Daniel Jasper [Fri, 18 Oct 2013 17:20:57 +0000 (17:20 +0000)]
clang-format: Be more aggressive on incorrect code.
Before, clang-format would not adjust leading indents if it found a
structural error (e.g. unmatched {}). It seems, however, that
clang-format has gotten good enough at parsing the code structure that
this hurts in almost all cases. Commonly, while writing code, it is
very useful to be able to correclty indent incomplete if statements or
for loops.
In case this leads to errors that we don't anticipate, we need to find
out and fix those.
Richard Smith [Fri, 18 Oct 2013 06:05:18 +0000 (06:05 +0000)]
Basic ODR checking for C++ modules:
If we have multiple definitions of the same entity from different modules, we
nominate the first definition which we see as being the canonical definition.
If we load a declaration from a different definition and we can't find a
corresponding declaration in the canonical definition, issue a diagnostic.
This is insufficient to prevent things from going horribly wrong in all cases
-- we might be in the middle of emitting IR for a function when we trigger some
deserialization and discover that it refers to an incoherent piece of the AST,
by which point it's probably too late to bail out -- but we'll at least produce
a diagnostic.
Alp Toker [Fri, 18 Oct 2013 05:54:24 +0000 (05:54 +0000)]
Check "late parsed" friend functions for redefinition
r177003 applied the late parsed template technique to friend functions
but omitted the corresponding check for redefinitions.
This patch adds the same check already in use for templates to the
new code path in order to diagnose and reject invalid redefinitions
that were being silently accepted.
Alp Toker [Fri, 18 Oct 2013 05:54:19 +0000 (05:54 +0000)]
Fix missed exception spec checks and crashes
Delayed exception specification checking for defaulted members and virtual
destructors are both susceptible to mutation during iteration so we need to
swap and process the worklists.
This resolves both accepts-invalid and rejects-valid issues and moreover fixes
potential invalid memory access as the contents of the vectors change during
iteration and recursive template instantiation.
Checking can be further delayed where parent classes aren't yet fully defined.
This patch adds two assertions at end of TU to ensure no specs are left
unchecked as was happenning before the fix, plus a test case from Marshall Clow
for the defaulted member crash extracted from the libcxx headers.
Consumed analysis: Add param_typestate attribute, which specifies that
function parameters must be in a particular state. Patch by
chris.wailes@gmail.com. Reviewed by delesley@google.com.
Consumed Analysis: Allow parameters that are passed by non-const reference
to be treated as return values, and marked with the "returned_typestate"
attribute. Patch by chris.wailes@gmail.com; reviewed by delesley@google.com.
ObjectiveC migrator. In infering NS_ENUM/NS_OPTIONS
macros, prefer the typedef immediately following the
enum declaration to the one preceeding it.
// rdar://15200915
Hans Wennborg [Thu, 17 Oct 2013 20:41:18 +0000 (20:41 +0000)]
Make .asm a valid extension for assembly files
This is a common extension on Windows, and now clang will assemble them
instead of treating them as linker input which is the default for unknown
file types.
Alp Toker [Thu, 17 Oct 2013 19:12:03 +0000 (19:12 +0000)]
Fix missed exception spec checks and crashes
Delayed exception specification checking for defaulted members and virtual
destructors are both susceptible to mutation during iteration so we need to
process the worklists fully.
This resolves both accepts-invalid and rejects-valid issues and moreover fixes
potential invalid memory access as the contents of the vectors change during
iteration and recursive template instantiation.
This patch also adds two assertions at end of TU to ensure no specs are left
unchecked as was happenning before the fix, plus a test case from Marshall Clow
for the defaulted member crash extracted from the libcxx headers.
Chad Rosier [Thu, 17 Oct 2013 18:12:50 +0000 (18:12 +0000)]
[AArch64] Add support for NEON scalar three register different instruction
class. The instruction class includes the signed saturating doubling
multiply-add long, signed saturating doubling multiply-subtract long, and
the signed saturating doubling multiply long instructions.
Hans Wennborg [Thu, 17 Oct 2013 16:16:23 +0000 (16:16 +0000)]
clang-cl: Add support for asm listings (/FA and /Fa)
This adds support for outputing the assembly to a file during compilation.
It does this by changing the compilation pipeling to not use the integrated
assembler, and keep the intermediate assembler file.
Daniel Sanders [Thu, 17 Oct 2013 14:55:58 +0000 (14:55 +0000)]
[mips] Added -mfp64 and -mfp32 options.
These options specify 64-bit FP registers and 32-bit FP registers respectively.
When using -mfp32, the FPU has 16x double-precision registers overlapping with
the 32x single-precision registers (each double-precision register overlaps
two single-precision registers).
When using -mfp64, the FPU has 32x double-precision registers overlapping with
the 32x single-precision registers (each double-precision register overlaps
with one single-precision register and has an additional 32-bits).
Chandler Carruth [Thu, 17 Oct 2013 03:28:24 +0000 (03:28 +0000)]
Add support to the Clang driver for forwarding all of gfortran's flags
to GCC when asked to compile a fortran input.
This fixes a regression with essentially every Fortran compile since we
started rejecting unknown flags. Also moves a mis-classified gfortran
flag into the nicely documented set.
Reid Kleckner [Thu, 17 Oct 2013 01:29:48 +0000 (01:29 +0000)]
Add implementations of the MSVC barrier intrinsics
Summary:
These are deprecated in VS 2012 according to MSDN. They don't actually
compile down to any code. They prevent the compiler from reordering
memory accesses across the barrier, which is what a memory-clobbering
volatile asm does.
Yunzhong Gao [Wed, 16 Oct 2013 19:07:02 +0000 (19:07 +0000)]
Enabling 3DNow! prefetch instruction support for a few AMD processors in the
clang front end. This change will allow the __PRFCHW__ macro to be set on these
processors and hence include prfchwintrin.h in x86intrin.h header. Support for
the intrinsic itself seems to have already been added in r178041.
Jordan Rose [Wed, 16 Oct 2013 17:45:35 +0000 (17:45 +0000)]
[analyzer] Don't draw edges to C++11 in-class member initializers.
Since these aren't lexically in the constructor, drawing arrows would
be a horrible jump across the body of the class. We could still do
better here by skipping over unimportant initializers, but this at least
keeps everything within the body of the constructor.
Rafael Espindola [Wed, 16 Oct 2013 01:40:34 +0000 (01:40 +0000)]
On 32 bit windows, mangle stdcall and fastcall decls in clang.
This removes the dependency on the llvm mangler doing it for us. In isolation,
the benefit is that the testing of what mangling is applied is all in one place:
(C, C++) X (Itanium, Microsoft) are all handled by clang.
This also gives me hope that in the future the llvm mangler (and llvm-ar) will
not depend on TargetMachine.
Yunzhong Gao [Tue, 15 Oct 2013 21:33:02 +0000 (21:33 +0000)]
Remove this regression test now that PR17578 is fixed (r192674). The fix is to
reject this test case at the Sema stage, so no CodeGen test should be needed.
Reid Kleckner [Tue, 15 Oct 2013 18:38:02 +0000 (18:38 +0000)]
ms-compat: Fix taking the address of a member of a dependent base
If unqualified id lookup fails while parsing a class template with a
dependent base, clang with -fms-compatibility will pretend the user
prefixed the name with 'this->' in order to delay the lookup. However,
if there was a unary ampersand, Sema::ActOnDependentIdExpression() will
create a DependentDeclRefExpr, which is not what we wanted at all. Fix
this by building the CXXDependentScopeMemberExpr directly instead.
In order to be fully MSVC compatible, we would have to defer all
attempts at name lookup to instantiation time. However, until we have
real problems with system headers that can't be parsed, we'll put off
implementing that.
ObjectiveC driver. Scrap -fno-objc-legacy-dispatch for NeXT
runtime. It will be silently ignored and regardless
of deployment target. // rdar://14803286
David Majnemer [Tue, 15 Oct 2013 09:30:14 +0000 (09:30 +0000)]
Sema: Don't crash when __try/__except/__finally appears in a template function
We wouldn't transform the compound statement in any of these forms,
causing crashes when it got time to act on them. Additionally, we
wouldn't check to see if the handler was invalid before deciding whether
or not we should continue acting on the __try.
David Majnemer [Tue, 15 Oct 2013 08:33:43 +0000 (08:33 +0000)]
docs: Remove incompatibility with Solaris shell
There doesn't seem to be a need in checking if a directory exists if we
will just rm -rf it once we affirm that it does. Instead, just blindly
try to delete it.
Reid Kleckner [Tue, 15 Oct 2013 01:18:02 +0000 (01:18 +0000)]
[ms-cxxabi] Fix assert in unspecified inheritance memptr emission
If a class is using the unspecified inheritance model for member
pointers and later we find the class is defined to use single
inheritance, zero out the vbptr offset field of the member pointer when
it is formed.
ObjectiveC migrator: Support for more possibility of
migration to NS_ENUM/NS_OPTIONS macros; when
typedef'ed to NSInteger/NSUInteger preceeds well
before of the enum declaration. // rdar://15201056
Richard Smith [Tue, 15 Oct 2013 00:00:26 +0000 (00:00 +0000)]
PR17567: Improve diagnostic for a mistyped constructor name. If we see something
that looks like a function declaration, except that it's missing a return type,
try typo-correcting it to the relevant constructor name.
In passing, fix a bug where the missing-type-specifier recovery codepath would
drop a preceding scope specifier on the floor, leading to follow-on diagnostics
and incorrect recovery for the auto-in-c++98 hack.
Reid Kleckner [Mon, 14 Oct 2013 21:14:05 +0000 (21:14 +0000)]
PR17576: Fix assertion on polymorphic classes with small alignment
We have to reserve at least the width of a pointer for the vfptr. For
classes with small alignment, we weren't reserving enough space, and
were overlapping the first field with the vfptr.
projects / clang / log