On Fri, Aug 15, 2014 at 10:45 AM, Joey Ye <joey.ye...@gmail.com> wrote:
> Running into an unexpected result with GCC with following case, but
> not sure if it is a valid C++ case.
>
> #define nullptr 0
> enum nonetype { none };
>
> template<typename T>
> class class_zoo {
> public:
> const T *data;
> int length;
>
> class_zoo (nonetype) : data (nullptr), length (0) {}
> class_zoo (const T &e) : data (&e), length (1) {}
Capturing a const referece via a pointer is error-prone as for
example literal constants class_zoo<const int *> zoo(0)
have associated objects that live only throughout the function
call.
So clearly your testcase is invalid.
Richard.
> };
>
> int bar (class_zoo<const int *> p1 = none,
> class_zoo<const unsigned *> p2 = none)
> {
> if (*p1.data==nullptr) return 1;
> return 0;
> }
>
> int foo (int *b) // If changing int to const int here, zoo will
> return 0 (pass)
> {
> class_zoo<const int *> zoo(b);
> return bar(zoo);
> }
>
> int g = 678;
> int main ()
> {
> return foo (&g);
> }
>
> $ g++ main.cpp -fstack-protector -o m
> $ ./m
> $ echo $?
> 1
>
> Expand shows:
> D.2320 = b_2(D);
> class_zoo<const int*>::class_zoo (&zoo, &D.2320);
> D.2320 ={v} {CLOBBER}; <------- D.2320 is dead, but it is address
> escapes to zoo
> <------- with
> stack-protector, D.2322 reuses stack slot of D.2320, thus overwrite
> D.2320
> class_zoo<const unsigned int*>::class_zoo (&D.2322, 0);
> _8 = bar (zoo, D.2322); <-------- D.2320 is accessed via its
> address, but value changed now
> _9 = _8;
> D.2322 ={v} {CLOBBER};
> zoo ={v} {CLOBBER};
> ;; succ: 3
>
> ;; basic block 3, loop depth 0
> ;; pred: 2
> <L2>:
> return _9;
> ;; succ: EXIT
>
> }
>
>
> Partition 0: size 16 align 16
> D.2322 D.2320 <------- GCC believes they are not
> conflict, which is what I not sure of here
> Partition 2: size 16 align 16
> zoo
>
> Questions
> 1. Is this a correct test case?
> 2. If not, why escaping of D.2320's address to a local is not
> accounted as conflicting to other local variables?
>
> Thanks,
> Joey
