Chandler Carruth [Wed, 10 Oct 2012 11:29:08 +0000 (11:29 +0000)]
Apply the same fundamental fix for PR14048 as was applied for PR11905.
The issue arises when coercing to/from types of different sizes. We need
to be certain that the allocation on either end has sufficient room for
the coerced type. When it doesn't, we need to make room, copy across,
and then proceed. PR11905 handled the case of storing function arguments
back into allocas in the function prolog, this patch handles the case of
setting up the function arguments in a call expression.
This is actually significantly simpler than the fix for PR11905. It ends
up being a trivial change to create a temporary alloca when the source
is too small and memcpy across. This should preserve the compile-time
fast-isel benefits of doing gep+load sequences and avoiding FCAs.
Reviewed by Benjamin and Evgeniy (who fixed PR11905).
Bill Wendling [Wed, 10 Oct 2012 07:36:56 +0000 (07:36 +0000)]
Remove the final bits of Attributes being declared in the Attribute
namespace. Use the attribute's enum value instead. No functionality change
intended.
When indexing a module file, for the ppIncludedFile callback give
an invalid location if the location points to the synthetic buffer
for the module input.
Douglas Gregor [Tue, 9 Oct 2012 23:05:51 +0000 (23:05 +0000)]
Rework the (de-)serialization of macros, as stored in
MacroInfo*. Instead of simply dumping an offset into the current file,
give each macro definition a proper ID with all of the standard
modules-remapping facilities. Additionally, when a macro is modified
in a subsequent AST file (e.g., #undef'ing a macro loaded from another
module or from a precompiled header), provide a macro update record
rather than rewriting the entire macro definition. This gives us
greater consistency with the way we handle declarations, and ties
together macro definitions much more cleanly.
Note that we're still not actually deserializing macro history (we
never were), but it's far easy to do properly now.
The clang driver has a fairly fancy support for executing gcc instead of
clang itself. This dates back to clang's early days and while it looks like
some of it is still used (for kext for example), other parts are probably dead.
Remove the -ccc-clang-archs option and associated code. I don't think there
is any remaining setup where clang doesn't support an architecture but it can
expect an working gcc cross compiler to be available.
A nice side effect is that tests no longer need to differentiate architectures
that are included in production builds of clang and those that are not.
Move the functionality that looks for ObjC overridden methods from
ASTContext to the ObjCMethodDecl, and have the more generic
ASTContext::getOverriddenMethods() use the ObjCMethodDecl::getOverriddenMethods()
function.
Douglas Gregor [Tue, 9 Oct 2012 17:21:28 +0000 (17:21 +0000)]
When we load a function or method body from an AST file, we check
whether that function/method already has a body (loaded from some
other AST file), as introduced in r165137. Delay this check until
after the redeclaration chains have been wired up.
While I'm here, make the loading of method bodies lazy.
Douglas Gregor [Tue, 9 Oct 2012 16:01:50 +0000 (16:01 +0000)]
If a macro has been #undef'd in a precompiled header, we still need to
write out the macro history for that macro. Similarly, we need to cope
with reading a macro definition that has been #undef'd.
Take advantage of this new ability so that global code-completion
results can refer to #undef'd macros, rather than losing them
entirely. For multiply defined/#undef'd macros, we will still get the
wrong result, but it's better than getting no result.
Bill Wendling [Tue, 9 Oct 2012 07:45:26 +0000 (07:45 +0000)]
We use the enums to query whether an Attributes object has that attribute. The
opaque layer is responsible for knowing where that specific attribute is stored.
In VarDecl::getSourceRange() make sure to check that the source location
of the initializer is valid before using it.
Fixes rdar://12455002&12449015 where local variables of objc objects in ARC mode
were not annotated because of the ImplicitValueInitExpr initializer having invalid
source range, resulting in the SourceRange of the VarDecl having invalid end location.
Make Bitrig's clang understand -stdlib= correctly.
With this patch Bitrig can use a different c++ library without pain and
within the normal commandline parameters.
Original patch by David Hill, with lots of fixes and cleanup by me.
Benjamin Kramer [Sat, 6 Oct 2012 14:42:22 +0000 (14:42 +0000)]
Expose __builtin_bswap16.
GCC has always supported this on PowerPC and 4.8 supports it on all platforms,
so it's a good idea to expose it in clang too. LLVM supports this on all targets.
Jordan Rose [Sat, 6 Oct 2012 01:19:36 +0000 (01:19 +0000)]
ParentMap: Restore the ability to update an existing map.
The Clang ASTs are a DAG, not a pure tree. However, ParentMap has to
choose a single parent for each object. In the main (only?) cases in
which the AST forms a DAG, it protects from multiple traversal by using
OpaqueValueExprs. Previously, ParentMap would just unconditionally look
through all OpaqueValueExprs when building its map.
In order to make this behavior better for the analyzer's diagnostics,
ParentMap was changed to not set a statement's parent if there already
was one in the map. However, ParentMap is supposed to allow updating
existing mappings by calling addStmt once again. This change makes the
"transparency" of OpaqueValueExprs explicit, and disables it when it
is not desired, rather than checking the current contents of the map.
This new code seems like a big change, but it should actually have
essentially the same performance as before. Only OpaqueValueExprs and
their users (PseudoObjectExpr and BinaryConditionalOperator) will
have any different behavior.
There should be no user-visible functionality change, though a test
has been added for the current behavior of BinaryConditionalOperator
source locations and accompanying Xcode arrows (which are not so great...).
Jordan Rose [Sat, 6 Oct 2012 01:19:30 +0000 (01:19 +0000)]
[analyzer] Handle implicit statements used for end-of-path nodes' source locs.
Some implicit statements, such as the implicit 'self' inserted for "free"
Objective-C ivar access, have invalid source locations. If one of these
statements is the location where an issue is reported, we'll now look at
the enclosing statements for a valid source location.
[libclang] Don't disable the preprocessing record while indexing so as
to not mess up with module building.
It was not worth trying to combine indexing without preprocessing record
and building modules with one because:
-just importing a module/PCH that was built with a pp record, enables it anyway
-the performance gain of indexing without the preprocessing record is insignificant.
David Blaikie [Fri, 5 Oct 2012 00:41:03 +0000 (00:41 +0000)]
Implement -Wshift-op-parentheses for: a << b + c
This appears to be consistent with GCC's implementation of the same warning
under -Wparentheses. Suppressing a << b + c for cases where 'a' is a user
defined type for compatibility with C++ stream IO. Otherwise suggest
parentheses around the addition or subtraction subexpression.
(this came up when MSVC was complaining (incorrectly, so far as I can tell)
about a perceived violation of this within the LLVM codebase, PR14001)
[preprocessing record] Have PPEntityID be independent of the size of the
loaded entities vector, otherwise its meaning will change when a module
is imported and the vector size changes.
Richard Smith [Thu, 4 Oct 2012 23:52:29 +0000 (23:52 +0000)]
If we flow off the end of a value-returning function:
- outside C++, return undef (behavior is not undefined unless the value is used)
- in C++, with -fcatch-undefined-behavior, perform an appropriate trap
- in C++, produce an 'unreachable' (behavior is undefined immediately)
Richard Smith [Thu, 4 Oct 2012 22:13:39 +0000 (22:13 +0000)]
Egriegious hack to support libstdc++4.6's broken <atomic> header, which defines
a non-inline namespace, then reopens it as inline to try to add its symbols to
the surrounding namespace. In this one special case, permit the namespace to be
reopened as inline, and patch up the name lookup tables to match.
Bill Schmidt [Thu, 4 Oct 2012 20:18:38 +0000 (20:18 +0000)]
Add a test case for r156143, which enabled general varargs support for the
64-bit PPC SVR4 ABI.
The test verifies passing of structures, items with 16-byte alignment, and
small items that are passed right-justified in the parameter save area slot.
Benjamin Kramer [Thu, 4 Oct 2012 19:42:20 +0000 (19:42 +0000)]
Driver: Link crtfastmath.o if it's available and -ffast-math is specified.
crtfastmath.o contains routines to set the floating point flags to a faster,
unsafe mode. Linking it in speeds up code dealing with denormals significantly
(PR14024).
For now this is only enabled on linux where I can test it and crtfastmath.o is
widely available. We may want to provide a similar file with compiler-rt
eventually and/or enable it on other platforms too.
Michael Han [Thu, 4 Oct 2012 16:42:52 +0000 (16:42 +0000)]
Fix scope location when parsing GNU attributes.
For GNU attributes, instead of reusing attribute source
location for the scope location, use SourceLocation() since
GNU attributes don not have scope tokens.
projects / clang / log