and OwningPtr instead of constructing only after all of sema is done.  This
has a couple of effects:
1. it fixes memory leaks from all the error cases in sema
2. it simplifies the code significantly.

The cost of this is that the error case now new's and delete's an expr where
it did not before, but we don't care about the perf of the error case.

llvm-svn: 45380

of conforming protocols (or not).

llvm-svn: 45276

More is yet to come.

llvm-svn: 45263

are due to arrive.

llvm-svn: 45244

llvm-svn: 45243

llvm-svn: 45239

llvm-svn: 45235

llvm-svn: 45234

both Preprocessor and ASTContext, we no longer need to explicitly pass
MainFileID around in function calls that also pass either Preprocessor or
ASTContext. This resulted in some nice cleanups in the ASTConsumers and the
driver.

llvm-svn: 45228

Various tweaks to the get/lookup instance/class method API's.

llvm-svn: 45224

llvm-svn: 45205

llvm-svn: 45204

protocol list.

llvm-svn: 45203

llvm-svn: 45202

Revert Anders r45191 commit...it broke several of the tests.

llvm-svn: 45199

llvm-svn: 45195

llvm-svn: 45191

llvm-svn: 45189

llvm-svn: 45188

llvm-svn: 45174

llvm-svn: 45152

Add DefaultFunctionArrayConversion() to the indirection operator in Sema::ActOnUnaryOp().

This fixes a bug Chris forwarded from Oliver Hunt...

typedef struct { char name[100]; } entry;
char f1(entry *e) { return *e->name; }

llvm-svn: 45148

Fixe bogus error for variable argument methods. Sema::ObjcGetTypeForMethodDefinition() wasn't preserving the isVariadic boolean. Another fix is to avoid synthsizing the function decl entirely, however this is a separate issue that I don't want to deal with now. Also added a FIXME to Sema::CheckFunctionCall(), which is currently emitting a bogus warning.

llvm-svn: 45146

Improve how we find private method decls. This involved:

- Changed Sema::ObjcActOnStartOfMethodDef() to register the methods with the global pools.
- Changed Sema::ActOnInstanceMessage() to look in global pools (should be much less error prone).
- Added a test case to message.m (for lookup that was broken).

Misc changes while I was investigating this...

- Changed Sema::ActOnAtEnd() to call AddFactoryMethodToGlobalPool (this looked like a cut/paste error).
- Added a comment and tweaked another where I was using the first person.

llvm-svn: 45142

the protocol list (id<P,...> types).

llvm-svn: 45121

following hold:

(1) A vprintf-like function is called that takes the argument list via a
    via_list argument.

(2) The format string is a non-literal that is the parameter value of
    the enclosing function, e.g:
    
    void logmessage(const char *fmt,...) { 
      va_list ap;
      va_start(ap,fmt);
      fprintf(fmt,ap);  // Do not emit a warning.
    }
    
In the future this special case will be enhanced to consult the "format"
attribute attached to a function declaration instead of just allowing a blank
check for all function parameters to be used as format strings to vprintf-like
functions. This will happen when more support for attributes becomes
available.

llvm-svn: 45114

llvm-svn: 45110

Sema::ActOnMemberReferenceExpr() needs to perform the default conversions.

Bug and test case provided by Carl Lewis.

llvm-svn: 45078

- Remove getInstanceMethods/getClassMethods API on ObjcInterfaceDecl, ObjcProtocolDecl, and ObjcCategoryDecl. These methods are replaced by the respective iterators on each class.
- Add getInstanceMethodForSelector to ObjcInterfaceDecl, ObjcProtocolDecl, and ObjcCatgoryDecl. This hook will do a "shallow" lookup. This is a convenience method that reducing some of the iterator usage.
- Various changes to convert all clients to the above API's...
 

llvm-svn: 45046

llvm-svn: 45014

Sema::ActOnFinishFunctionBody(): Since we no longer synthesize a FunctionDecl for each method, remove the unconditional initialization to 0 and corresponding comment.

llvm-svn: 45008

Don't do integer promotions of LHS for compound shift assignment.  The LHS has to be a modifiable lvalue.

llvm-svn: 44993

llvm-svn: 44964

Moved all clients of Diagnostics to use FullSourceLoc instead of SourceLocation.
Added many utility methods to FullSourceLoc to provide shorthand for:

    FullLoc.getManager().someMethod(FullLoc.getLocation());
    
instead we have:

    FullLoc.someMethod();
    
Modified TextDiagnostics (and related classes) to use this short-hand.

llvm-svn: 44957

llvm-svn: 44945

llvm-svn: 44941

llvm-svn: 44938

llvm-svn: 44925

int test(int x, long long y) {
  return x << y;
}

we now realize the type of the shift is int, not long long.

This fixes a fixme from june.

llvm-svn: 44912

this is passed to sema and ignored there, so the second part of the
string will not make it into the AST.  Passing to Fariborz to finish
Sema + AST construction.

llvm-svn: 44898