Re: [PATCHv4] gdb: fix handling of DW_AT_entry_pc of inlined subroutines

[Bernd Edlinger <bernd.edlinger@hotmail.de>]

On 10/29/24 15:49, Andrew Burgess wrote:
> @@ -344,6 +357,18 @@ struct block : public allocate_on_obstack<block>
>       startaddr and endaddr above.  */
>  
>    struct blockranges *m_ranges = nullptr;
> +
> +  /* The entry address for this block.  The value 0 is special and
> +     indicates that the entry address has not been set.
> +
> +     Using 0 as a special value is not ideal, targets for which 0 is a
> +     valid code address might run into problems if they want to use 0 as a
> +     block's entry address, but the alternative is to carry a flag
> +     indicating if m_entry_pc is valid or not, but that would make 'struct
> +     block' even bigger, and we want to keep 'struct block' as small as
> +     possible (we might have a lot of blocks).  */
> +
> +  CORE_ADDR m_entry_pc = 0;
>  };
>  

Aehm, I've just had an idea how you can get this much simpler:

Instead of using an exceptional value, and fiddle with the later relocation steps,
you should simply store here the offset from ranges[0].start, or block.start
if no subranges, this is just a constant value, that just has to be added in
block.entry_pc(), 0 means just the current behavour, and if >0 it is always a
constant value, even after objfile relocation.  That means there is no chance of any
special value that cannot be handled here, and a lot of complications can be avoided.

What do you think?


Thanks
Bernd.