Re: [PATCHv5] gdb: fix handling of DW_AT_entry_pc of inlined subroutines

[Andrew Burgess <aburgess@redhat.com>]

Bernd Edlinger <bernd.edlinger@hotmail.de> writes:

> On 11/5/24 12:25, Andrew Burgess wrote:
>> Bernd Edlinger <bernd.edlinger@hotmail.de> writes:
>> 
>>> On 10/31/24 16:00, Andrew Burgess wrote:
>>>> +  /* Set this block's entry-pc to ADDR, which must lie between start() and
>>>> +     end().  */
>>>> +
>>>> +  void set_entry_pc (CORE_ADDR addr)
>>>> +  {
>>>> +    gdb_assert (addr >= this->start () && addr < this->end ());
>>>
>>> note that my patch series part 3/3 has an example where
>>> the inline funcion has an empty instruction range, I wonder
>>> if this assertion will break, maybe addr <= this->end ?
>> 
>> I ran the tests from the full optimised debug series.  Obviously there's
>> plenty of failures at this point, but there are no assertions triggered.
>> 
>
> Ah, okay, then this is probably just a complaint.
>
>> Using 'addr <= this->end' would, I think not be correct, as the block's
>> end() address is not inclusive.  That is, if 'addr == this->end ()' then
>> the entry-pc would be the first address outside the block's range, which
>> is clearly not correct.
>> 
>
> Yeah likely, but maybe with the exception when start==end && start==entry_pc?

But right now such empty ranges (when start == end) are discarded, so
this would appear as an entry_pc outside of the block's range, which
would mean we don't attempt to set the entry-pc.

If/when we later "fix" the handling of empty ranges then we might have
to do something here, but that will depend on how we choose to handle
the empty range problem.  I don't think we should preemptively try to
adjust for that future fix at this point.  But it's certainly something
we can keep in mind that might need investigating later on.

Thanks,
Andrew