[PATCH v1 1/1] gas/write.c: add explicitly flag SEC_HAS_CONTENT to object attributes section

[Jan Beulich]

On 30.06.2025 18:08, Matthieu Longo wrote:
> On 2025-06-27 07:55, Jan Beulich wrote:
>> On 25.06.2025 16:57, Matthieu Longo wrote:
>>> The object attributes section:
>>> - is readonly (SEC_READONLY=1).
>>> - only contains data (SEC_DATA=1).
>>> - does not require the OS to allocate space at runtime
>>>    (SEC_ALLOC=0).
>>> - does not require the OS to load the section at runtime
>>>    (SEC_LOAD=0).
>>> - is written out to the output object if it is not empty
>>>    (SEC_HAS_CONTENT=1).
>>>
>>> Today, the setting of SEC_HAS_CONTENT (if not specified by
>>> bfd_set_section_flags) occurs a few lines below inside
>>> size_seg (see code below), the last function called in
>>> create_obj_attrs_section().
>>>
>>> // If this section is not empty (size > 0) and this section
>>> // is not .bss, the flag is set.
>>>    if (size > 0 && ! seginfo->bss)
>>>      flags |= SEC_HAS_CONTENTS;
>>>
>>> Setting the flag SEC_HAS_CONTENT via bfd_set_section_flags()
>>> instead of relying on the hidden side-effect of size_seg()
>>> seems a more maintainable approach due to its explicitness.
>>
>> One can also look at this the other way around: It may be undue open-
>> coding that you suggest to add. Look at maybe_generate_build_notes():
>> It sets the flag unconditionally, but calls bfd_set_section_contents()
>> only conditionally. If it (properly) left the setting of the flag to
>> size_seg(), things would be more consistent overall. (And yes,
>> create_obj_attrs_section() doesn't have this same issue as it checks
>> for zero size early on. The more abstract open-coding aspect remains,
>> though.)
> 
> If I put in parallel this function with maybe_generate_build_notes(), 
> why is SEC_HAS_CONTENTS set explicitly and still size_seg() is called at 
> the end ?

Did you read my reply carefully, though? I actually mention that function
there, indicating it probably isn't quite right.

Jan