[cxx-abi-dev] Question about tail padding

[scott douglass]

At 05:39 PM 15-08-02 +0100, you wrote:
>mendell at ca.ibm.com wrote:
>> 
>> With allocation of derived classes in tail padding, what happens in the
>> following case
>> 
>>   struct S1 {
>>     virtual void f();
>>     int i;
>>     char c1;
>>   };                    // sizeof (S1) == 16 in 64 bit mode
>> 
>>   struct S2 : public S1 {
>>     char c2;            // allocated next to c1
>>   };                    // sizeof (S2) == 16 in 64 bit mode
>> 
>> S2 s2;
>> S1& rs1 = &s2;
>> S1 s1;
>> 
>> rs1 = s1;  // this will clobber the char member of s2.
>> 
>> Is this a reasonable interpretation?  Doesn't the assignment of S1s copy
>> the tail padding?
>No. You must arrange that S1::operator= does not copy tail padding.
>(S1 is non-pod, pods are different).

If the tail padding must not be changed this will be quite inefficient when there is a trailing bit-field -- it makes you wish there were separate whole-object op= and a sub-object op=.