http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49115

           Summary: invalid return value optimization (?) when exception
                    is thrown and caught
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassig...@gcc.gnu.org
        ReportedBy: ar...@pvv.ntnu.no


we have some code that started failing when I tried upgrading our tool chain to
GCC 4.5.2; it gets stale data in an object when an exception is thrown during
assignment.

It looks to me like return value optimization will cause the object constructor
to be optimized away, but I'm not totally certain if that is the actual
problem, nor what the C++ language guarantees in this situation.

A self-contained test program is below, failing for me on x86_64 using:

GNU C++ (GCC) version 4.5.2 (x86_64-unknown-linux-gnu)
    compiled by GNU C version 4.5.2, GMP version 4.3.2, MPFR version 3.0.1, MPC
version 0.9

----- test program follows -----

extern "C" { extern long write(int, const void *, unsigned long); }

struct MyException {};
int simfail = 7;

struct Data {
    int nr;
    Data() : nr(66) {}
};

Data getData(int i) {
    if (simfail == i) throw MyException();
    Data data;
    data.nr = i;
    return data;
}

bool verify() {
    char bad[7] = "BAD x\n";
    for (int i = 0; i < 10; i++) {
        Data data;
        try {
            data = getData(i);
        } catch (MyException& e) {
            if (data.nr != 66) {
                bad[4] = '0' + data.nr;
                write(2, bad, 6);
                return false;
            }
        }
    }
    return true;
}

int main(int, char **) {
    simfail = 4;
    return verify() ? 0 : 1;
}

Reply via email to