You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
[analyzer] Improve handling of placement new in PointerArith (#155855)
This pull improves the handling of placement new in`PointerArith`,
fixing one family of false positives, and one of negatives:
### False Positives
```cpp
Buffer buffer;
int* array = new (&buffer) int[10];
++array; // there should be no warning
```
The code above should flag the memory region `buffer` as reinterpreted,
very much as `reinterpret_cast` would do. Note that in this particular
case the placement new is inlined so the engine can track that `*array`
points to the same region as `buffer`.
This is no-op if the placement new is opaque.
### False Negatives
```cpp
Buffer buffer;
int* array = new (&buffer) int;
++array; // there should be a warning
```
In this case, there is an implicit cast to `void*` when calling
placement new. The memory region was marked as reinterpreted, and
therefore later pointer arithmetic will not raise. I have added a
condition to not consider a cast to `void*` as a reinterpretation, as an
array of voids does not make much sense.
There are still some limitations, of course. For starters, if a single
`int` is created in place of an array of `unsigned char` of exactly the
same size, it will still be considered as an array. A convoluted example
to make the point that I think it makes sense *not* to raise in this
situation is in the test `checkPlacementNewSlices`.
CPP-6868
0 commit comments