[cfe-commits] r170825 - in /cfe/trunk: include/clang/Analysis/CallGraph.h lib/Analysis/CallGraph.cpp lib/StaticAnalyzer/Frontend/AnalysisConsumer.cpp

[Anna Zaks]

On Dec 20, 2012, at 5:53 PM, Jordan Rose <jordan_rose at apple.com> wrote:

> Forgot to say: nice work for yet more speedup!
> 
> 
> On Dec 20, 2012, at 17:19 , Anna Zaks <ganna at apple.com> wrote:
> 
>> 
>> +STATISTIC(NumObjCCallEdges, "Number of objective C call edges");
> 
> Super-nitpicking: s/objective C/Objective-C method/
> 
> 
>> +  // Adds may-call edges for the ObjC message sends.
> 
> Looks like there's nothing special about "may-call" edges anymore?
> 

What do you mean? It's a "may-call" because we are not 100% sure that is what will be called at runtime (We don't do the same checking as in CallEvent to be 100% sure.).

> 
>> +  void VisitObjCMessageExpr(ObjCMessageExpr *ME) {
>> +    if (ObjCInterfaceDecl *IDecl = ME->getReceiverInterface()) {
>> +      Selector Sel = ME->getSelector();
>> +      
>> +      // Fild the callee definition within the same translation unit.
> 
> Typo: "Fild"
> 
> 
>> void CallGraphNode::print(raw_ostream &os) const {
>> -  os << getName();
>> +  if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(FD))
>> +      return ND->printName(os);
>> +  os << "< >";
>> } 
> 
> It'd be nice to have a way to tell blocks apart; some parts of our code say "anonymous block at <line>:<col>", but even just a pointer value would be nice.
> 
> 
>> -    return Node->getName();
>> +    if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(Node->getDecl()))
>> +      return ND->getNameAsString();
>> +    else
>> +      return "< >";
> 
> This duplication is an unfortunate downside of removing getName(). Maybe print to a local buffer?
>