================
@@ -31,6 +31,20 @@ void test_unclaimed_use(int *p) { // expected-warning{{'p'
is an unsafe pointer
p[5] = 5; // expected-note{{used in buffer access here}}
}
+namespace std {
+inline namespace __1 {
+template <class T> class unique_ptr {
+public:
+ T &operator[](long long i) const;
+};
+} // namespace __1
+} // namespace std
+
+void basic_unique_ptr() {
+ std::unique_ptr<int[]> p1;
+ p1[0]; // expected-warning{{direct access using operator[] on
std::unique_ptr<T[]> is unsafe due to lack of bounds checking}}
----------------
jkorous-apple wrote:
Accessing the first element (index 0) is ... special.
We need to keep the C++ Safe Buffers programming model rules consistent for
both pointers and arrays to not confuse the users.
Pointer dereference and accessing the first element are identical operations
(`buf[0]` being a different syntax for `*buf`).
Our view of built-in pointer dereference is that it is NOT unsafe from the
bounds-correctness perspective. While it can lead to invalid memory access it
won't be caused by an invalid bound but rather the pointer itself being
invalid. For those issues we need a different approach.
Because of that the programming model allows accessing the first element of
built-in arrays and pointers and pointer dereference.
Also, the programming model is very intrusive and our goal is to keep it from
becoming prohibitively intrusive - in other words we don't want the warning to
be overly noisy. By prohibiting pointer dereference and access to the first
element we would essentially ban all pointers and arrays (and now
`std::unique_ptr<T[]>`).
While I am open to having a discussion about the above I suggest that for now
we keep the rules consistent for built-in pointers, arrays and
`std::unique_ptr<T[]>` and if we were to prohibit `buff[0]` let's discuss it
and possibly make the change also consistently for all three buffer types.
https://github.com/llvm/llvm-project/pull/156773
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
