https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88565
Bug ID: 88565
Summary: enhance -Warray-bounds for C++ trailing class member
arrays
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Severity: enhancement
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: msebor at gcc dot gnu.org
Target Milestone: ---
While thinking about pr84051 and about the test case in attachment 20033 to
pr43270, it occurred to me that:
GCC doesn't differentiate between out-of-bounds accesses to trailing member
arrays in C or C++: in both languages it refrains from issuing -Warray-bounds=1
regardless of the array's bound. The rationale in C is that such arrays could
be treated as flexible array members by the client code.
But in the case of C++ classes this same logic is much less likely to apply
because such uses are much less common there. It almost certainly doesn't
apply when the array is a member of a class with a user-defined ctor. Objects
of such classes are typically only initialized using the ctor which largely
rules out allocating more space for the object than the size of its type
implies.
With this in mind, it would be appropriate to issue -Warray-bounds even at
level 1 for the following code:
struct S
{
S ();
int f ();
int a[3];
};
int S::f ()
{
return a[7]; // request: issue -Warray-bounds=1
}
This line of reasoning would also suggest that accesses to such arrays could be
diagnosed even for zero-length arrays as requested in pr43270. Almost
certainly when such an array is not the last member of a class as in the test
case provided in that bug.
