[PATCH] D20943: [LibFuzzer] [WIP] Declare and use sanitizer functions in `fuzzer::ExternalFunctions`

[Kostya Serebryany via llvm-commits]

kcc added inline comments.

================
Comment at: lib/Fuzzer/FuzzerDriver.cpp:428
@@ +427,3 @@
+// Storage for global ExternalFunctions object.
+ExternalFunctions EF;
+
----------------
delcypher wrote:
> kcc wrote:
> > delcypher wrote:
> > > kcc wrote:
> > > > make this ExternalFunctions *EF
> > > We could also use  `std::unique_ptr<ExternalFunctions>` instead or are you fine with an explicit delete?
> > I am *not* ok with std::unique_ptr<ExternalFunctions>
> > And there is no need for an explicit delete. 
> > Just 
> >     EF = new  .... ;
> > in main is enough 
> > 
> > In tests you may have new/delete or even 
> >      std::unique_ptr<ExternalFunctions> t;
> >      EF = t.get()
> > I am *not* ok with std::unique_ptr<ExternalFunctions>
> > And there is no need for an explicit delete. 
> > Just
> > 
> > EF = new  .... ;
> > in main is enough
> 
> I don't understand. Why do you want a deliberate leak?
> 
> > In tests you may have new/delete or even
> > 
> > std::unique_ptr<ExternalFunctions> t;
> > EF = t.get()
> 
> Again I don't understand. Why would I want to assign nullptr (that is what `t.get()` would return here) to `EF`?
> 
> 
This won't be a leak. It will be a global variable which is alive throughout the process. 

I meant, in test you could either do 

    FE = new ...
     ....
     delete EF;
     EF = nullptr; // just in case

or, better
       
        std::unique_ptr<ExternalFunctions> t(new ...);
        EF = t.get()
        ...
        EF = nullptr;  // just tin case



http://reviews.llvm.org/D20943