Re: [PATCH PR gdb/16841] virtual inheritance via typedef cannot find base

[Weimin Pan <weimin.pan@oracle.com>]

On 7/24/2018 3:18 PM, Simon Marchi wrote:
> On 2018-07-20 07:16 PM, Weimin Pan wrote:
>> Finding data member in virtual base class
>>
>> This patch fixes the original problem - printing member in a virtual base,
>> using various expressions, do not yield the same value. Simple test case
>> below demonstrates the problem:
>>
>> % cat t.cc
>> struct base { int i; };
>> typedef base tbase;
>> struct derived: virtual tbase { void func() { } };
>> int main() { derived().func(); }
>> % g++ -g t.cc
>> % gdb a.out
>> (gdb) break derived::func
>> (gdb) run
>> (gdb) p i
>> $1 = 0
>> (gdb) p base::i
>> $2 = 0
>> (gdb) p derived::base::i
>> $3 = 0
>> (gdb) p derived::i
>> $4 = 4196392
>>
>> To fix the problem, the virtual-base offset, relative to its derived class,
>> needs to be fetched, via baseclass_offset(), and used in calculating the
>> address of its data member in value_struct_elt_for_reference().
> Hi Weimin,
>
> I have looked at this a little bit, but unfortunately I don't really understand
> what's going on (maybe somebody else does and could review it?).  I understand
> the issue, and can see that your patch fixes it, but I don't understand how it
> does it.  It would help if you could walk us through your code.  It maybe also
> means that some comments would be appropriate, to explain what's going on.
>
> Thanks,
>
> Simon

Hi Simon,

Since a virtual base offset is stored in its derived class's vtable, my 
code simply (1) checks if base
class "curtype" is virtual in its derived class "domain" and (2) calls 
baseclass_offset() to get its
offset if it is.

Please check out the following article which provides a detailed 
explanation about the layout:

Memory Layout for Multiple and Virtual Inheritance

https://web.archive.org/web/20160413064252/http://www.phpcompiler.org/articles/virtualinheritance.html

Thanks,
Weimin