Serge Pavlov [Thu, 23 Jan 2014 15:05:00 +0000 (15:05 +0000)]
Fix to PR8880 (clang dies processing a for loop)
Due to statement expressions supported as GCC extension, it is possible
to put 'break' or 'continue' into a loop/switch statement but outside
its body, for example:
for ( ; ({ if (first) { first = 0; continue; } 0; }); )
This code is rejected by GCC if compiled in C mode but is accepted in C++
code. GCC bug 44715 tracks this discrepancy. Clang used code generation
that differs from GCC in both modes: only statement of the third
expression of 'for' behaves as if it was inside loop body.
This change makes code generation more close to GCC, considering 'break'
or 'continue' statement in condition and increment expressions of a
loop as it was inside the loop body. It also adds error for the cases
when 'break'/'continue' appear outside loop due to this syntax. If
code generation differ from GCC, warning is issued.
Tim Northover [Thu, 23 Jan 2014 15:00:01 +0000 (15:00 +0000)]
MachO embedded: default to soft float without issuing warning
This is a simpler rule, broadly in line with previous Darwin (which chose
between "soft" and "softfp") but probably safer. In practice the only real
reason for "softfp" is ABI compatibility, not usually an issue on limited chips
like these, so anyone who wanted hard-float should already be saying so.
Anton Yartsev [Thu, 23 Jan 2014 14:12:48 +0000 (14:12 +0000)]
[analyzer] Strip trailing whitespace characters from input.
More universal way of removing trailing whitespace characters then 'chomp' does. Chomp "removes any trailing string that corresponds to the current value of $/" (quote from perldoc). In my case an input ended with '\r\r\n', chomp left '\r' at the end of input and the script ended up with an error "Use of uninitialized value in concatenation (.) or string"
Jordan Rose [Thu, 23 Jan 2014 03:59:10 +0000 (03:59 +0000)]
[analyzer] Tighten up sanity checks on Objective-C property getter synthesis.
If there are non-trivially-copyable types /other/ than C++ records, we
won't have a synthesized copy expression, but we can't just use a simple
load/return.
Also, add comments and shore up tests, making sure to test in both ARC
and non-ARC.
Justin Bogner [Thu, 23 Jan 2014 02:54:23 +0000 (02:54 +0000)]
test/CodeGenCXX: Give instr-profile exception tests their own file
This test requires asserts for now, and exception handling has an
awkward structure that leads to extra run lines. Because of this, the
test file's not a great place for other C++ PGO tests, but
instr-profile.cpp is obviously the better name for them.
Ben Langmuir [Wed, 22 Jan 2014 23:19:39 +0000 (23:19 +0000)]
Require a module.map file to load a module
Removes some old code that allowed a module to be loaded from a pcm file
even if the module.map could not be found. Also update a number of
tests that relied on the old behavior.
Richard Smith [Wed, 22 Jan 2014 23:07:19 +0000 (23:07 +0000)]
Don't forget about a builtin if we're about to redeclare it and we couldn't
create an implicit declaration of it (because some type it depends on is
unavailable). This had the effect of causing us to not implicitly give it the
right attributes. It turns out that glibc's __sigsetjmp is declared before
sigjmp_buf is declared, and this resulted in us not implicitly giving it
__attribute__((returns_twice)), which in turn resulted in miscompiles in any C
code calling glibc's sigsetjmp.
(See also the vaguely-related sourceware.org/PR4662.)
Aaron Ballman [Wed, 22 Jan 2014 21:51:20 +0000 (21:51 +0000)]
If an attribute has a semantically meaningful spelling (such as ArgumentWithTypeTagAttr or MSInheritanceAttr), display the spelling used for the attribute as part of the AST dump. This should ease debugging the AST for these attributes. Attributes without semantically meaningful spelling variations are not affected.
Mark Seaborn [Wed, 22 Jan 2014 20:11:01 +0000 (20:11 +0000)]
Handle va_arg on struct types for the le32 target (PNaCl and Emscripten)
PNaCl and Emscripten can both handle va_arg IR instructions with
struct type.
Also add a test to cover generating a va_arg IR instruction from
va_arg in C on le32 (as already handled by VisitVAArgExpr() in
CGExprScalar.cpp), which was not covered by a test before.
Richard Smith [Wed, 22 Jan 2014 20:09:10 +0000 (20:09 +0000)]
When a special member is explicitly defaulted outside its class, and we reject
the defaulting because it would delete the member, produce additional notes
explaining why the member is implicitly deleted.
ObjectiveC. When issuing property implementation is
not using backing ivar warning, ignore when
property is not being synthesized (user declared its
implementation @dynamic). // rdar://1583425
Alp Toker [Wed, 22 Jan 2014 07:29:52 +0000 (07:29 +0000)]
Introduce and use Decl::getAsFunction() to simplify templated function checks
Lift the getFunctionDecl() utility out of the parser into a general
Decl::getAsFunction() and use it to simplify other parts of the implementation.
Reduce isFunctionOrFunctionTemplate() to a simple type check that works the
same was as the other is* functions and move unwrapping of shadowed decls to
callers so it doesn't get run twice.
Shuffle around canSkipFunctionBody() to reduce virtual dispatch on ASTConsumer.
There's no need to query when we already know the body can't be skipped.
Ted Kremenek [Wed, 22 Jan 2014 06:10:28 +0000 (06:10 +0000)]
Add basic checking for returning null from functions/methods marked 'returns_nonnull'.
This involved making CheckReturnStackAddr into a static function, which
is now called by a top-level return value checking routine called
CheckReturnValExpr.
Richard Smith [Wed, 22 Jan 2014 01:43:19 +0000 (01:43 +0000)]
Enforce restrictions that 'main' is not allowed to be deleted, or to be used by
the program, in C++. (We allow the latter as an extension, since we've always
permitted it, and GCC does the same, and our supported C++ ABIs don't do
anything special in main.)
Richard Smith [Wed, 22 Jan 2014 00:30:17 +0000 (00:30 +0000)]
Build an appropriate (albeit trivial) TypeSourceInfo for a destructor name, so
AST consumers can determine where the destructor name was written. Patch by
Olivier Goffart!
Richard Smith [Wed, 22 Jan 2014 00:27:42 +0000 (00:27 +0000)]
When formatting a C++-only declaration name, enable C++ mode in the formatter's
language options. This is not really ideal -- we should require the right
language options to be passed in, or not require language options to format a
name -- but it fixes a number of *obviously* wrong formattings. Patch by
Olivier Goffart!
Nico Rieck [Tue, 21 Jan 2014 23:54:36 +0000 (23:54 +0000)]
Delay attribute checking until auto types are deduced
Checking in ActOnVariableDeclarator computes and caches the linkage using
the non-deduced auto type which defaults to external linkage. Depending on
how the auto type is deduced linkage can change and conflict with the
cached linkage, hitting asserts.
Chandler Carruth [Tue, 21 Jan 2014 22:49:05 +0000 (22:49 +0000)]
Teach Clang to look in its installation libdir for libraries (such as
libc++) when the installation is within the system root.
This doesn't really help cross compiles much, but we don't (currently)
have a great story around libc++, cross compiles, and who is responsible
for building and/or installing the libraries. However, it handles the
very common case of non-cross builds in a way entirely consistent with
GCC, so I'm hopeful this won't really hose anyone.
This is the second patch that I think should be backported to 3.4 to
give folks an easy to checkout and install working Clang+libc++
toolchain.
Adrian Prantl [Tue, 21 Jan 2014 18:42:27 +0000 (18:42 +0000)]
Debug info: use the file a typedef is defined in as its decl_file instead
of the current compilation unit.
As a side effect this enables many more LTO uniquing opportunities.
Adrian Prantl [Tue, 21 Jan 2014 18:20:52 +0000 (18:20 +0000)]
Debug info: use the file a typedef is defined in as its decl_file instead
of the current compilation unit.
As a side effect this enables many more LTO uniquing opportunities.
Bob Wilson [Tue, 21 Jan 2014 00:17:10 +0000 (00:17 +0000)]
Stop linking with crt1.o etc. for the iOS simulator. <rdar://problem/11800973>
Recent versions of the iOS simulator no longer require linking with the
crt1.o, dylib1.o, or bundle1.o files. The relevant code is now included in
libSystem for the simulator.
Rafael Espindola [Mon, 20 Jan 2014 20:33:18 +0000 (20:33 +0000)]
Give explicit sections for string constants used in NSStrings.
Without them they can be merged with non unnamed_addr constants during LTO.
The resulting constant is not unnamed_addr and goes in a different section,
which causes ld64 to crash.
Alp Toker [Mon, 20 Jan 2014 20:26:09 +0000 (20:26 +0000)]
Rename FunctionProtoType accessors from 'arguments' to 'parameters'
Fix a perennial source of confusion in the clang type system: Declarations and
function prototypes have parameters to which arguments are supplied, so calling
these 'arguments' was a stretch even in C mode, let alone C++ where default
arguments, templates and overloading make the distinction important to get
right.
Readability win across the board, especially in the casting, ADL and
overloading implementations which make a lot more sense at a glance now.
Will keep an eye on the builders and update dependent projects shortly.
Aaron Ballman [Mon, 20 Jan 2014 18:07:09 +0000 (18:07 +0000)]
Exposed a declarative way to specify that an attribute can be duplicated when merging attributes on a declaration. This replaces some hard-coded functionality from Sema.
Aaron Ballman [Mon, 20 Jan 2014 17:18:35 +0000 (17:18 +0000)]
Remove some hard-coded specialness for thread-safety attributes from the parser, and made it more declarative. If an attribute is allowed to appear on a function definition when late parsed, it can now use the FunctionDefinition attribute subject. It's treated as a FunctionDecl for most purposes, except it also gets exposed on the AttributeList so that it can be used while parsing.
Aaron Ballman [Mon, 20 Jan 2014 14:18:40 +0000 (14:18 +0000)]
HasFunctionProto is a more strict version of FunctionLike. Since attribute subjects are inclusive (passing a single subject test means no subject-related diagnostic will fire), these two subjects should not be combined.
Chandler Carruth [Mon, 20 Jan 2014 09:42:24 +0000 (09:42 +0000)]
Make the Linux support for finding libc++ somewhat less braindead.
Now instead of just looking in the system root for it, we also look
relative to the clang binary's directory. This should "just work" in
almost all cases. I've added test cases accordingly.
This is probably *very* worthwhile to backport to the 3.4 branch so that
folks can check it out, build it, and use that as their host compiler
going forward.
Ted Kremenek [Mon, 20 Jan 2014 05:50:47 +0000 (05:50 +0000)]
Wire up basic parser/sema support for attribute 'returns_nonnull'.
This attribute is supported by GCC. More generally it should
probably be a type attribute, but this behavior matches 'nonnull'.
This patch does not include warning logic for checking if a null
value is returned from a function annotated with this attribute.
That will come in subsequent patches.
Alp Toker [Mon, 20 Jan 2014 00:23:47 +0000 (00:23 +0000)]
Fix nothrow trait with multiple default constructors
Check all default ctors, not just the first one we see. This brings
__has_nothrow_constructor() in line with the other unary type traits.
A C++ class can have multiple default constructors but clang was only checking
the first one written, presumably due to ambiguity in the GNU specification.
MSVC has the same bug, while g++ has the correct implementation which we now
match.
Alp Toker [Sat, 18 Jan 2014 21:49:02 +0000 (21:49 +0000)]
docs: clarify the role of compiler-builtin type trait primitives
Update the documentation to clarify the intent of clang's built-in type trait
facilities, their relation to user-facing C++ type traits and means to check
for availability.
Also explain that __has_feature() is not currently up to date and should not
generally be used in user code (there's a proposal to provide more consistent
checks via __has_builtin(), see cfe-dev).