[cfe-commits] plug a memory leak in the -E code

[Chris Lattner]

>>>
>>> When clang is acting on the -E option, it creates a
>>> PrintPPOutputPPCallbacks object but never deletes it.  The object is
>>> passed to the Preprocessor object, but it doesn't delete it either.
>>>
>>> It seems to me that the Driver code ought to do the deleting.  So I
>>> have attached a patch which does just that.
>>
>> Nice catch.  However, I think it makes more sense for the PP to take
>> ownership of the callbacks object.  What do you think?
>
> Do the LLVM and clang projects have a general rule of thumb for making
> these kinds of decisions?  "The object/module that calls new calls
> delete" is the rule I have always gone with, but I can see your point
> and could be easily swayed.

Not specifically.  This is why it's really important to add good  
comments (like you did!) saying "this method takes ownership" or not.

-Chris