On 9/18/06, Miles Bader <miles@xxxxxxx> wrote:
"Peng Yu" <pengyu.ut@xxxxxxxxx> writes: > I think the compiler should suspect there is something wrong, because > the buggy code assigns the reference of a temp variable to a const > reference. It should be that a const reference be assigned to another > const reference. It's not "assigning a temp to a const reference" (that's done all the time, and is perfectly innocent), it's "assigning a reference to a variable with a certain extent to a reference with a longer-lived extent." Does gcc have the information easily available to detect that?
Assuming it happens in part of a program we can see, we could detect that it is stored away in a variable that may live longer than the original reference. We could prove that in this case, the store always happens, and thus, there will always be a dangling reference. However, unless you made this a flow-sensitive, context-sensitive interprocedural analysis (which would probably be darn slow given a large amount of code), it would be very easy to turn that "must" information into "may" information.
