs).
llvm-svn: 160218
uninitialized variable use, walk back over branches where we've reached all the
non-null successors, not just cases where we've reached all successors.
llvm-svn: 160206
very simple semantic analysis that just builds the AST; minor changes for lexer
to pick up source locations I didn't think about before.
Comments AST is modelled along the ideas of HTML AST: block and inline content.
* Block content is a paragraph or a command that has a paragraph as an argument
or verbatim command.
* Inline content is placed within some block. Inline content includes plain
text, inline commands and HTML as tag soup.
llvm-svn: 159790
duplicates attributes on the declaration. Also eliminates a false negative in
ReleasableMutexLock. Fixing this bug required some refactoring.
llvm-svn: 159780
This required moving the ctors for IntegerLiteral and FloatingLiteral out of
line which shouldn't change anything as they are usually called through Create
methods that are already out of line.
ASTContext::Deallocate has been a nop for a long time, drop it from ASTVector
and make it independent from ASTContext.h
Pass the StorageAllocator directly to AccessedEntity so it doesn't need to
have a definition of ASTContext around.
llvm-svn: 159718
we are encountering some scalability issues with memory usage. The
appropriate long term fix is to make the analysis more scalable, but
this will at least prevent the analyzer swapping when
analyzing very large functions.
llvm-svn: 159578
properly if there is a join point in the control flow graph that involves
a trylock. Also changes the source locations of some warnings to be
more consistent.
llvm-svn: 159008
express library-level dependencies within Clang.
This is no more verbose really, and plays nicer with the rest of the
CMake facilities. It should also have no change in functionality.
llvm-svn: 158888
In addition, I've made the pointer and reference typedef 'void' rather than T*
just so they can't get misused. I would've omitted them entirely but
std::distance likes them to be there even if it doesn't use them.
This rolls back r155808 and r155869.
Review by Doug Gregor incorporating feedback from Chandler Carruth.
llvm-svn: 158104
-Wsometimes-uninitialized. This detects cases where an explicitly-written branch
inevitably leads to an uninitialized variable use (so either the branch is dead
code or there is an uninitialized use bug).
This chunk of warnings tentatively lives within -Wuninitialized, in order to
give it more visibility to existing Clang users.
llvm-svn: 157458
For "%hhx", printf expects an unsigned char. This makes Clang
accept a 'char' argument for that also when using -funsigned-char.
This fixes PR12761.
llvm-svn: 156388
Teach ASTContext about WIntType, and have it taken from TargetInfo like WCharType. Should fix test/Sema/format-strings.c for ARM, with the exception of one subtest which will fail if wint_t and wchar_t are the same size and wint_t is signed, wchar_t is unsigned.
There'll be a followup commit to fix that.
Reviewed by Chandler and Hans at http://llvm.org/reviews/r/8
llvm-svn: 156165
cases in switch statements. Also add a [[clang::fallthrough]] attribute, which
can be used to suppress the warning in the case of intentional fallthrough.
Patch by Alexander Kornienko!
The handling of C++11 attribute namespaces in this patch is temporary, and will
be replaced with a cleaner mechanism in a subsequent patch.
llvm-svn: 156086
filter_decl_iterator had a weird mismatch where both op* and op-> returned T*
making it difficult to generalize this filtering behavior into a reusable
library of any kind.
This change errs on the side of value, making op-> return T* and op* return
T&.
(reviewed by Richard Smith)
llvm-svn: 155808
g++4.7, which reuses stack space allocated for temporaries. CFGElement::getAs
returns a suitably-cast version of 'this'. Patch by Markus Trippelsdorf!
No test: this code has the same observable behavior as the old code when built
with most compilers, and the tests were already failing when built with a
compiler for which this produced a broken binary.
llvm-svn: 155803
This is needed to ensure that we always report issues in the correct
function. For example, leaks are identified when we call remove dead
bindings. In order to make sure we report a callee's leak in the callee,
we have to run the operation in the callee's context.
This change required quite a bit of infrastructure work since:
- We used to only run remove dead bindings before a given statement;
here we need to run it after the last statement in the function. For
this, we added additional Program Point and special mode in the
SymbolReaper to remove all symbols in context lower than the current
one.
- The call exit operation turned into a sequence of nodes, which are
now guarded by CallExitBegin and CallExitEnd nodes for clarity and
convenience.
(Sorry for the long diff.)
llvm-svn: 155244
We have a new flavor of exception specification, EST_Uninstantiated. A function
type with this exception specification carries a pointer to a FunctionDecl, and
the exception specification for that FunctionDecl is instantiated (if needed)
and used in the place of the function type's exception specification.
When a function template declaration with a non-trivial exception specification
is instantiated, the specialization's exception specification is set to this
new 'uninstantiated' kind rather than being instantiated immediately.
Expr::CanThrow has migrated onto Sema, so it can instantiate exception specs
on-demand. Also, any odr-use of a function triggers the instantiation of its
exception specification (the exception specification could be needed by IRGen).
In passing, fix two places where a DeclRefExpr was created but the corresponding
function was not actually marked odr-used. We used to get away with this, but
don't any more.
Also fix a bug where instantiating an exception specification which refers to
function parameters resulted in a crash. We still have the same bug in default
arguments, which I'll be looking into next.
This, plus a tiny patch to fix libstdc++'s common_type, is enough for clang to
parse (and, in very limited testing, support) all of libstdc++4.7's standard
headers.
llvm-svn: 154886
attached. Since we do not support any attributes which appertain to a statement
(yet), testing of this is necessarily quite minimal.
Patch by Alexander Kornienko!
llvm-svn: 154723
We should not deserialize unused declarations from the PCH file. Achieve
this by storing the top level declarations during parsing
(HandleTopLevelDecl ASTConsumer callback) and analyzing/building a call
graph only for those.
Tested the patch on a sample ObjC file that uses PCH. With the patch,
the analyzes is 17.5% faster and clang consumes 40% less memory.
Got about 10% overall build/analyzes time decrease on a large Objective
C project.
A bit of CallGraph refactoring/cleanup as well..
llvm-svn: 154625
during construction of branches for chained logical operators.
This makes -fsyntax-only for test/Sema/many-logical-ops.c about 32x times faster.
With measuring SemaExpr.cpp I see differences below the noise level.
llvm-svn: 153297
Specifically, we use the last store of the leaked symbol in the leak diagnostic.
(No support for struct fields since the malloc checker doesn't track those
yet.)
+ Infrastructure to track the regions used in store evaluations.
This approach is more precise than iterating the store to
obtain the region bound to the symbol, which is used in RetainCount
checker. The region corresponds to what is uttered in the code in the
last store and we do not rely on the store implementation to support
this functionality.
llvm-svn: 153212
The one difference between ObjCMethodDecl::getMethodFamily and Selector::getMethodFamily is that the former will do some additional sanity checking, and since CoreFoundation types don't look like Objective-C objects, an otherwise interesting method will get a method family of OMF_None. Future clients that use method families should consider how they want to handle CF types.
llvm-svn: 153000
track whether the referenced declaration comes from an enclosing
local context. I'm amenable to suggestions about the exact meaning
of this bit.
llvm-svn: 152491
as aborted, but didn't treat such cases as sinks in the ExplodedGraph.
Along the way, add basic support for CXXCatchStmt, expanding the set of code we actually analyze (hopefully correctly).
Fixes: <rdar://problem/10892489>
llvm-svn: 152468
This renames the -Wformat-non-standard flag to -Wformat-non-iso,
rewords the current warnings a bit (pointing out that a format string
is not supported by ISO C rather than being "non standard"),
and adds a warning about positional arguments.
llvm-svn: 152403
The final graph contains a single root node, which is a parent of all externally available functions(and 'main'). As well as a list of Parentless/Unreachable functions, which are either truly unreachable or are unreachable due to our analyses imprecision.
The analyzer checkers debug.DumpCallGraph or debug.ViewGraph can be used to look at the produced graph.
Currently, the graph is not very precise, for example, it entirely skips edges resulted from ObjC method calls.
llvm-svn: 152272
analysis to make the AST representation testable. They are represented by a
new UserDefinedLiteral AST node, which is a sugared CallExpr. All semantic
properties, including full CodeGen support, are achieved for free by this
representation.
UserDefinedLiterals can never be dependent, so no custom instantiation
behavior is required. They are mangled as if they were direct calls to the
underlying literal operator. This matches g++'s apparent behavior (but not its
actual mangling, which is broken for literal-operator-ids).
User-defined *string* literals are now fully-operational, but the semantic
analysis is quite hacky and needs more work. No other forms of user-defined
literal are created yet, but the AST support for them is present.
This patch committed after midnight because we had already hit the quota for
new kinds of literal yesterday.
llvm-svn: 152211
This adds the -Wformat-non-standard flag (off by default,
enabled by -pedantic), which warns about non-standard
things in format strings (such as the 'q' length modifier,
the 'S' conversion specifier, etc.)
llvm-svn: 151154
This is in preparation for being able to warn about 'q' and other
non-standard format string features.
It also allows us to print its name correctly.
llvm-svn: 150697
This commit makes PrintfSpecifier::fixType() and ScanfSpecifier::fixType()
only fix a conversion specification enough that Clang wouldn't warn about it,
as opposed to always changing it to use the "canonical" conversion specifier.
(PR11975)
This preserves the user's choice of conversion specifier in cases like:
printf("%a", (long double)1);
where we previously suggested "%Lf", we now suggest "%La"
printf("%x", (long)1);
where we previously suggested "%ld", we now suggest "%lx".
llvm-svn: 150578
* When we detect that a CFG block has inconsistent lock sets, point the
diagnostic at the location where we found the inconsistency, and point a note
at somewhere the inconsistently-locked mutex was locked.
* Fix the wording of the normal (non-loop, non-end-of-function) case of this
diagnostic to not suggest that the mutex is going out of scope.
* Fix the diagnostic emission code to keep a warning and its note together when
sorting the diagnostics into source location order.
llvm-svn: 149669
This fixes the case where Clang would output:
error: format specifies type 'wchar_t *' (aka 'wchar_t *')
ArgTypeResult::getRepresentativeTypeName needs to take into account
that wchar_t can be a built-in type (as opposed to in C, where it is a
typedef).
llvm-svn: 149387
block-level expr. Currently CXXConstructExpr is always added as a block-level
expr. This caused two problems for the analyzer (and potentially for the
CFG-based codegen).
1. We have no way to know whether a ctor call is base or complete.
2. We have no way to know the destination object being contructed.
llvm-svn: 147306
declarations and definitions) as ObjCInterfaceDecls within the same
redeclaration chain. This new representation matches what we do for
C/C++ variables/functions/classes/templates/etc., and makes it
possible to answer the query "where are all of the declarations of
this class?"
llvm-svn: 146679
in addition to underlying type.
For example, the warning for printf("%zu", 42.0);
changes from "conversion specifies type 'unsigned long'" to "conversion
specifies type 'size_t' (aka 'unsigned long')"
(This is a second attempt after r145697, which got reverted.)
llvm-svn: 146032
For example, the warning for printf("%zu", 42.0);
changes from "conversion specifies type 'unsigned long'" to "conversion
specifies type 'size_t' (aka 'unsigned long')"
llvm-svn: 145697
lifetimes have been extended via reference binding. The type of the
reference and the type of the temporary are not necessarily the same,
which could cause a crash. Fixes <rdar://problem/10398199>.
llvm-svn: 144646
property references to use a new PseudoObjectExpr
expression which pairs a syntactic form of the expression
with a set of semantic expressions implementing it.
This should significantly reduce the complexity required
elsewhere in the compiler to deal with these kinds of
expressions (e.g. IR generation's special l-value kind,
the static analyzer's Message abstraction), at the lower
cost of specifically dealing with the odd AST structure
of these expressions. It should also greatly simplify
efforts to implement similar language features in the
future, most notably Managed C++'s properties and indexed
properties.
Most of the effort here is in dealing with the various
clients of the AST. I've gone ahead and simplified the
ObjC rewriter's use of properties; other clients, like
IR-gen and the static analyzer, have all the old
complexity *and* all the new complexity, at least
temporarily. Many thanks to Ted for writing and advising
on the necessary changes to the static analyzer.
I've xfailed a small diagnostics regression in the static
analyzer at Ted's request.
llvm-svn: 143867
implicitly perform an lvalue-to-rvalue conversion if used on an lvalue
expression. Also improve the documentation of Expr::Evaluate* to indicate which
of them will accept expressions with side-effects.
llvm-svn: 143263
The code had it backwards, thinking size_t was signed, and using that for "%zd".
Also let the analysis get the types for (u)intmax_t while we are at it.
llvm-svn: 143099
For PR11152. Make PrintSpecifier::fixType() suggest "%zu" for size_t, etc.
rather than looking at the underlying type and suggesting "%llu" or other
platform-specific length modifiers. Applies to C99 and C++11.
llvm-svn: 142342
to take a FunctionDecl* instead of an llvm::StringRef. Eventually
we might push more logic in there, like using slightly different
conventions for C++ methods.
Also, fix a bug where 'copy' and 'create' were being caught in
non-camel-cased strings. We want copyFoo and CopyFoo and XCopy
but not Xcopy or xcopy.
llvm-svn: 140911
- Speed of "merge()", which merged data flow facts. This was doing a set canonicalization on every insertion, which was super slow.
To fix this, we use ImmutableSetRef.
- Visit CFGBlocks in reverse postorder. This is a huge speedup, as on some test cases the algorithm would take many iterations
to converge.
This contains a bunch of copy-paste from UninitializedValues.cpp and ThreadSafety.cpp. The idea
was to get something working first, and then refactor the common logic for all three files into
a separate analysis/library entry point.
llvm-svn: 139968
temporary objects and local variables. When detected, these split the
block, marking the new one as having only the exit block as a successor.
This prevents a large number of false positives in warnings sensitive to
no-return constructs such as -Wreturn-type, and fixes the remainder of
PR10063 along with several variations of this bug that had not been
reported. The test cases are extended across the board to cover these
patterns.
This also checks in a stress test for these types of CFGs. The stress
test declares some 32k variables, a mixture of no-return and normal
destructors. Previously, this resulted in roughly 2500 CFG blocks, but
didn't model any of the no-return destructors. With this patch, it
results in over 33k blocks, many of them now unreachable.
The nice thing about how the analyzer is set up? This causes *no*
regression in performance of building the CFG. It actually in some cases
makes it faster, as best I can benchmark. The analysis for -Wreturn-type
(and any other that cares about no-return code paths) is technically
slower now as it has to look at many more candidate blocks, but it
computes the correct answer. I have more test cases to follow, I think
they all work now. Also I have further work that should dramatically
simplify analyses in the presence of no-return.
llvm-svn: 139586
and case statements. Use this to make the logic in the CFG builder more
robust at finding the actual statements within a compound statement,
even when there are many layers of labels obscuring it.
Also extend the test cases for a large chunk of PR10063. Still more work
to do here though.
llvm-svn: 139437
incorrectly in the CFG, and also the static analyzer. This patch regresses the analyzer a bit, but
that needs to be followed up with a better solution.
Fixes <rdar://problem/10008112>.
llvm-svn: 138372
Having a notion of an actual ProgramPointTag will aid in introspection of the analyzer's behavior.
For example, the GraphViz output of the analyzer will pretty-print the tags in a useful manner.
llvm-svn: 137529
The motivation of this large change is to drastically simplify the logic in ExprEngine going forward.
Some fallout is that the output of some BugReporterVisitors is not as accurate as before; those will
need to be fixed over time. There is also some possible performance regression as RemoveDeadBindings
will be called frequently; this can also be improved over time.
llvm-svn: 136419
AnalysisBasedWarnings Sema layer and out of the Analysis library itself.
This returns the uninitialized values analysis to a more pure form,
allowing its original logic to correctly detect some categories of
definitely uninitialized values. Fixes PR10358 (again).
Thanks to Ted for reviewing and updating this patch after his rewrite of
several portions of this analysis.
llvm-svn: 135748
This is accomplished by forcing the needed expressions for -Wuninitialized to always be CFGElements in the CFG.
This allows us to remove a fair amount of the code for -Wuninitialized.
Some fallout:
- AnalysisBasedWarnings.cpp now specifically toggles the CFGBuilder to create a CFG that is suitable for -Wuninitialized. This
is a layering violation, since the logic for -Wuninitialized is in libAnalysis. This can be fixed with the proper refactoring.
- Some of the source locations for -Wunreachable-code warnings have shifted. While not ideal, this is okay because that analysis
already needs some serious reworking.
llvm-svn: 135480
patch, we actually move the state-machine for the value set backwards
one step. This can pretty easily lead to infinite loops where we
continually try to propagate a bit, succeed for one iteration, but then
back up because we find an uninitialized use.
A reduced test case from PR10379 is included.
llvm-svn: 135359
Previously, despite the names 'enqueue' and 'dequeue', it behaved as
a stack and visited blocks in a LIFO fashion. This interacts badly with
extremely broad CFGs *inside* of a loop (such as a large switch inside
a state machine) where every block updates a different variable.
When encountering such a CFG, the checker visited blocks in essentially
a "depth first" order due to the stack-like behavior of the work list.
Combined with each block updating a different variable, the saturation
logic of the checker caused it to re-traverse blocks [1,N-1] of the
broad CFG inside the loop after traversing block N. These re-traversals
were to propagate the variable values derived from block N. Assuming
approximately the same number of variables as inner blocks exist, the
end result is O(N^2) updates. By making this a queue, we also make the
traversal essentially "breadth-first" across each of the N inner blocks
of the loop. Then all of this state is propagated around to all N inner
blocks of the loop. The result is O(N) updates.
The truth is in the numbers:
Before, gcc.c: 96409 block visits (max: 61546, avg: 591)
After, gcc.c: 69958 block visits (max: 33090, avg: 429)
Before, PR10183: 2540494 block vists (max: 2536495, avg: 37360)
After, PR10183: 137803 block visits (max: 134406, avg: 2026)
The nearly 20x reduction in work for PR10183 corresponds to a roughly
100x speedup in compile time.
I've tested it on all the code I can get my hands on, and I've seen no
slowdowns due to this change. Where I've collected stats, the ammount of
work done is on average less. I'll also commit shortly some synthetic
test cases useful in analyzing the performance of CFG-based warnings.
Submitting this based on Doug's feedback that post-commit review should
be good. Ted, please review! Hopefully this helps compile times until
then.
llvm-svn: 134697
Special detail is added for uninitialized variable analysis as this has
serious performance problems than need to be tracked.
Computing some of this data is expensive, for example walking the CFG to
determine its size. To avoid doing that unless the stats data is going
to be used, we thread a bit into the Sema object to track whether
detailed stats should be collected or not. This bit is used to avoid
computations whereever the computations are likely to be more expensive
than checking the state of the flag. Thus, counters are in some cases
unconditionally updated, but the more expensive (and less frequent)
aggregation steps are skipped.
With this patch, we're able to see that for 'gcc.c':
*** Analysis Based Warnings Stats:
232 functions analyzed (0 w/o CFGs).
7151 CFG blocks built.
30 average CFG blocks per function.
1167 max CFG blocks per function.
163 functions analyzed for uninitialiazed variables
640 variables analyzed.
3 average variables per function.
94 max variables per function.
96409 block visits.
591 average block visits per function.
61546 max block visits per function.
And for the reduced testcase in PR10183:
*** Analysis Based Warnings Stats:
98 functions analyzed (0 w/o CFGs).
8526 CFG blocks built.
87 average CFG blocks per function.
7277 max CFG blocks per function.
68 functions analyzed for uninitialiazed variables
1359 variables analyzed.
19 average variables per function.
1196 max variables per function.
2540494 block visits.
37360 average block visits per function.
2536495 max block visits per function.
That last number is the somewhat scary one that indicates the problem in
PR10183.
llvm-svn: 134494
specifiers. Fixes <rdar://problem/9607158>." because it causes false positives
on some code that uses CF toll free bridging.
- I'll let Doug or Ted figure out the right fix here, possibly just to accept
any pointer type.
llvm-svn: 134041
MaterializeTemporaryExpr captures a reference binding to a temporary
value, making explicit that the temporary value (a prvalue) needs to
be materialized into memory so that its address can be used. The
intended AST invariant here is that a reference will always bind to a
glvalue, and MaterializeTemporaryExpr will be used to convert prvalues
into glvalues for that binding to happen. For example, given
const int& r = 1.0;
The initializer of "r" will be a MaterializeTemporaryExpr whose
subexpression is an implicit conversion from the double literal "1.0"
to an integer value.
IR generation benefits most from this new node, since it was
previously guessing (badly) when to materialize temporaries for the
purposes of reference binding. There are likely more refactoring and
cleanups we could perform there, but the introduction of
MaterializeTemporaryExpr fixes PR9565, a case where IR generation
would effectively bind a const reference directly to a bitfield in a
struct. Addresses <rdar://problem/9552231>.
llvm-svn: 133521
Language-design credit goes to a lot of people, but I particularly want
to single out Blaine Garst and Patrick Beard for their contributions.
Compiler implementation credit goes to Argyrios, Doug, Fariborz, and myself,
in no particular order.
llvm-svn: 133103
Ted Kremenek