[cfe-dev] CFG blocks and variable scope

[Ted Kremenek]

On Mar 28, 2009, at 5:25 PM, Chris Lattner wrote:

>
> On Mar 28, 2009, at 5:23 PM, Ted Kremenek wrote:
>
>>> Would it make sense for the CFG to contain a "virtual statement"  
>>> saying "variable x destroyed here"?  This would be a natural way  
>>> to handle C++ dtors and would also be useful in C, because you'd  
>>> know the end of the variable's life.  CFG construction would  
>>> handle this as it is walking the scopes.
>>>
>> Yes, this is the approach I was envisioning.  We still need the  
>> scope information to be available during CFG construction.
>
>
> For declstmt's?  I think the AST has enough information to know  
> scopes. What is missing?

I would like a solution that will work well with C++ (e.g., the  
construction/destruction of temporaries, etc.), so having an ad hoc  
solution to reconstruct scope and other places where objects get  
destroyed seems suboptimal.    I suppose the CFG could construct that,  
but it might be nice if that information was affixed to the AST in  
some way.  Other clients (e.g., CodeGen) might want to use that as well.