Re: [PATCH] readelf: Warn zero-sized relocation sections

[Jan Beulich <jbeulich@suse.com>]

On 20.04.2020 12:28, Nick Clifton wrote:
>>>   I am of a mind to allow this patch, given that one of readelf's
> 
>> While I can see the "diagnose problems" aspect, zero-sized relocation
>> sections aren't something the ELF standard disallows. So there's no
>> problem with the ELF files in this case, just a _possible_ problem
>> with consumers thereof.
> 
> True - but in this case we are talking about relocation sections, and
> what possible use is an empty relocation section ?  Its presence is
> legal sure, but it can also be an indication that the tool that created
> the binary is not quite doing its job properly.

It is still legitimate for a tool to create _all_ sections it may
possibly put data into. While now causing headaches, that ppc
behavior mentioned further down actually demonstrates this.

> I personally like the idea of giving users as much information as possible
> and allowing them to decide if they want it.  After all it is relatively
> easy to remove any warning out that is undesired via command line tools.

Is it, even when considering the messages getting translated to
other languages? What would my pattern be to filter them out
regardless of language?

Anyway, I merely meant to voice my opinion. You're the maintainer,
so you've got to judge.

>>>     readelf: Warning: Section '.rela.dyn': zero-sized relocation section
>>
>> Aren't these a good enough indication that zero-sized relocation
>> sections aren't this unusual, and hence a default-on warning is
>> going too far?
> 
> Actually no - these failures have now prompted H.J. and Alan to have a look
> at the code producing these empty sections and see about fixing them.

Yes, I've seen this. Not emitting these sections is certainly a
good thing to do, but aiui current behavior is still legitimate
(and hence I'd call this clean up, not "fixing").

Jan