Manuel Klimek [Tue, 1 Mar 2016 10:56:19 +0000 (10:56 +0000)]
Optionally demote fatal errors to non-fatal errors.
This behavior is enabled when the new CXTranslationUnit_KeepGoing
option is passed to clang_parseTranslationUnit{,2}. It is geared
towards use by IDEs and similar consumers of the clang-c API where
fatal errors may arise when parsing incomplete code mid-edit, or
when include paths are not properly configured yet. In such situations
one still wants to get as much information as possible about a TU.
Previously, the semantic analysis would not instantiate templates
or report additional fatal errors after the first fatal error was
encountered.
Sean Silva [Tue, 1 Mar 2016 01:20:15 +0000 (01:20 +0000)]
Make test more robust.
Really all these tests are checking is that we find a file path. The
behavior when we don't find one will have `"ps4-ld"` in it. We just need
a path separator to know that a path has been found.
The root cause of the flakiness of these tests is the same on Windows
and non-Windows: setting the PATH environment variable is not sufficient
to guarantee that a particular path is looked up first.
Driver::GetProgramPath checks some paths before deferring to PATH
(in particular, the directory containing the clang binary itself).
I initally ran into this on Windows when putting a PS4 linker in
build-dir/bin/ps4-ld for testing.
After digging for a while thinking that it was some windows path search
oddity (the Windows SearchPathW documentation indicates that its
behavior varies depending on a registry setting...).
I eventually tried reproducing the issue on Mac and to my surprise found
the same issue.
Ultimately I traced it down to the extra lookups in
Driver::GetProgramPath.
Steven Wu [Tue, 1 Mar 2016 01:07:58 +0000 (01:07 +0000)]
Introduce -fembed-bitcode driver option
Summary:
This is the clang driver part of the change to embedded bitcode. This
includes:
1. -fembed-bitcode option which breaks down the compilation into two
stages. The first stage emits optimized bitcode and the second stage
compiles bitcode into object file.
2. -fembed-bitcode-marker option which doesn't really break down to
two stages to speedup the compilation flow.
3. pass the correct linker flag to darwin linker if tool chains supports
embedded bitcode.
Devin Coughlin [Mon, 29 Feb 2016 21:44:08 +0000 (21:44 +0000)]
[analyzer] Don't treat calls to system headers as escaping in CheckObjCDealloc.
This prevents false negatives when a -dealloc method, for example, removes itself as
as an observer with [[NSNotificationCenter defaultCenter] removeObserver:self]. It is
unlikely that passing 'self' to a system header method will release 'self''s instance
variables, so this is unlikely to produce false positives.
A challenge here is that while CheckObjCDealloc no longer treats these calls as
escaping, the rest of the analyzer still does. In particular, this means that loads
from the same instance variable before and after a call to a system header will
result in different symbols being loaded by the region store. To account for this,
the checker now treats different ivar symbols with the same instance and ivar decl as
the same for the purpose of release checking and more eagerly removes a release
requirement when an instance variable is assumed to be nil. This was not needed before
because when an ivar escaped its release requirement was always removed -- now the
requirement is not removed for calls to system headers.
Craig Topper [Mon, 29 Feb 2016 06:51:38 +0000 (06:51 +0000)]
[X86] Disabling avx512f should also disable avx512vbmi and avx512ifma. Enabling avx512vbmi or avx512ifma should enable avx512f. Add command line switches and header defines for avx512ifma and avx512vbmi.
Alexey Bataev [Mon, 29 Feb 2016 05:54:20 +0000 (05:54 +0000)]
[OPENMP 4.5] Initial support for data members in 'reduction' clauses.
OpenMP 4.5 allows to privatize non-static data members of current class
in non-static member functions. Patch adds initial parsing/semantic
analysis for data members support in 'reduction' clauses.
David Majnemer [Mon, 29 Feb 2016 01:40:36 +0000 (01:40 +0000)]
[clang-cl] /EHc should not effect functions with explicit exception specifications
Functions with an explicit exception specification have their behavior
dictated by the specification. The additional /EHc behavior only comes
into play if no exception specification is given.
Vassil Vassilev [Sun, 28 Feb 2016 19:08:24 +0000 (19:08 +0000)]
[modules] Prefer more complete array types.
If we import a module that has a complete array type and one that has an
incomplete array type, the declaration found by name lookup might be the one with
the incomplete type, possibly resulting in rejects-valid.
Now, the name lookup prefers decls with a complete array types. Also,
diagnose cases when the redecl chain has array bound, different from the merge
candidate.
Logan Chien [Sun, 28 Feb 2016 15:01:42 +0000 (15:01 +0000)]
Add ARM EHABI-related constants to unwind.h.
Adds a number of constants, defined in the ARM EHABI spec, to the Clang
lib/Headers/unwind.h header. This is prerequisite for landing
http://reviews.llvm.org/D15781, as previously discussed there.
Richard Smith [Sat, 27 Feb 2016 02:36:43 +0000 (02:36 +0000)]
Fix bug in using shadow decl checking: a using shadow decl should not conflict
with a prior UsingDecl -- those should not even really be found by the lookup
here, except that we use the same lookup results for two different checks, and
the other check needs them.
This happens to work in *almost all* cases, because either the lookup results
list the UsingDecl first (and the NonTag result gets replaced by something
else) or because the problematic declaration is a function (which causes us to
use different logic to detect conflicts). This can also be triggered from a
state only reachable through modules (where the name lookup results can contain
multiple UsingDecls in the same scope).
Chris Bieneman [Fri, 26 Feb 2016 21:23:59 +0000 (21:23 +0000)]
[CMake] Only configure Native target in stage 1, configure all in other stages
This patch causes the 3-stage build pipeline to only build a host compiler in the first stage, and to build all targets for subsequent stages. The host target is determined via the Native target specifier added in r262070.
Reid Kleckner [Fri, 26 Feb 2016 19:51:02 +0000 (19:51 +0000)]
[dllexport] Sort out emission order of delayed exported classes
Relands r260194 with a fix. If we have a template that transitions from
an extern template to an explicitly instantiated dllexport template, we
would add that class to the delayed exported class list without flushing
it.
For explicit instantiations, we can just flush the list of delayed
classes immediately. We don't have to worry about the bug fixed in
r260194 in this case because explicit instantiations can only occur at
file and namespace scope.
SemaCXX: Support templates in availability attributes
If the availability context is `FunctionTemplateDecl`, we should look
through it to the `FunctionDecl`. This prevents a diagnostic in the
following case:
class C __attribute__((unavailable));
template <class T> void foo(C&) __attribute__((unavailable));
This adds tests for availability in templates in many other cases, but
that was the only case that failed before this patch.
I added a feature `__has_feature(attribute_availability_in_templates)`
so users can test for this.
Andy Gibbs [Fri, 26 Feb 2016 15:35:16 +0000 (15:35 +0000)]
Reduce false positives in printf/scanf format checker
Summary:
The printf/scanf format checker is a little over-zealous in handling the conditional operator. This patch reduces work by not checking code-paths that are never used and reduces false positives regarding uncovered arguments, for example in the code fragment:
David Majnemer [Fri, 26 Feb 2016 04:23:19 +0000 (04:23 +0000)]
[MSVC Compat] Don't evaluate member base expressions w/o side effects
A member expression's base doesn't always have an impact on what the
member decl would evaluate to. In such a case, the base is used as a
poor man's scope qualifier.
Change "use of 'self' after it has been freed with call to [super dealloc]" to
"use of 'self' after it has been deallocated" and "use of instance variable
'_ivar' after the instance has been freed with call to [super dealloc]" to
"use of instance variable '_ivar' after 'self' has been deallocated".
Devin Coughlin [Thu, 25 Feb 2016 23:36:52 +0000 (23:36 +0000)]
[analyzer] Warn on use of 'self' after call to to [super dealloc].
Referring to 'self' after a call to [super dealloc] is a use-after-free in
Objective-C because NSObject's -dealloc frees the memory pointed to by self.
This patch extends the ObjCSuperDeallocChecker to catch this error.
Devin Coughlin [Thu, 25 Feb 2016 21:15:16 +0000 (21:15 +0000)]
[analyzer] Reapply r261917 with a fix.
This reapplies "[analyzer] Make ObjCDeallocChecker path sensitive." (r261917)
with a fix for an error on some bots about specializing a template
from another namespace.
Devin Coughlin [Thu, 25 Feb 2016 18:55:24 +0000 (18:55 +0000)]
[analyzer] Make ObjCDeallocChecker path sensitive.
Convert the ObjCDeallocChecker to be path sensitive. The primary
motivation for this change is to prevent false positives when -dealloc calls
helper invalidation methods to release instance variables, but it additionally
improves precision when -dealloc contains control flow. It also reduces the need
for pattern matching. The check for missing -dealloc methods remains AST-based.
Aaron Ballman [Thu, 25 Feb 2016 15:14:09 +0000 (15:14 +0000)]
Replace a compiler-specific approach to determining the presence of a getDecl() member function with one that does not require compiler-specific workarounds; NFC.
Akira Hatanaka [Thu, 25 Feb 2016 07:08:33 +0000 (07:08 +0000)]
[Sema] Remove assert in TreeTransform<Derived>::TransformObjCObjectType.
The assert isn't correct since TypeLoc::ObjCObjectTypeLoc doesn't
indicate whether the type is a dependent type. The function returns
false for a type like "<SomeProtocol>" which is a synonym for
"id<SomeProtocol>".
Alexey Bataev [Thu, 25 Feb 2016 05:25:57 +0000 (05:25 +0000)]
[OPENMP 4.5] Codegen for member decls in 'lastprivate' clause.
OpenMP 4.5 allows to privatize non-static member decls in non-static
member functions. Patch captures such decls by reference in general (for
bitfields, by value) and then operates with this capture. For bitfields,
at the end of codegen for lastprivates original bitfield is updated with the value of captured copy.
Faisal Vali [Thu, 25 Feb 2016 05:09:30 +0000 (05:09 +0000)]
Twek fix for PR24473 (r261506) - Preserve the template arguments as written
(within the DeclRefExpr Node) when creating AST nodes that reference specializations of static data member templates. While we pass the template args through for all non-instance members, they should only be relevant (i.e. non-null) for variable template ids (assertion added for that assumption)
Also preserve the FoundDecl that refers to the canonical Decl (the primary VarTemplateDecl for a variable template specialization) that we are referencing in our DeclRefExpr. Not sure why this was not being done for non-variable template-ids.
No functionality change - so no tests added.
Thanks to Richard Smith for drawing my attention to this!
Alexey Bataev [Thu, 25 Feb 2016 03:59:29 +0000 (03:59 +0000)]
[OPENMP 4.5] Support fielddecls in 'shared' clause.
OpenMP 4.5 allows to use non-static data members of current class in
non-static member functions in different kind of clauses. Patch adds
support for data members in 'shared' clause.
Manman Ren [Wed, 24 Feb 2016 23:05:43 +0000 (23:05 +0000)]
Fix assertion failure on MaybeODRUseExprs.
In VisitNonTypeTemplateParamDecl, before SubstExpr with the default argument,
we should create a ConstantEvaluated ExpressionEvaluationContext. Without this,
it is possible to use a PotentiallyEvaluated ExpressionEvaluationContext; and
MaybeODRUseExprs will not be cleared when popping the context, causing
assertion failure.
This is similar to how we handle the context before SubstExpr with the
default argument, in SubstDefaultTemplateArgument.
David Majnemer [Wed, 24 Feb 2016 22:07:26 +0000 (22:07 +0000)]
Revert "Don't convert a char to a const char *"
This reverts commit r261780. It turns out the original code was just
fine. An overload for ltrim which takes char was added but the Doxygen
docs haven't seemed to pick it up.
Richard Smith [Wed, 24 Feb 2016 21:59:10 +0000 (21:59 +0000)]
PR26237: Fix iterator invalidation bug that occurs if serializing
specializations of a template manages to trigger deserialization of more
specializations of the same template.
No test case provided: this is hard to reliably test due to standard library
differences.
Without this change, if the optimizer does not inline sync() (which it
won't because __syncthreads is also marked as noduplicate, for now
anyway), it is free to perform optimizations on sync() that it would not
be able to perform on __syncthreads(), because sync() is not marked as
convergent.
Similarly, we need a notion of convergent calls, since in the case when
we can't statically determine a call's target(s), we need to know
whether it's safe to perform optimizations around the call.
This change is conservative; the optimizer will remove these attrs where
it can, see r260318, r260319.
Justin Lebar [Wed, 24 Feb 2016 21:49:31 +0000 (21:49 +0000)]
[CUDA] Add hack so code which includes "curand.h" doesn't break.
Summary:
curand.h includes curand_mtgp32_kernel.h. In host mode, this header
redefines threadIdx and blockDim, giving them their "proper" types of
uint3 and dim3, respectively.
clang has its own plan for these variables -- their types are magic
builtin classes. So these redefinitions are incompatible.
As a hack, we force-include the offending CUDA header and use #defines
to get the right types for threadIdx and blockDim.
Justin Lebar [Wed, 24 Feb 2016 21:49:28 +0000 (21:49 +0000)]
Bail on compilation as soon as a job fails.
Summary:
(Re-land of r260448, which was reverted in r260522 due to a test failure
in Driver/output-file-cleanup.c that only showed up in fresh builds.)
Previously we attempted to be smart; if one job failed, we'd run all
jobs that didn't depend on the failing job.
Problem is, this doesn't work well for e.g. CUDA compilation without
-save-temps. In this case, the device-side and host-side Assemble
actions (which actually are responsible for preprocess, compile,
backend, and assemble, since we're not saving temps) are necessarily
distinct. So our clever heuristic doesn't help us, and we repeat every
error message once for host and once for each device arch.
The main effect of this change, other than fixing CUDA, is that if you
pass multiple cc files to one instance of clang and you get a compile
error, we'll stop when the first cc1 job fails.
Reviewers: echristo
Subscribers: cfe-commits, jhen, echristo, tra, rafael
Nico Weber [Wed, 24 Feb 2016 20:58:14 +0000 (20:58 +0000)]
Fix rejects-valid caused by r261297.
r261297 called hasUserProvidedDefaultConstructor() to check if defining a
const object is ok. This is incorrect for this example:
struct X { template<typename ...T> X(T...); int n; };
const X x; // formerly OK, now bogus error
Instead, track if a class has a defaulted default constructor, and disallow
a const object for classes that either have defaulted default constructors or
if they need an implicit constructor.
Bug report and fix approach by Richard Smith, thanks!
James Y Knight [Wed, 24 Feb 2016 02:59:33 +0000 (02:59 +0000)]
Default vaarg lowering should support indirect struct types.
Fixes PR11517 for SPARC.
On most targets, clang lowers va_arg itself, eschewing the use of the
llvm vaarg instruction. This is necessary (at least for now) as the type
argument to the vaarg instruction cannot represent all the ABI
information that is needed to support complex calling conventions.
However, on targets with a simpler varrags ABIs, the LLVM instruction
can work just fine, and clang can simply lower to it. Unfortunately,
even on such targets, vaarg with a struct argument would fail, because
the default lowering to vaarg was naive: it didn't take into account the
ABI attribute computed by classifyArgumentType. In particular, for the
DefaultABIInfo, structs are supposed to be passed indirectly and so
llvm's vaarg instruction should be emitted with a pointer argument.
Now, vaarg instruction emission is able to use computed ABIArgInfo for
the provided argument type, which allows the default ABI support to work
for structs too.
I haven't touched the EmitVAArg implementation for PPC32_SVR4 or XCore,
although I believe both are now redundant, and could be switched over to
use the default implementation as well.
Devin Coughlin [Tue, 23 Feb 2016 22:26:04 +0000 (22:26 +0000)]
[analyzer] Find ObjC 'self' decl even when block captures local named 'self'.
When looking up the 'self' decl in block captures, make sure to find the actual
self declaration even when the block captures a local variable named 'self'.