Richard Smith [Fri, 23 Mar 2012 23:09:08 +0000 (23:09 +0000)]
When defining a forward-declared enum, don't try to attach the definition to
a previous declaration if the redeclaration is invalid. That way lies madness.
Fixes a crash-on-invalid reported by Abramo.
Ted Kremenek [Fri, 23 Mar 2012 06:26:56 +0000 (06:26 +0000)]
Avoid applying retain/release effects twice in RetainCountChecker when a function call was inlined (i.e., we do not need to apply summaries in such cases).
Richard Smith [Fri, 23 Mar 2012 03:33:32 +0000 (03:33 +0000)]
Support for definitions of member enumerations of class templates outside the
class template's definition, and for explicit specializations of such enum
members.
[libclang] For a CXCursor_ObjCInstanceMethodDecl/CXCursor_ObjCClassMethodDecl cursor,
return from clang_getCursorLocation the start location of the method name.
Bill Wendling [Thu, 22 Mar 2012 23:25:07 +0000 (23:25 +0000)]
When an MMX output variable is tied to the input variable, we have to implicitly
cast the value to x86_mmx. This gives the ASM string the correct call signature.
<rdar://problem/10919182>
Eli Friedman [Thu, 22 Mar 2012 22:36:39 +0000 (22:36 +0000)]
Make sure we correctly set the alignment for vector loads and stores associated with vector element lvalues. Patch by Kevin Schoedel (with some minor modifications by me).
[PCH] When we are replacing a decl in a chained PCH that is also a DeclContext,
make sure to fully load its external lexical and visible declarations before
re-writing it.
Chandler Carruth [Thu, 22 Mar 2012 06:22:13 +0000 (06:22 +0000)]
Relax the FileCheck assertions in this test a touch. Previously the test
relied on an artifact of how the inliner and subsequent passes in
clang's -O3 mode happen to treat basic blocks and the labels for the
basic blocks. In my work on the inliner, and changed this fundamental
assumption, and the label that was being checked on the entry basic
block will no longer appear in opt builds. There was no reason to expect
the label to always be present anyways, much to my regret.
I've changed the test to just ensure that we return an immediate
constant. If there are intervening instructions, that's bad, but not
really that relevant to the test.
I'd love it if others have a better way of checking that a function body
contains only a 'ret' instruction that isn't dependent on whether or not
the entry block receives a label...
Richard Smith [Thu, 22 Mar 2012 03:35:28 +0000 (03:35 +0000)]
During the instantiation of a class template specialization, that
specialization is known to be incomplete. If we're asked to try to
complete it, don't attempt to instantiate it again -- that can lead
to stack overflow, and to rejects-valids if the class being incomplete
is not an error.
Simplify DataRecursiveIntBinOpEvaluator::VisitBinOp() a bit and make sure we don't
evaluate RHS if LHS could not be evaluated and keepEvaluatingAfterFailure() is false.
Richard Trieu [Wed, 21 Mar 2012 23:30:30 +0000 (23:30 +0000)]
Change the binary operator data recursive evaluator to not stop at the first
non-constant value encountered. This allows the evaluator to deduce that
expressions like (x < 5 || true) is equal to true. Previously, it would visit
x and determined that the entire expression is could not evaluated to a
constant.
Anna Zaks [Wed, 21 Mar 2012 19:45:08 +0000 (19:45 +0000)]
[analyzer] Malloc: Utter the name of the leaked variable.
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.
Bob Wilson [Wed, 21 Mar 2012 17:19:12 +0000 (17:19 +0000)]
Duplicated code is bad. At least make it consistent.
The getARMTargetCPU and getLLVMArchSuffixForARM functions exist in both
Toolchain.cpp and Tools.cpp. This stuff needs a thorough overhaul. In the
meantime, this patch at least makes them consistent. One version had been
converted to use StringSwitch, and the other version had new Cortex M-series
processors added.
Allow void blocks to return witn a void expression in
c-mode to match behavior with void functions in c. Issue
warning with -pedantic. // rdar://11069896
Bob Wilson [Wed, 21 Mar 2012 16:31:37 +0000 (16:31 +0000)]
For Darwin, do not let -mcpu override the -arch option. <rdar://11059238>
On Darwin the architecture and the corresponding Mach-O slice is typically
specified with -arch. If not, it defaults to the current host architecture.
Do not use -mcpu to override the -arch value. This is only an issue when
people need to use specialized code for a non-default CPU (hopefully guarded
by run-time checks to detect the current processor). The -mcpu option is
still used for the -target-cpu option to clang, but this patch causes it to
not be used to set the architecture in the target triple.
John McCall [Wed, 21 Mar 2012 06:57:19 +0000 (06:57 +0000)]
For the annals of subtle but terrible bugs: fix a longstanding bug
in vtable layout where virtual methods inherited from virtual bases
could be assigned the same vcall adjustment slot if they shared
a name and parameter signature but differed in their
cv-qualification. The code was already trying to handle this
case, but unfortunately used the ordinary type qualifiers
(which are always empty here) instead of the method qualifiers.
This seems like something that the API should discourage, but
I don't know how to carry that principle out in this instance.
Eliminate this function's need for an ASTContext while we're at it.
This bug affects the ABI, and fixing it brings us into accord with
the Itanium ABI (and GCC's implementation of it), but, obviously,
technically breaks full compatibility with previous releases of Clang.
Just letting you know.
John McCall [Wed, 21 Mar 2012 00:45:33 +0000 (00:45 +0000)]
Test case for rdar://problem/11055105, a bug with the instantiation
of references to function template parameters in noexcept clauses when
the instantiation is forced from a point during parsing when a block
is in scope.
objective-c modern translator: Refactor intername ivar names to
one place and use it throughout. Also, change ivar name to avoid
name collisions. // rdar://11079366
Chad Rosier [Tue, 20 Mar 2012 16:40:00 +0000 (16:40 +0000)]
[avx] Define the _mm256_loadu2_xxx and _mm256_storeu2_xxx intrinsics.
From the Intel Optimization Reference Manual, Section 11.6.2. When data cannot
be aligned or alignment is not known, 16-byte memory accesses may provide better
performance.
rdar://11076953
Ted Kremenek [Mon, 19 Mar 2012 23:48:41 +0000 (23:48 +0000)]
Fix crash when querying the CFG reported when using the thread safety analysis
on code using multi-dimensional arrays. Fix by DeLesley Hutchins, and reported in
PR 12271.
Jordy Rose [Sun, 18 Mar 2012 07:43:35 +0000 (07:43 +0000)]
[analyzer] Mark a failed-realloc's result as an interesting symbol between the realloc call and the null check, so we get nicer path notes. Fixes a regression introduced by the diagnostic pruning added in r152361.
This is accomplished by calling markInteresting /during/ path diagnostic generation, and as such relies on deterministic ordering of BugReporterVisitors -- namely, that BugReporterVisitors are run in /reverse/ order from how they are added. (Right now that's a consequence of storing visitors in an ImmutableList, where new items are added to the front.) It's a little hacky, but it works for now.
I think this is the best we can do without storing the relation between the old and new symbols, and that would be a hit whether or not there ends up being an error.
David Blaikie [Sun, 18 Mar 2012 02:56:47 +0000 (02:56 +0000)]
Use character literals for vexing initialization fixit hints.
Instead of suggesting " = 0" for "char c();", suggest " = '\0'", and similarly
for other char types (wide, 16, and 32). Add tests for all these, and since
this means testing such hints under C++0x, add tests for some untested C++0x
hint cases in the existing code, including suggesting nullptr for pointer
initialization.
This sets up the initialization helper to provide better type fidelity that
will be especially helpful for non-assignment cases (such as fixit-correcting
NULL usage in function calls (eg: foo(char) + foo(NULL) => foo('\0') instead
of the less informative foo(0)))
Jordy Rose [Sun, 18 Mar 2012 01:26:10 +0000 (01:26 +0000)]
[analyzer] Use a FoldingSet to cache simple RetainSummary instances, rather than explicitly keeping DoNothing and StopTracking summaries and nothing else.
I tried to test the effects of this change on memory usage and run time, but what I saw on retain-release.m was indistinguishable from noise (debug and release builds). Even so, some caveman profiling showed 101 cache hits that we would have generated new summaries for before (i.e. not default or stop summaries), and the more code we analyze, the more memory we should save.
Maybe we should have a standard project for benchmarking the retain count checker's memory and time?
Jordy Rose [Sat, 17 Mar 2012 20:51:32 +0000 (20:51 +0000)]
Kill cocoa::deriveNamingConvention and cocoa::followsFundamentalRule. They are now just simple wrappers around method families, and method decls can cache method family lookups. Also, no one is using them right now.
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.
Jordy Rose [Sat, 17 Mar 2012 19:53:04 +0000 (19:53 +0000)]
[analyzer] Remove duplicate work on deriving method behavior. No functionality change.
The cocoa::deriveNamingConventions helper is just using method families anyway now, and the way RetainSummaryTemplate works means we're allocating an extra summary for every method with a relevant family.
Also, fix RetainSummaryTemplate to do the right thing w/r/t annotating an /existing/ summary. This was probably the real cause of <rdar://problem/10824732> and the fix in r152448.
modern objective-c rewriter: further improvement in
writing @synchronized statement; do not call locking
expression more than once and support early exits in
@synchronized's statement block (such as return).