On Fri, May 22, 2015 at 4:12 PM, Aldy Hernandez <al...@redhat.com> wrote:
> On 05/22/2015 07:23 AM, Richard Biener wrote:
>>
>> On Wed, May 20, 2015 at 5:50 PM, Aldy Hernandez <al...@redhat.com> wrote:
>>>
>>> On 05/18/2015 06:56 AM, Richard Biener wrote:
>
>
>>>> diff --git a/gcc/tree-core.h b/gcc/tree-core.h
>>>> index ad1bb23..2a9f417 100644
>>>> --- a/gcc/tree-core.h
>>>> +++ b/gcc/tree-core.h
>>>> @@ -1334,6 +1334,9 @@ struct GTY(()) tree_block {
>>>> tree abstract_origin;
>>>> tree fragment_origin;
>>>> tree fragment_chain;
>>>> +
>>>> + /* Pointer to the DWARF lexical block. */
>>>> + struct die_struct *die;
>>>> };
>>>>
>>>> struct GTY(()) tree_type_common {
>>>>
>>>> Ick - do we need this? dwarf2out.c has a hashtable to map blocks to
>>>> DIEs (which you don't remove in turn).
>>>
>>>
>>>
>>> We need a way to reference the early created DIE from late debugging, and
>>> we
>>> can't use block_map because it gets cloberred across functions. It's
>>> currently being released in late debug (dwarf2out_function_decl),
>>> that's why you see it not set to NULL in dwarf2out_c_finalize.
>>>
>>> Also, it uses BLOCK_NUMBERs, which according to the documentation in
>>> tree.h, are not guaranteed to be unique across functions.
>>>
>>> As Honza mentioned, we're already using a DIE map in types through
>>> TYPE_SYMTAB_DIE. See lookup_type_die() in dwarf2out.c.
>>>
>>> Could we leave this as is?
>>
>>
>> But why then not eliminate block_map in favor of using the new ->die
>> member?
>> Having both looks very odd to me.
>
>
> Oh, I would love to. I just didn't want to rip things apart elsewhere until
> I was sure you guys were on board with the approach.
>
>> Can you cook up a patch for trunk adding that field to tree_block and
>> removing
>> the block_map map in favor of sth like what we do for
>> lookup_type_die/equate_type_number_to_die
>> and TYPE_SYMTAB_DIE?
>
>
> Absolutely! The attached patch removes block_map in favor of BLOCK_DIE. I
> did not add lookup_block_die/equate_block_number_to_die abstractions because
> I think BLOCK_DIE is pretty straightforward.
>
> The attached patch is against mainline. I also ported it to the branch for
> testing, and neither the branch nor mainline exhibit any regressions.
>
> Tested on x86-64 Linux with --enable-languages=all,go,ada.
>
> OK for trunk?
Ok.
Thanks,
Richard.
> Aldy
>
