Eli Friedman [Thu, 19 Sep 2013 00:41:32 +0000 (00:41 +0000)]
Make Preprocessor::Lex non-recursive.
Before this patch, Lex() would recurse whenever the current lexer changed (e.g.
upon entry into a macro). This patch turns the recursion into a loop: the
various lex routines now don't return a token when the current lexer changes,
and at the top level Preprocessor::Lex() now loops until it finds a token.
Normally, the recursion wouldn't end up being very deep, but the recursion depth
can explode in edge cases like a bunch of consecutive macros which expand to
nothing (like in the testcase test/Preprocessor/macro_expand_empty.c in this
patch).
Hans Wennborg [Wed, 18 Sep 2013 22:26:39 +0000 (22:26 +0000)]
clang-cl: Don't warn about overriding /MD with /MT, /Fo with another /Fo, etc.
I put in the warnings because MSVC has them, but I don't think they're very
useful.
Clang does not warn about overriding flags in general, e.g. it's perfectly
fine to have -fomit-frame-pointer followed by -fno-omit-frame-pointer.
We should focus on warning where things get confusing, such as with the
/TP and /TC options. In "clang-cl /TC a.c /TP b.cc", the user might not
realize that the /TP flag will apply to both files, and we warn about that.
Adrian Prantl [Wed, 18 Sep 2013 22:08:57 +0000 (22:08 +0000)]
Debug info: Explicitly set the DIVariable::isIndirect() flag for VLAs.
This allows us to get rid of an ugly hack in the backend.
Paired commit with LLVM.
ObjectiveC migrator: For consistency, also infer
'instancetype' for known family of methods
with related result type; such as 'init'
methods. // rdar://14987948
Jordan Rose [Wed, 18 Sep 2013 18:58:58 +0000 (18:58 +0000)]
[analyzer] Don't even try to convert floats to booleans for now.
We now have symbols with floating-point type to make sure that
(double)x == (double)x comes out true, but we still can't do much with
these. For now, don't even bother trying to create a floating-point zero
value; just give up on conversion to bool.
PR16752: 'mode' attribute for unusual targets doesn't work properly
Description:
Troubles could be happened due to some assumptions in handleModeAttr function (see SemaDeclAttr.cpp).
For example, it assumes that 32 bit integer is 'int', while it could be 16 bit only.
Instead of asking target: 'which type do you want to use for int32_t ?' it just hardcodes general opinion. That doesn't looks pretty correct.
Please consider the next solution:
1. In Basic/TargetInfo add getIntTypeByWidth and getRealTypeByWidth methods. Methods asks target for proper type for given bit width.
2. Fix handleModeAttr according to new methods in TargetInfo.
Fixes:
1st Commit (Done): Add new methods for TargetInfo:
getRealTypeByWidth and getIntTypeByWidth
for ASTContext names are almost same(invokes new methods from TargetInfo):
getIntTypeForBitwidth and getRealTypeForBitwidth
Hal Finkel [Wed, 18 Sep 2013 06:19:40 +0000 (06:19 +0000)]
Add new libm LIBBUILTIN definitions, synchronizing with __builtin_*
For all libm __builtin_* functions that are defined, this adds the
corresponding LIBBUILTIN definitions (tagged, as necessary, with "e" instead of
"c" when the function may set errno).
Note that this changes the current definitions for lrint and fma
(unfortunately). The Linux man page documents that these don't set errno, but
the POSIX standard says that they should.
Hal Finkel [Wed, 18 Sep 2013 03:29:45 +0000 (03:29 +0000)]
Add the intrinsic __builtin_convertvector
LLVM supports applying conversion instructions to vectors of the same number of
elements (fptrunc, fptosi, etc.) but there had been no way for a Clang user to
cause such instructions to be generated when using builtin vector types.
C-style casting on vectors is already defined in terms of bitcasts, and so
cannot be used for these conversions as well (without leading to a very
confusing set of semantics). As a result, this adds a __builtin_convertvector
intrinsic (patterned after the OpenCL __builtin_astype intrinsic). This is
intended to aid the creation of vector intrinsic headers that create generic IR
instead of target-dependent intrinsics (in other words, this is a generic
_mm_cvtepi32_ps). As noted in the documentation, the action of
__builtin_convertvector is defined in terms of the action of a C-style cast on
each vector element.
This will define _MSC_VER to 1700 by default and avoid linker errors
from /failifmismatch linker directives in the C++ standard headers.
Most people trying out the Visual Studio integration are using 2012,
since that's the only version that clang-format works with. This way
they don't have to pass funky -Xclang -fmsc-version=1700 flags just to
link against the standard C++ runtime.
Hans Wennborg [Tue, 17 Sep 2013 23:27:39 +0000 (23:27 +0000)]
clang-cl: ignore ths /GS- flag
The /GS- flag is used to turn off run-time buffer security checks (/GS).
Since no such checks are enabled in the first place, I think we should just
ignore this flag.
David Majnemer [Tue, 17 Sep 2013 22:21:27 +0000 (22:21 +0000)]
[-cxx-abi microsoft] Mangle local TagDecls appropriately
Summary:
When selecting a mangling for an anonymous tag type:
- We should first try it's typedef'd name.
- If that doesn't work, we should mangle in the name of the declarator
that specified it as a declaration specifier.
- If that doesn't work, fall back to a static mangling of
<unnamed-type>.
This should make our anonymous type mangling compatible.
This partially fixes PR16994; we would need to have an implementation of
scope numbering to get it right (a separate issue).
Push contents of X86TargetInfo::setFeatureEnabled down to a static function called by the virtual version and all the places in getDefaultFeatures. This way getDefaultFeatures doesn't make so many virtual calls.
Anna Zaks [Tue, 17 Sep 2013 01:30:57 +0000 (01:30 +0000)]
[analyzer] More reliably detect property accessors.
This has a side effect of preventing a crash, which occurs because we get a
property getter declaration, which is overriding but is declared inside
@protocol. Will file a bug about this inconsistency internally. Getting a
small test case is very challenging.
Hans Wennborg [Tue, 17 Sep 2013 00:25:08 +0000 (00:25 +0000)]
Fix test/Driver/cl-options.c test
The test builds an object file to be able to get into linking mode
with a valid obj file on the command-line. Using clang-cl for this,
which targets win32, caused problems on some buildbots, so just
use regular clang.
Hans Wennborg [Tue, 17 Sep 2013 00:03:41 +0000 (00:03 +0000)]
clang-cl: ignore compile-only options in link-only invocations.
Previously we would warn about unused arguments such as /MD when linking.
Clang already has logic to ignore compile-only options, e.g. for -D and -U.
This patch extends that to include clang-cl's compile-only options too.
Also, some clang-cl options should always be ignored. Doing this earlier
means they get ignored both for compilation and link-only invocations.
Summary:
This fixes several issues with the original implementation:
- Win32 entry points cannot be in namespaces
- A Win32 entry point cannot be a function template, diagnose if we it.
- Win32 entry points cannot be overloaded.
- Win32 entry points implicitly return, similar to main.
Richard Trieu [Mon, 16 Sep 2013 21:54:53 +0000 (21:54 +0000)]
Move the uninitialized field check to after all the field initializers are added
to the CXXConstructorDecl so that information from the constructor can be used.
Anton Yartsev [Mon, 16 Sep 2013 17:51:25 +0000 (17:51 +0000)]
New message for cases when ownership is taken:
"+method_name: cannot take ownership of memory allocated by 'new'."
instead of the old
"Memory allocated by 'new' should be deallocated by 'delete', not +method_name"
Ed Schouten [Sat, 14 Sep 2013 16:17:20 +0000 (16:17 +0000)]
Make __has_extension(c_thread_local) work.
Unlike C++11's "thread_local" keyword, C11's "_Thread_local" is in the
reserved namespace, meaning we provide it unconditionally; it is marked
as KEYALL in TokenKinds.def.
This means that like all the other C11 keywords, we can expose its
presence through __has_extension().
Avoid getting an argument of allocation function if it does not exist.
This is a fix to PR12778: in erroneous code an allocation function
can be declared with no arguments, quering the first argument in this case
causes assertion violation.
Summary:
This fixes PR17145 and avoids unknown pragma warnings.
This change does not attempt to map MSVC warning numbers to clang
warning flags. Perhaps in the future we will implement a mapping for
some common subset of Microsoft warnings, but for now we don't.
David Blaikie [Fri, 13 Sep 2013 18:45:00 +0000 (18:45 +0000)]
DebugInfo: omit debug info for friends
GCC ToT doesn't do this & it's worth about 3.2% on Clang's DWO file size
with Clang. Some or all of this may be due to things like r190715 which
could have source fixes/improvements, but it's not clear that's the case
and that doesn't help other source bases.
David Blaikie [Fri, 13 Sep 2013 18:32:52 +0000 (18:32 +0000)]
Remove unnecessary inclusion of Sema.h
Let me tell you a tale...
Within some twisted maze of debug info I've ended up implementing an
insane man's Include What You Use device. When the debugger emits debug
info it really shouldn't, I find out why & then realize the code could
be improved too.
In this instance CIndexDiagnostics.cpp had a lot more debug info with
Clang than GCC. Upon inspection a major culprit was all the debug info
describing clang::Sema. This was emitted because clang::Sema is
befriended by DiagnosticEngine which was rightly required, but GCC
doesn't emit debug info for friends so it never emitted anything for
Clang. Clang does emit debug info for friends (will be fixed/changed to
reduce debug info size).
But why didn't Clang just emit a declaration of Sema if this entire TU
didn't require a definition?
1) Diagnostic.h did the right thing, only using a declaration of Sema
and not including Sema.h at all.
2) Some other dependency of CIndexDiagnostics.cpp didn't do the right
thing. ASTUnit.h, only needing a declaration, still included Sema.h
(hence this commit which removes that include and adds the necessary
includes to the cpp files that were relying on this)
3) -flimit-debug-info didn't save us because of
EnterExpressionEvaluationContext, defined inline in Sema.h which fires
the "requiresCompleteType" check/flag (since it uses nested types from
Sema and calls Sema member functions) and thus, if debug info is ever
emitted for the type, the whole type is emitted and not just a
declaration.
Improving -flimit-debug-info to account for this would be... hard.
Modifying the code so that's not 'required to be complete' might be
possible, but probably only by moving EnterExpressionEvaluationContext
either into Sema, or out of Sema.h. That might be a bit too much of a
contortion to be bothered with.
Also, this is only one of the cases where emitting debug info for
friends caused us to emit a lot more debug info (this change reduces
Clang's DWO size by 0.93%, dropping friends entirely reduces debug info
by 3.2%) - I haven't hunted down the other cases, but I assume they
might be similar (Sema or something like it). IWYU or a similar tool
might help us reduce build times a bit, but analyzing debug info to find
these differences isn't worthwhile. I'll take the 3.2% win, provide this
small improvement to the code itself, and move on.
projects / clang / log