Hans Wennborg [Tue, 14 Jan 2014 19:35:09 +0000 (19:35 +0000)]
Remove the -cxx-abi command-line flag.
This makes the C++ ABI depend entirely on the target: MS ABI for -win32 triples,
Itanium otherwise. It's no longer possible to do weird combinations.
To be able to run a test with a specific ABI without constraining it to a
specific triple, new substitutions are added to lit: %itanium_abi_triple and
%ms_abi_triple can be used to get the current target triple adjusted to the
desired ABI. For example, if the test suite is running with the i686-pc-win32
target, %itanium_abi_triple will expand to i686-pc-mingw32.
Tim Northover [Tue, 14 Jan 2014 19:26:03 +0000 (19:26 +0000)]
Darwin: add __sinpi (etc) and __exp10 libbuiltins
These functions have the same constness properties of the normal libm
functions, which allows LLVM to optimise code better in general. There
are also a couple of specific optimisations that only trigger when
these are properly marked.
Alp Toker [Tue, 14 Jan 2014 19:13:00 +0000 (19:13 +0000)]
Clarify driver/frontend -fms-compatibility help text
Better describe the flag that enables drop-in MSVC compatibility, including
ability to parse MS standard headers. This is intended to distinguish it from
-fms-extensions, the more established and 'gentler' flag also supported by GCC.
The new wording matches up with the internal description introduced in r198936.
Still room for improvement (e.g. C++ is part of the product name, yet the flag
also applies to C) but it's a step forward from "Microsoft mode".
Aaron Ballman [Tue, 14 Jan 2014 17:41:53 +0000 (17:41 +0000)]
Simplifying the OpenCL image attribute. It does not need a semantic integer parameter because the required information is encoded in the spelling. Added an appropriate subject to the attribute, and simplified the semantic checking (which will likely be expanded upon in a future patch). Also, removed the GNU spelling since it was unsupported in the first place.
Jordan Rose [Tue, 14 Jan 2014 17:29:00 +0000 (17:29 +0000)]
Use a proper lvalue-to-rvalue conversion in Objective-C++ property accessors.
Previously, the synthesized AST contained an rvalue DeclRefExpr for 'self'.
Now, it has an lvalue DeclRefExpr wrapped in an lvalue-to-rvalue
ImplicitCastExpr, which is what's generated when an ivar access is written
in the source.
Alp Toker [Tue, 14 Jan 2014 12:51:41 +0000 (12:51 +0000)]
Rename language option MicrosoftMode to MSVCCompat
There's been long-standing confusion over the role of these two options. This
commit makes the necessary changes to differentiate them clearly, following up
from r198936.
MicrosoftExt (aka. fms-extensions):
Enable largely unobjectionable Microsoft language extensions to ease
portability. This mode, also supported by gcc, is used for building software
like FreeBSD and Linux kernel extensions that share code with Windows drivers.
MSVCCompat (aka. -fms-compatibility, formerly MicrosoftMode):
Turn on a special mode supporting 'heinous' extensions for drop-in
compatibility with the Microsoft Visual C++ product. Standards-compilant C and
C++ code isn't guaranteed to work in this mode. Implies MicrosoftExt.
Note that full -fms-compatibility mode is currently enabled by default on the
Windows target, which may need tuning to serve as a reasonable default.
See cfe-commits for the full discourse, thread 'r198497 - Move MS predefined
type_info out of InitializePredefinedMacros'
Daniel Jasper [Tue, 14 Jan 2014 09:53:07 +0000 (09:53 +0000)]
clang-format: Fix bug introduced in r198871.
We cannot simply change the start column to accomodate for the @ in an
ObjC string literal as that will make clang-format happily violate the
column limit.
Use a different workaround instead. However, a better long-term
solution might be to join the @ and the rest of the literal into a
single token.
David Majnemer [Tue, 14 Jan 2014 08:18:49 +0000 (08:18 +0000)]
Switch to ssize_t from size_t to unbreak windows builders.
Builders that have -fms-compatibility on by default define size_t implicitly.
Tests that provide conflicting definitions would cause unintended failures.
David Majnemer [Tue, 14 Jan 2014 06:19:35 +0000 (06:19 +0000)]
Sema: Predefine size_t in MSVC Compatibility mode
MSVC defines size_t without any explicit declarations. This change
allows us to be compatible with TUs that depend on this declaration
appearing from nowhere.
Aaron Ballman [Tue, 14 Jan 2014 01:29:54 +0000 (01:29 +0000)]
Removing some attribute magic related to the OpenCL keyword attributes. Instead of mapping them to their semantics as a custom part of the parser, they instead map declaratively through the rest of the attribute system.
Warren Hunt [Tue, 14 Jan 2014 00:54:36 +0000 (00:54 +0000)]
[ms-abi] Small Change to pack+alignment interaction.
This patch makes a small behavioral change to the interaction between
pack and alignment. Specifically it makes __declspec(align()) on a
field change that field's alignment without respect to pack but the
alignment change to the record alignment as a whole still obeys pack.
Consumed analysis: add two new attributes which fine-tune the behavior of
consumable objects. These are useful for implementing error codes that
must be checked. Patch also includes some significant refactoring, which was
necesary to implement the new behavior.
Warren Hunt [Tue, 14 Jan 2014 00:31:30 +0000 (00:31 +0000)]
[ms-abi] Report VBPtrOffset correctly
Although VBPtrs were being placed correctly by the ms-abi layout engine,
their offsets were being improperly reported to the ASTRecordLayout
builder due to a bug. This patch fixes that and fixes the test cases to
use the correct values.
y
Reid Kleckner [Mon, 13 Jan 2014 22:57:31 +0000 (22:57 +0000)]
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI
Fixes PR18435, where we generated a base ctor instead of a complete
ctor, and so failed to construct virtual bases when constructing the
complete object.
This patch moves the check for pragma pack until after the application
of __declspec align to before pragma pack. This causes observable
changes in the use of tail padding for bases. A test case is included.
Chandler Carruth [Mon, 13 Jan 2014 22:23:58 +0000 (22:23 +0000)]
LLVM's CMake is now using a feature that wasn't preasant in CMake 2.8.7,
so bump the minimum version in the standalone Clang CMake project as
well.
As I mentioned on the LLVM commit version of this, if this causes any
trouble for folks, just let me know. I'm trying to avoid re-implementing
functionality in CMake, but I will if there are problems using the newer
versions.
Chandler Carruth [Mon, 13 Jan 2014 22:22:24 +0000 (22:22 +0000)]
Teach the standalone Clang CMake bits about the flag to force using an
old toolchain to build. The toolchain version is now checked in
HandleLLVMOptions.
Aaron Ballman [Mon, 13 Jan 2014 21:42:39 +0000 (21:42 +0000)]
When determining the attribute's parsed kind, pay attention to the syntax used. This fixes bugs where an attribute has differing GNU and Declspec spellings, but they are treated as the same. Eg) __declspec(aligned) when it should be __attribute__((aligned)), and __attribute__((align)) when it should be __declspec(align).
Aaron Ballman [Mon, 13 Jan 2014 21:40:16 +0000 (21:40 +0000)]
__forceinline is a keyword, and not a GNU-style attribute. This FIXME appears to be out-dated, and the attribute syntax is becoming more important these days.
Aaron Ballman [Mon, 13 Jan 2014 21:32:48 +0000 (21:32 +0000)]
Updating the attribute declarations to have the correct syntaxes. This means giving a __declspec spelling to: deprecated, naked, noinline, noreturn, and nothrow. uuid has no GNU spelling, so it was switched to __declspec. dllexport and dllimport both are now supported with GNU spellings.
There are a number of places where we do PGO.setCurrentRegionCount(0)
directly after an unconditional branch. Give this operation a name so
that it's clearer why we're doing this.
Justin Bogner [Mon, 13 Jan 2014 21:24:15 +0000 (21:24 +0000)]
CodeGen: Remove a superfluous setCurrentRegionCount
This call looks like it was an artifact of an earlier change, and
doesn't actually make sense. We begin a new region immediately anyway,
so it was mostly harmless.
Warren Hunt [Mon, 13 Jan 2014 19:55:52 +0000 (19:55 +0000)]
[ms-abi] Leading VFPtrs don't suppress the leading zero sized flag
The MS-ABI tracks a bit that asserts that the first sub-object is zero
sized. This bit is used to add padding between objects if there's the
potential for zero sized objects to alias. The bit is still true even
if the zero sized base is lead by a VFPtr. This patch makes clang mimic
that behavior.
Hans Wennborg [Mon, 13 Jan 2014 19:48:18 +0000 (19:48 +0000)]
Use the MS ABI for Win32 targets by default
In addition to being a sensible default, this is a huge improvement
in test coverage for the MS ABI: any bot that targets Win32 will
now run the test suite using the MS ABI by default.
Jordan Rose [Mon, 13 Jan 2014 17:59:19 +0000 (17:59 +0000)]
[analyzer] Add a CFG node for the allocator call in a C++ 'new' expression.
In an expression like "new (a, b) Foo(x, y)", two things happen:
- Memory is allocated by calling a function named 'operator new'.
- The memory is initialized using the constructor for 'Foo'.
Currently the analyzer only models the second event, though it has special
cases for both the default and placement forms of operator new. This patch
is the first step towards properly modeling both events: it changes the CFG
so that the above expression now generates the following elements.
1. a
2. b
3. (CFGNewAllocator)
4. x
5. y
6. Foo::Foo
The analyzer currently ignores the CFGNewAllocator element, but the next
step is to treat that as a call like any other.
The CFGNewAllocator element is not added to the CFG for analysis-based
warnings, since none of them take advantage of it yet.
Hans Wennborg [Mon, 13 Jan 2014 17:23:24 +0000 (17:23 +0000)]
[ms-cxxabi] Elide dtor access checks for pass-by-val objects in callees
The ABI requires the destructor to be invoked in the callee, but the
standard does not require access checks here so we avoid doing direct
access checks on the destructor.
If we end up needing to define an implicit destructor, we don't skip
access checks for the base class, etc. Those checks are effectively part
of generating the destructor definition, and aren't affected by which TU
the check is performed in.
Chandler Carruth [Mon, 13 Jan 2014 10:56:17 +0000 (10:56 +0000)]
[PM] Update Clang to reflect LLVM r199095 which moves the core DomTree
algorithms and datastructures into the fully generic support library,
separating them (almost) entirely from the LLVM IR. This makes the
reliance on domtrees here *much* cleaner.
It might be worthwhile for someone to use extern templates and other
tools to sink a lot more of this code into the .cpp files instead of the
.h files, but leaving that for someone other than me.
NAKAMURA Takumi [Mon, 13 Jan 2014 05:25:13 +0000 (05:25 +0000)]
[CMake] Move BUG_REPORT_URL from clang to llvm.
It was too late to set BUG_REPORT_URL after configure_file(config.h).
BUG_REPORT_URL in config.h.cmake would be updated at 2nd run of cmake.
It caused many recompilations.
FYI, configure handles BUG_REPORT_URL in llvm side.
Nico Weber [Sun, 12 Jan 2014 23:12:35 +0000 (23:12 +0000)]
Add help text for -MMD, -MD, -MM, -M, -MF.
Also regroup these flags so that alike flags are next to each other, while
keeping the list still mostly alphabetical.
The help text isn't ideal, but I feel it's less maze-like than
http://gcc.gnu.org/onlinedocs/gcc-4.1.2/gcc/Preprocessor-Options.html (look
at the entry for '-MMD' and count how many other entries you need to look up
until you know what it does).
And it looks like -M / -MM are mostly an historical accident and most people
use -MD or -MMD for deps tracking these days, so make -M / -MM refer to
-MD / -MMD instead of the other way round.
Alp Toker [Sun, 12 Jan 2014 15:18:06 +0000 (15:18 +0000)]
Clarify warn_cxx98_compat_attribute diagnostic
Various attribute flavours are supported in C++98. Make it clear that this
compatibility warning relates specifically to C++11-style generalized
attributes.
Richard Smith [Sat, 11 Jan 2014 01:24:05 +0000 (01:24 +0000)]
PR12208: Under -fno-elide-constructors, don't forget to actually copy an NRVO
variable to the return slot. Patch by David Wiberg, with test case alterations
by me.
Warren Hunt [Sat, 11 Jan 2014 01:16:40 +0000 (01:16 +0000)]
[ms-abi] Change the way alignment is tracked
This patch more cleanly seperates the concepts of Preferred Alignment
and Required Alignment. Most notable that changes to Required Alignment
do *not* impact preferred alignment until late in struct layout. This
is observable when using pragma pack and non-virtual bases and the use
of tail padding when laying them out.
Richard Smith [Sat, 11 Jan 2014 00:53:35 +0000 (00:53 +0000)]
Fix "regression" caused by updating our notion of POD to better match the C++11
rules: instead of requiring flexible array members to be POD, require them to
be trivially-destructible. This seems to be the only constraint that actually
matters here (and even then, it's questionable whether this matters).
Warren Hunt [Fri, 10 Jan 2014 23:32:32 +0000 (23:32 +0000)]
[ms-abi] Adjusting Rules for Padding Between Bases
The presence of a VBPtr suppresses the presence of zero sized
sub-objects in the non-virtual portion of the object in the context of
determining if two base objects need alias-avoidance padding placed
between them.
Jordan Rose [Fri, 10 Jan 2014 20:06:06 +0000 (20:06 +0000)]
[analyzer] Model getters of known-@synthesized Objective-C properties.
...by synthesizing their body to be "return self->_prop;", with an extra
nudge to RetainCountChecker to still treat the value as +0 if we have no
other information.
This doesn't handle weak properties, but that's mostly correct anyway,
since they can go to nil at any time. This also doesn't apply to properties
whose implementations we can't see, since they may not be backed by an
ivar at all. And finally, this doesn't handle properties of C++ class type,
because we can't invoke the copy constructor. (Sema has actually done this
work already, but the AST it synthesizes is one the analyzer doesn't quite
handle -- it has an rvalue DeclRefExpr.)
Modeling setters is likely to be more difficult (since it requires
handling strong/copy), but not impossible.
ObjectiveC. Remove warning on mismatched methods
which may belong to unrelated classes. It was
primarily intended for miuse of @selector expression.
But warning is too noisy and will be issued when
an actual @selector is used. // rdar://15740134
Alp Toker [Fri, 10 Jan 2014 14:37:02 +0000 (14:37 +0000)]
Remove unexpected code completion handling from ConsumeToken()
With this change tok::code_completion is finally handled exclusively as a
special token kind like other tokens that need special treatment.
All callers have been updated to use the specific token consumption methods and
the parser has a clear idea the current token isn't special by the time
ConsumeToken() gets called, so this has been unreachable for some time.
ConsumeAnyToken() behaviour is unchanged and will continue to support
unexpected code completion as part of the special token path.
This survived an amount of fuzzing and validation, but please ping the list if
you hit a code path that previously relied on the old unexpected handler and
now asserts.
Alp Toker [Fri, 10 Jan 2014 11:19:45 +0000 (11:19 +0000)]
Update LangOpt descriptions
Based on recent discussions, attempt to provide a clearer distinction between
MicrosoftMode and MicrosoftExt. This still doesn't feel perfect but gives a
better idea which is which.
Also update the CPlusPlus11 description which got missed in r171367.
reapply r198858: Disable LeakSanitizer in TableGen binaries, see PR18325; this time LeakSanitizerIsTurnedOffForTheCurrentProcess is used instead of __lsan_is_turned_off
Richard Trieu [Fri, 10 Jan 2014 04:38:09 +0000 (04:38 +0000)]
Make the tautological out of range warning use Sema::DiagRuntimeBehavior so that
the warning will not trigger on code protected by compile time checks.
Warren Hunt [Fri, 10 Jan 2014 01:28:05 +0000 (01:28 +0000)]
[ms-abi] Handle __declspec(align) on bitfields "properly"
__declspec(align), when applied to bitfields affects their perferred
alignment instead of their required alignment. We don't know why.
Also, #pragma pack(n) turns packing *off* if n is greater than the
pointer size. This is now observable because of the impact of
declspec(align) on bitfields.
projects / clang / log