Reid Kleckner [Wed, 17 Dec 2014 19:34:15 +0000 (19:34 +0000)]
Don't build invalid AST nodes during recovery
A DependentScopeDeclRefExpr should always have a nested name specifier.
During template instantiation, if we found that the named context was
incomplete, we would previously build a DependentScopeDeclRefExpr with
an empty qualifier.
This error recovery path has been asserting for some time. The other
error codepaths use ExprError, so we can do the same.
For MSVC compatibility, add the `__emit' builtin. This is used in the Windows
SDK headers, and must therefore be implemented as a builtin rather than an
intrinsic.
The `__emit' builtin provides a mechanism to emit a 16-bit opcode instruction
into the stream. The value must be a compile time constant expression. No
guarantees are made about the CPU and memory states after the execution of the
instruction.
Due to the unchecked nature of the builtin, only support this on Windows on ARM.
Toma Tabacu [Wed, 17 Dec 2014 12:02:58 +0000 (12:02 +0000)]
[mips] Always clobber $1 for MIPS inline asm.
Summary:
Because GCC doesn't use $1 for code generation, inline assembly code can use $1 without having to add it to the clobbers list.
LLVM, on the other hand, does not shy away from using $1, and this can cause conflicts with inline assembly which assumes GCC-like code generation.
A solution to this problem is to make Clang automatically clobber $1 for all MIPS inline assembly.
This is not the optimal solution, but it seems like a necessary compromise, for now.
Anna Zaks [Wed, 17 Dec 2014 00:34:07 +0000 (00:34 +0000)]
[CallGraph] Make sure the edges are not missed due to re-declarations
A patch by Daniel DeFreez!
We were previously dropping edges on re-declarations. Store the
canonical declarations in the graph to ensure that different
references to the same function end up reflected with the same call graph
node.
(Note, this might lead to performance fluctuation because call graph
is used to determine the function analysis order.)
David Blaikie [Tue, 16 Dec 2014 22:49:17 +0000 (22:49 +0000)]
DebugInfo: Generalize debug info location handling
This is a more scalable (fixed in mostly one place, rather than many
places that will need constant improvement/maintenance) solution to
several commits I've made recently to increase source fidelity for
subexpressions.
This resetting had to be done at the DebugLoc level (not the
SourceLocation level) to preserve scoping information (if the resetting
was done with CGDebugInfo::EmitLocation, it would've caused the tail end
of an expression's codegen to end up in a potentially different scope
than the start, even though it was at the same source location). The
drawback to this is that it might leave CGDebugInfo out of sync. Ideally
CGDebugInfo shouldn't have a duplicate sense of the current
SourceLocation, but for now it seems it does... - I don't think I'm
going to tackle removing that just now.
I expect this'll probably cause some more buildbot fallout & I'll
investigate that as it comes up.
Also these sort of improvements might be starting to show a weakness/bug
in LLVM's line table handling: we don't correctly emit is_stmt for
statements, we just put it on every line table entry. This means one
statement split over multiple lines appears as multiple 'statements' and
two statements on one line (without column info) are treated as one
statement.
I don't think we have any IR representation of statements that would
help us distinguish these cases and identify the beginning of each
statement - so that might be something we need to add (possibly to the
lexical scope chain - a scope for each statement). This does cause some
problems for GDB and possibly other DWARF consumers.
Nick Lewycky [Tue, 16 Dec 2014 22:02:06 +0000 (22:02 +0000)]
Look at whether TransformTypos returned a different Expr instead of looking at the number of uncorrected typos before and after. Correcting one typo may produce an expression with another TypoExpr in it, leading to matching counts even though a typo was corrected.
Nick Lewycky [Tue, 16 Dec 2014 21:39:02 +0000 (21:39 +0000)]
Add a new flag, -fspell-checking-limit=<number> to control how many times we'll do spell checking. Note that spell checking will change the produced AST, so we don't automatically change this value when someone sets -ferror-limit=. With this, merge test typo-correction-pt2.cpp into typo-correction.cpp.
Remove Sema::UnqualifiedTyposCorrected, a cache of corrected typos. It would only cache typo corrections that didn't provide ValidateCandidate of which there were few left, and it had a bug when we had the same identifier spelled wrong twice. See the last two tests in typo-correction.cpp for cases this fires.
Nico Weber [Tue, 16 Dec 2014 21:16:10 +0000 (21:16 +0000)]
Move -Wkeyword-macro into -pedantic, remove -Wreserved-id-macro.
As discussed on the post-commit review thread for r224012, -Wkeyword-macro fires
mostly on headers trying to set up portable defines and doesn't find much bad
stuff in practice. But [macro.names]p2 does disallow defining or undefining
keywords, override and final, and alignas, so keep the warning but move it
into -pedantic.
-Wreserved-id-macro warns on
#define __need_size_t
which is more or less public api for glibc headers. Since this warning isn't
motivated by a standard, remove it.
(See also r223114 for a previous follow-up to r224012.)
Rafael Espindola [Tue, 16 Dec 2014 21:00:30 +0000 (21:00 +0000)]
Put static local variables of inline functions in the function comdat.
The variable (and the GV) is only ever used if the function is. Putting it
in the function's comdat make it easier for the linker to discard them.
The motivating example is
struct S {
static const int x;
};
// const int S::x = 42;
inline const int *f() {
static const int y = S::x;
return &y;
}
const int *g() { return f(); }
With S::x commented out, _ZZ1fvE1y is a variable with a guard variable
that is initialized by f.
Above the implicit copy assignment operator was inferred as host device but
there was only a host assignment copy defined which is an error in device
compilation mode.
Alexey Bataev [Tue, 16 Dec 2014 08:01:48 +0000 (08:01 +0000)]
Renamed RefersToEnclosingLocal bitfield to RefersToCapturedVariable.
Bitfield RefersToEnclosingLocal of Stmt::DeclRefExprBitfields renamed to RefersToCapturedVariable to reflect latest changes introduced in commit 224323. Also renamed method Expr::refersToEnclosingLocal() to Expr::refersToCapturedVariable() and comments for constant arguments.
No functional changes.
Alexey Bataev [Tue, 16 Dec 2014 07:00:22 +0000 (07:00 +0000)]
[OPENMP] Bugfix for processing of global variables in OpenMP regions.
Currently, if global variable is marked as a private OpenMP variable, the compiler crashes in debug version or generates incorrect code in release version. It happens because in the OpenMP region the original global variable is used instead of the generated private copy. It happens because currently globals variables are not captured in the OpenMP region.
This patch adds capturing of global variables iff private copy of the global variable must be used in the OpenMP region.
Differential Revision: http://reviews.llvm.org/D6259
David Majnemer [Tue, 16 Dec 2014 06:31:17 +0000 (06:31 +0000)]
Sema: Don't crash converting to bool from _Atomic
Turning our _Atomic L-value into an R-value removes its _Atomic-ness.
However, we didn't update our 'FromType' which made
ScalarTypeToBooleanCastKind think we were trying to pass it a
non-scalar.
David Majnemer [Tue, 16 Dec 2014 04:52:14 +0000 (04:52 +0000)]
AST: Fix the linkage of static vars in fn template specializations
We that static variables in function template specializations were
externally visible. The manglers assumed that externally visible static
variables were numbered in Sema. We would end up mangling static
variables in the same specialization with the same mangling number which
would give all of them the same name.
Hans Wennborg [Tue, 16 Dec 2014 01:15:01 +0000 (01:15 +0000)]
Clarify the code in checkDLLAttribute()
Update the comments to make it more clear what's going on, and address
Richard's comments from PR21718. This doesn't fix that bug, but hopefully
makes the code easier to understand.
Reid Kleckner [Mon, 15 Dec 2014 23:16:32 +0000 (23:16 +0000)]
Diagnose function template definitions inside functions
The parser can only be tricked into parsing a function template
definition by inserting a typename keyword before the function template
declaration. This used to make us crash, and now it's fixed.
While here, remove an unneeded boolean parameter from ParseDeclGroup.
This boolean always corresponded to non-typedef declarators at file
scope. ParseDeclGroup already has precise diagnostics for the function
definition typedef case, so we can let that through.
David Majnemer [Mon, 15 Dec 2014 10:00:35 +0000 (10:00 +0000)]
Sema: Don't diagnose string + int if the int is value dependent
Don't send a value dependent expression into the expression evaluator,
HandleSizeof would crash. Making HandleSizeof handle dependent types
would noisily warn about the operation even if everything turns out OK
after instantiation.
David Majnemer [Mon, 15 Dec 2014 09:03:58 +0000 (09:03 +0000)]
Preprocessor: Recover instead of mutating a token in ExpandBuiltinMacro
We would CreateString on arbitrary garbage instead of just skipping to
the end of the builtin macro. Eventually, this would cause us to crash
because we would end up replacing the contents of a character token with
a numeric literal.
Alexander Musman [Mon, 15 Dec 2014 07:07:06 +0000 (07:07 +0000)]
First patch with codegen of the 'omp for' directive. It implements
the simplest case, which is used when no chunk_size is specified in
the schedule(static) or no 'schedule' clause is specified - the
iteration space is divided by the library into chunks that are
approximately equal in size, and at most one chunk is distributed
to each thread. In this case, we do not need an outer loop in each
thread - each thread requests once which iterations range it should
handle (using __kmpc_for_static_init runtime call) and then runs the
inner loop on this range.
Alexey Bataev [Mon, 15 Dec 2014 05:25:25 +0000 (05:25 +0000)]
Bugfix for Codegen of atomic load/store/other ops.
Currently clang fires assertions on x86-64 on any atomic operations for long double operands. Patch fixes codegen for such operations.
Differential Revision: http://reviews.llvm.org/D6499
Alexey Bataev [Mon, 15 Dec 2014 04:18:11 +0000 (04:18 +0000)]
MSVC: A wide string literal from L#macro_arg in a macro
Clang should form a wide string literal from L#macro_arg in a function-like macro in -fms-compatibility mode.
Fix for http://llvm.org/PR9984.
Differential Revision: http://reviews.llvm.org/D6604
David Blaikie [Sun, 14 Dec 2014 18:48:18 +0000 (18:48 +0000)]
DebugInfo: More accurate line information for placement new.
This actually came up as a break in UBSan tests (look for a follow-up
commit to this one to see the UBSan test fallout) when I tried a broader
fix to location information.
I have some other ideas about how to do that broader change & will keep
looking into it.
David Majnemer [Sun, 14 Dec 2014 12:16:43 +0000 (12:16 +0000)]
CodeGen: Compound literals with funny types shouldn't crash
CodeGen assumed that a compound literal with array type should have a
corresponding LLVM IR array type.
We had two bugs in this area:
- Zero sized arrays in compound literals would lead to the creation of
an opaque type. This is unnecessary, we should just create an array
type with a bound of zero.
- Funny record types (like unions) lead to exotic IR types for compound
literals. In this case, CodeGen must be prepared to deal with the
possibility that it might not have an array IR type.
David Majnemer [Sun, 14 Dec 2014 04:53:11 +0000 (04:53 +0000)]
Lex: Don't crash if both conflict markers are on the same line
We would check if the terminator marker is on a newline. However, the
logic would end up out-of-bounds if the terminator marker immediately
follows the start marker.
John McCall [Sun, 14 Dec 2014 01:46:53 +0000 (01:46 +0000)]
If a non-template constructor instantiated to X(X),
ignore it during overload resolution when initializing
X from a value of type cv X.
Previously, our rule here only ignored specializations
of constructor templates. That's probably because the
standard says that constructors are outright ill-formed
if their first parameter is literally X and they're
callable with one argument. However, Clang only
enforces that prohibition against non-implicit
instantiations; I'm not sure why, but it seems to be
deliberate. Given that, the most sensible thing to
do is to just ignore the "illegal" constructor
regardless of where it came from.
Also, stop ignoring such constructors silently:
print a note explaining why they're being ignored.
David Majnemer [Sun, 14 Dec 2014 01:05:01 +0000 (01:05 +0000)]
Sema: attribute((annotate)) must have at least one argument
Sema::handleAnnotateAttr expects that some basic validation is done on
the given AttributeList. However, ProcessAccessDeclAttributeList called
it directly. Instead, pass the list to ProcessDeclAttribute.
David Majnemer [Sat, 13 Dec 2014 08:12:56 +0000 (08:12 +0000)]
Sema: Constexpr functions must have return statements which have an expr
clang lets programmers be pretty cavalier when it comes to void return
statements in functions which have non-void return types. However, we
cannot be so forgiving in constexpr functions: evaluation will go off
the rails very quickly.
Instead, keep the return statement in the AST but mark the function as
invalid. Doing so gives us nice diagnostics while making constexpr
evaluation halt.
Nick Lewycky [Sat, 13 Dec 2014 02:54:28 +0000 (02:54 +0000)]
Fix two small bugs in typo correction. One assertion failure building member expressions because the lookup finds a different name than the original, fixed by updating the LookupResult's name with the name of the found decl. Second is that we also diagnose delayed typo exprs in the index of an array subscript expression.
The testcase shows a third bug with a FIXME in it.
Implement the __builtin_call_with_static_chain GNU extension.
The extension has the following syntax:
__builtin_call_with_static_chain(Call, Chain)
where Call must be a function call expression and Chain must be of pointer type
This extension performs a function call Call with a static chain pointer
Chain passed to the callee in a designated register. This is useful for
calling foreign language functions whose ABI uses static chain pointers
(e.g. to implement closures).
Justin Bogner [Fri, 12 Dec 2014 23:12:27 +0000 (23:12 +0000)]
Frontend: Canonicalize to native paths when dumping module dependencies
Mixed path separators (ie, both / and \\) can mess up the sort order
of the VFS map when dumping module dependencies, as was recently
exposed by r224055 and papered over in r224145. Instead, we should
simply use native paths for consistency.
This also adds a TODO to add handling of .. in paths. There was some
code for this before r224055, but it was untested and probably broken.
Fix the issue of mangling of local anonymous unions (Itanium C++ ABI):
A discriminator is used for the first occurrence of a name.
inline int f1 () {
static union {
int a;
long int b;
};
static union {
int c;
double d;
};
return a+c;
}
The name of the second union is mangled as _ZZ2f1vE1c_0 instead of _ZZ2f1vE1c.
Don't break string literals in Java and JavaScript.
The proper way to break string literals in these languages is by inserting a "+"
between parts which we don't support yet. So we disable string literal breaking
until then.
[Objective-C]. This patch extends objc_bridge attribute to support objc_bridge(id).
This means that a pointer to the struct type to which the attribute appertains
is a CF type (and therefore an Objective-C object of some type), but not of any
specific class. rdar://19157264
Richard Smith [Thu, 11 Dec 2014 20:50:24 +0000 (20:50 +0000)]
[modules] When constructing paths relative to a module, strip out /./ directory
components. These sometimes get synthetically added, and we don't want -Ifoo
and -I./foo to be treated fundamentally differently here.
Paul Robinson [Thu, 11 Dec 2014 20:14:04 +0000 (20:14 +0000)]
Instead of having -Os/-Oz add OptimizeForSize/MinSize first, and later
having OptimizeNone remove them again, just don't add them in the
first place if the function already has OptimizeNone.
Note that MinSize can still appear due to attributes on different
declarations; a future patch will address that.
David Majnemer [Thu, 11 Dec 2014 19:36:24 +0000 (19:36 +0000)]
AST: Incomplete types might be zero sized
Comparing the address of an object with an incomplete type might return
true with a 'distinct' object if the former has a size of zero.
However, such an object should compare unequal with null.
Aaron Ballman [Thu, 11 Dec 2014 19:35:42 +0000 (19:35 +0000)]
When checking for nonnull parameter attributes, also check the ParmVarDecl since the attribute may reside there, instead of just on the FunctionDecl. Fixes PR21668.
David Majnemer [Thu, 11 Dec 2014 01:00:48 +0000 (01:00 +0000)]
Parse: Concatenated string literals should be verified in inline asm
While we would correctly handle asm("foo") and reject asm(L"bar"), we
weren't careful to handle cases where an ascii literal could be
concatenated with a wide literal.
Eric Christopher [Wed, 10 Dec 2014 22:58:32 +0000 (22:58 +0000)]
Revert the default changing behavior part of r216662 until we
can change the backend to be the same default. Leave the
modified/new testcases with the exception of the default behavior
since it increases our testing footprint.
David Majnemer [Wed, 10 Dec 2014 22:58:14 +0000 (22:58 +0000)]
AST: Properly calculate the linkage of a IndirectFieldDecl
getLVForNamespaceScopeDecl believed that it wasn't possible for it to
ever see an IndirectFieldDecl. However, this can occur when determining
whether or not something is a redeclaration of a member of an anonymous
static union.