[cfe-commits] r61147 - in /cfe/trunk: lib/Analysis/CFRefCount.cpp test/Analysis/uninit-vals-ps.c

[Zhongxing Xu]

On Tue, Mar 10, 2009 at 2:09 PM, Ted Kremenek <kremenek at apple.com> wrote:

>
> On Mar 9, 2009, at 10:48 PM, Zhongxing Xu wrote:
>
>  Hi Ted,
>>
>> We don't have any casts for this example. So why would p and q point to
>> different region?
>>
>> typedef struct s* MyPointer;
>>
>> struct s* foo();
>>
>> void bar() {
>>  MyPointer p = foo();
>>  struct s* q = p;
>> }
>>
>
>
> That's a fair point.  There are no casts in this example.  That doesn't
> mean, however, that we don't have to reason about them when they are
> present.  Put another way, I'm not convinced it isn't a problem.
>
> For example, considered the following contrived example:
>
>  struct s* p = ...
>  p->f = 5;
>
>  void* r = p;
>  MyPointer *q = r;
>  if (q->f == 5)
>    // do something
>
>  if (q == p)
>    // do something
>
> While contrived, we should be able to readily determine that q->f is indeed
> 5 at the first branch as well as the fact that q == p at the second branch.
>  We have a variety of hacks in place to handle some of these cases; it would
> be nice if there was just a clear, consistent model of what different kinds
> of region views mean and how they should be interpreted.
>
> The point of having a clear model for region views and their semantics that
> handles all cases consistently (and inductively) means that the corner cases
> just get naturally handled without us having to specifically consider them.


I see your point. But how can we recognize which casts are semantic changes
and which are sugar typed?
I doubt there is a consistent rule for this.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20090310/d121949e/attachment.html>