collisions.
 - Provide CodeGenModule::CreateRuntimeFunction which guarantees that
   the function it creates will have the provided name in the final
   module. This allows the runtime to have its functions protected
   from declarations of the same name in the source code.

 - One could argue that this is a reason to abuse the llvm::Module
   namespace for dealing with function redeclarations. However, that
   approach seems conceptually flawed to me. This one also happens to
   be somewhat more efficient.

No functionality change.

llvm-svn: 56899

Use this updated interface when invalidating arguments passed by reference; the type of symbol is of the object passed by reference, not the reference itself.

llvm-svn: 56894

llvm-svn: 56880

llvm-svn: 56879

llvm-svn: 56876

like "10.3.9"

llvm-svn: 56873

llvm-svn: 56859

Fix <rdar://problem/6191148> [sema] Objective-C method lookup (at global scope) fails to handle overloaded selectors properly.

Long standing bug in Sema::ActOnInstanceMessage(). We now warn when messaging an "id" with multiple method signatures in scope. The diags are a little verbose, however they can be streamlined if necessary. 

llvm-svn: 56843

Needed to make sure the relaxed type checking only applies to "readonly" properties.

llvm-svn: 56838

llvm-svn: 56836

llvm-svn: 56832

from the subexpression type to the expression type.

llvm-svn: 56831

llvm-svn: 56830

 - Provides a basic primitive to jump to an arbitrary basic block,
   through the finally code.

 - Only used for return statements and rethrow currently. Still need
   to handle break, continue and goto.

 - Code still needs to be shuffled around to live elsewhere.

llvm-svn: 56827

the target triple on darwin.  For example i386-apple-darwin9 -> 1050 because
darwin9 is "10.5".

llvm-svn: 56826

llvm-svn: 56824

llvm-svn: 56822

llvm-svn: 56821

NumericLiteral parser is not careful about overrun because
it should never be possible.  It implicitly expects that its
input matched the regex for pp-constant.  Because of this, it
knows it can't be pointing to a prefix of something that
looks like a number.  This is all fine, except that __LINE__
does not prevent implicit concatenation from happening.  Fix
__LINE__ to not do this.

llvm-svn: 56818

fixes problems handling references of va_list, which happens on x86_64.
This fixes PR2841 and rdar://6252231

llvm-svn: 56809

and add a dump method to FullSourceLoc!  Patch by Nico Weber!

llvm-svn: 56806

llvm-svn: 56800

Teach Sema::CheckAssignmentConstraints() to allow assignments between id and block pointer types (^{}).

llvm-svn: 56793

llvm-svn: 56791

llvm-svn: 56785

llvm-svn: 56776

llvm-svn: 56775

Fix <rdar://problem/6252129> implementation of method in category doesn't effectively declare it for methods below.

llvm-svn: 56771

llvm-svn: 56770

by decaying __builtin_va_list's type when forming builtins.  On
x86-64 (and other targets) __builtin_va_list is a typedef for
an array.

llvm-svn: 56768

wasn't covering checking builtins like __builtin___vsnprintf_chk) with
a check that won't get out of date. 

llvm-svn: 56767

don't need a BlockDeclRefExpr.

llvm-svn: 56766

llvm-svn: 56764

Comment exception-handling code generation strategy.

llvm-svn: 56763

llvm-svn: 56762

llvm-svn: 56761

readable and to generate more readable IR.

llvm-svn: 56758

 - This should be rejected by sema (IMHO) but that needs some
   investigation.

llvm-svn: 56756

llvm-svn: 56755

 - There is no need to branch on the rethrow variable to determine if
   we need to call objc_exception_try_exit. We know whether an
   exception was thrown, so just branch to a different target if we
   want to skip the try_exit. 

   This is a slight semantic departure from gcc, but only for throwing
   nil, which is undefined (and for which gcc emits broken code).

 - Also fixes a bug in current code which was calling try_exit too
   many times when an exception was uncaught (but there were some
   handlers).

Fix bug introduced in prev. commit, the type of the @catch parameter
was uninitialized.

llvm-svn: 56754