Reid Kleckner [Tue, 10 Nov 2015 22:23:58 +0000 (22:23 +0000)]
[COFF] Don't try to emit weak aliases on COFF
This comes up when a derived class destructor is equivalent to a base
class destructor defined in the same TU, and we try to alias them.
A COFF weak alias cannot satisfy a normal undefined symbol reference
from another TU. The other TU must also mark the referenced symbol as
weak, and we can't rely on that.
Clang already has a special case here for dllexport, but we failed to
realize that the problem also applies to other non-discardable symbols
such as those from explicit template instantiations.
Evgeniy Stepanov [Tue, 10 Nov 2015 21:28:44 +0000 (21:28 +0000)]
Implement __attribute__((internal_linkage)).
The attrubite is applicable to functions and variables and changes
the linkage of the subject to internal.
This is the same functionality as C-style "static", but applicable to
class methods; and the same as anonymouns namespaces, but can apply
to individual methods of a class.
Following the proposal in
http://lists.llvm.org/pipermail/cfe-dev/2015-October/045580.html
Use the generic Sparc CPU handling for Linux, FreeBSD and OpenBSD, too.
This currently changes the default toward the more historic -Av8/-Av9,
but as discussed with James Y Knight, consistency is for now more
important than figuring out which default CPU each OS should be using.
Reorganise CPU handling for Sparc. When using -mcpu=v9 and co, __sparcv8
is not defined for 32bit mode, but __sparcv9 is. Pass down the correct
-target-cpu flags to the backend, so that instruction restrictions are
applied correctly. Pass down the correct -A flag when not using IAS.
The latter is limited to NetBSD targets in this commit.
Tim Northover [Mon, 9 Nov 2015 19:56:35 +0000 (19:56 +0000)]
Atomics: support __c11_* calls on _Atomic struct types.
When a struct's size is not a power of 2, the corresponding _Atomic() type is
promoted to the nearest. We already correctly handled normal C++ expressions of
this form, but direct calls to the __c11_atomic_whatever builtins ended up
performing dodgy operations on the smaller non-atomic types (e.g. memcpy too
much). Later optimisations removed this as undefined behaviour.
This patch converts EmitAtomicExpr to allocate its temporaries at the full
atomic width, sidestepping the issue.
Renato Golin [Mon, 9 Nov 2015 12:40:41 +0000 (12:40 +0000)]
[EABI] Add Clang support for -meabi flag
The -meabi flag to control LLVM EABI version.
Without '-meabi' or with '-meabi default' imply LLVM triple default.
With '-meabi gnu' sets EABI GNU.
With '-meabi 4' or '-meabi 5' set EABI version 4 and 5 respectively.
Devin Coughlin [Sat, 7 Nov 2015 18:27:35 +0000 (18:27 +0000)]
[analyzer] Update SATestBuild.py to enable a 'download and patch' model for projects.
Currently the SATestBuild.py and SATestAdd.py buildbot scripts expect project
sources to be checked into the project repository. This commit changes these
scripts to additionally support a model where project sources are downloaded
rather than checked into the repository. Sometimes projects may need to be
modified (for example, to support a newer versions of clang), so the updated scripts
also allow for an optional patch file that will be applied to the downloaded
project source before analysis.
To support this workflow, this commit changes the expected layout of
a project in the repository. The project-specific helper scripts will stay
in the root of each project directory, but the benchmark source itself (if
checked into the repo) should now be stored in a subdirectory named
'CachedSource':
If the 'CachedSource' source directory is not present, the download script will
be executed. This script should download the project source into 'CachedSource'.
Then, if 'changes_for_analyzer.patch' is present its changes will
be applied to a copy of 'CachedSource' before analysis.
Akira Hatanaka [Fri, 6 Nov 2015 23:56:15 +0000 (23:56 +0000)]
Add support for function attribute 'not_tail_called'.
This attribute is used to prevent tail-call optimizations to the marked
function. For example, in the following piece of code, foo1 will not be
tail-call optimized:
int __attribute__((not_tail_called)) foo1(int);
int foo2(int a) {
return foo1(a); // Tail-call optimization is not performed.
}
The attribute has effect only on statically bound calls. It has no
effect on indirect calls. Also, virtual functions and objective-c
methods cannot be marked as 'not_tail_called'.
Kuba Brecka [Fri, 6 Nov 2015 15:09:20 +0000 (15:09 +0000)]
[tsan] Add Clang frontend support for TSan on OS X
We're currently in process of porting TSan to OS X, and quite a few of the initial support in the runtime library has already landed in trunk. This patch actually enables "-fsanitize=thread" in the frontend.
Yury Gribov [Fri, 6 Nov 2015 11:16:31 +0000 (11:16 +0000)]
[analyzer] Add VforkChecker to find unsafe code in vforked process.
This checker looks for unsafe constructs in vforked process:
function calls (excluding whitelist), memory write and returns.
This was originally motivated by a vfork-related bug in xtables package.
Keno Fischer [Thu, 5 Nov 2015 23:18:44 +0000 (23:18 +0000)]
Fix crash in EmitDeclMetadata mode
Summary: This fixes a bug that's easily encountered in LLDB
(https://llvm.org/bugs/show_bug.cgi?id=22875). The problem here is that we
mangle a name during debug info emission, but never actually emit the actual
Decl, so we run into problems in EmitDeclMetadata (which assumes such a Decl
exists). Fix that by just skipping metadata emissions for mangled names that
don't have associated Decls.
Richard Smith [Thu, 5 Nov 2015 21:16:22 +0000 (21:16 +0000)]
PR25368: Replace workaround for build failure with modules enabled with a fix
for the root cause. The 'using llvm::isa;' declaration in Basic/LLVM.h only
pulls the declarations of llvm::isa that were declared prior to it into
namespace clang. In a modules build, this is a hermetic set of just the
declarations from LLVM. In a non-modules build, we happened to also pull the
declaration from lib/CodeGen/Address.h into namespace clang, which made the
code in question accidentally compile.
Devin Coughlin [Thu, 5 Nov 2015 18:56:42 +0000 (18:56 +0000)]
[analyzer] Update RegionStoreManager::getBinding to handle BlockDataRegions
Update RegionStoreManager::getBinding() to return UnknownVal when trying to get
the binding for a BlockDataRegion. Previously, getBinding() would try to cast the
BlockDataRegion to a TypedValueRegion and crash. This happened when a block
was passed as a parameter to an inlined function for which
StackHintGeneratorForSymbol::getMessage() tried to generate a stack hint message.
Richard Smith [Thu, 5 Nov 2015 00:54:55 +0000 (00:54 +0000)]
[modules] If we're given a module file, via -fmodule-file=, for a module, but
we can't load that file due to a configuration mismatch, and implicit module
building is disabled, and the user turns off the error-by-default warning for
that situation, then fall back to textual inclusion for the module rather than
giving an error if any of its headers are included.
Aaron Ballman [Thu, 5 Nov 2015 00:06:05 +0000 (00:06 +0000)]
The control expression for a _Generic selection expression should have
its type decayed and qualifiers stripped when determining which
selection it matches. Fixes PR16340.
Devin Coughlin [Wed, 4 Nov 2015 21:33:41 +0000 (21:33 +0000)]
[analyzer] Add 'optin' checker package and move localizability checkers into it.
This commit creates a new 'optin' top-level checker package and moves several of
the localizability checkers into it.
This package is for checkers that are not alpha and that would normally be on by
default but where the driver does not have enough information to determine when
they are applicable. The localizability checkers fit this criterion because the
driver cannot determine whether a project is localized or not -- this is best
determined at the IDE or build-system level.
This new package is *not* intended for checkers that are too noisy to be on by
default.
The hierarchy under 'optin' mirrors that in 'alpha': checkers under 'optin'
should be organized in the hierarchy they would have had if they were truly top
level (e.g., optin.osx.cocoa.MyOptInChecker).
Richard Smith [Wed, 4 Nov 2015 19:26:32 +0000 (19:26 +0000)]
[modules] Generalize the workaround for multiple ambiguous definitions of
internal linkage entities in different modules from r250884 to apply to all
names, not just function names.
This is really awkward: we don't want to merge internal-linkage symbols from
separate modules, because they might not actually be defining the same entity.
But we don't want to reject programs that use such an ambiguous symbol if those
internal-linkage symbols are in fact equivalent. For now, we're resolving the
ambiguity by picking one of the equivalent definitions as an extension.
Douglas Gregor [Tue, 3 Nov 2015 18:33:07 +0000 (18:33 +0000)]
Introduce module file extensions to piggy-back data onto module files.
Introduce the notion of a module file extension, which introduces
additional information into a module file at the time it is built that
can then be queried when the module file is read. Module file
extensions are identified by a block name (which must be unique to the
extension) and can write any bitstream records into their own
extension block within the module file. When a module file is loaded,
any extension blocks are matched up with module file extension
readers, that are per-module-file and are given access to the input
bitstream.
Note that module file extensions can only be introduced by
programmatic clients that have access to the CompilerInvocation. There
is only one such extension at the moment, which is used for testing
the module file extension harness. As a future direction, one could
imagine allowing the plugin mechanism to introduce new module file
extensions.
Tim Northover [Tue, 3 Nov 2015 18:14:42 +0000 (18:14 +0000)]
Revert "Support watchOS and tvOS in compiler-rt builds"
The required compiler-rt changes aren't present yet so attempting to
build with compiler-rt breaks. And since we're trying to deprecate
autotools we actually want to fix this in CMake primarily anyway.
Richard Smith [Tue, 3 Nov 2015 03:13:11 +0000 (03:13 +0000)]
[modules] Rationalize the behavior of Decl::declarationReplaces, and in
particular don't assume that two declarations of the same kind in the same
context are declaring the same entity. That's not true when the same name is
declared multiple times as internal-linkage symbols within a module.
(getCanonicalDecl is cheap now, so we can just use it here.)
Richard Smith [Tue, 3 Nov 2015 01:19:56 +0000 (01:19 +0000)]
Switch to using an explicit scope object to ensure we don't forget to pop ObjC
type parameters off the scope, and fix the cases where we failed to do so.
Douglas Gregor [Tue, 3 Nov 2015 01:15:46 +0000 (01:15 +0000)]
Stop back-patching 'readonly' Objective-C properties with 'readwrite' ones.
A 'readonly' Objective-C property declared in the primary class can
effectively be shadowed by a 'readwrite' property declared within an
extension of that class, so long as the types and attributes of the
two property declarations are compatible.
Previously, this functionality was implemented by back-patching the
original 'readonly' property to make it 'readwrite', destroying source
information and causing some hideously redundant, incorrect
code. Simplify the implementation to express how this should actually
be modeled: as a separate property declaration in the extension that
shadows (via the name lookup rules) the declaration in the primary
class. While here, correct some broken Fix-Its, eliminate a pile of
redundant code, clean up the ARC migrator's handling of properties
declared in extensions, and fix debug info's naming of methods that
come from categories.
A wonderous side effect of doing this write is that it eliminates the
"AddedObjCPropertyInClassExtension" method from the AST mutation
listener, which in turn eliminates the last place where we rewrite
entire declarations in a chained PCH file or a module file. This
change (which fixes rdar://problem/18475765) will allow us to
eliminate the rewritten-decls logic from the serialization library,
and fixes a crash (rdar://problem/23247794) illustrated by the
test/PCH/chain-categories.m example.
Make hasLHS and hasRHS matchers available for ArraySubscriptExpr
Summary:
The hasBase and hasIndex don't tell anything about the position of the
base and the index in the code, so we need hasLHS and hasRHS in some cases.
Reid Kleckner [Mon, 2 Nov 2015 20:47:31 +0000 (20:47 +0000)]
[PTH] Fix data length used for stat cache entries
This came up in a boost build, which apparently uses PTH. This was
broken in r187619 when we migrated it to uses llvm::fs instead of raw
stat calls.
Constructing a test case with a hash table collision in-tree is tough.
Instead, I have a pending change to OnDiskChainedHashTable that asserts
that the reported length of the data agrees with the data actually
written. All of the existing in-tree tests find the bug with this
assert.
projects / clang / log