llvm dev - May 2017 - clang assembler keeps .Linfo_string symbols?

On 3 May 2017 at 08:01, Rafael Avila de Espindola via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Can you provide a .s file that shows the difference?
I managed to reproduce it by compiling 'char *var =
"whatever";' in
debug mode (so clang -target arm-none-eabi tmp.c -g"). Direct compile
leaves no assembler-local strings, but for some reason going via -S
does. I've not investigated further I'm afraid.

Tim.

Tim Northover <t.p.northover at gmail.com> writes:
> On 3 May 2017 at 08:01, Rafael Avila de Espindola via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
>> Can you provide a .s file that shows the difference?
>
> I managed to reproduce it by compiling 'char *var =
"whatever";' in
> debug mode (so clang -target arm-none-eabi tmp.c -g"). Direct compile
> leaves no assembler-local strings, but for some reason going via -S
> does. I've not investigated further I'm afraid.
Sorry, had somehow replied off list:

   This reduces to

           .section        .debug_str,"MS",%progbits,1
   .Linfo_string8:
           .long   .Linfo_string8

   the difference comes from ELFObjectWriter::shouldRelocateWithSymbol. In
   particular

       // It looks like gold has a bug (http://sourceware.org/PR16794) and can
       // only handle section relocations to mergeable sections if using RELA.
       if (!hasRelocationAddend())
         return true;

   If I remove that we produce a relocation with the section, as does
   gas. The gold bug has not been fixed :-(

   Cheers,
   Rafael

That was probably my fault - not sure of the niceties of the list. Also 
not sure I understand why the integrated assembler behaves differently. 
I guess gold is never used with object files produced this way?

Regards
Michael
> Tim Northover <t.p.northover at gmail.com> writes:
>
>> On 3 May 2017 at 08:01, Rafael Avila de Espindola via llvm-dev
>> <llvm-dev at lists.llvm.org> wrote:
>>> Can you provide a .s file that shows the difference?
>> I managed to reproduce it by compiling 'char *var =
"whatever";' in
>> debug mode (so clang -target arm-none-eabi tmp.c -g"). Direct
compile
>> leaves no assembler-local strings, but for some reason going via -S
>> does. I've not investigated further I'm afraid.
> Sorry, had somehow replied off list:
>
>     This reduces to
>
>             .section        .debug_str,"MS",%progbits,1
>     .Linfo_string8:
>             .long   .Linfo_string8
>
>     the difference comes from ELFObjectWriter::shouldRelocateWithSymbol. In
>     particular
>
>         // It looks like gold has a bug (http://sourceware.org/PR16794) and
can
>         // only handle section relocations to mergeable sections if using
RELA.
>         if (!hasRelocationAddend())
>           return true;
>
>     If I remove that we produce a relocation with the section, as does
>     gas. The gold bug has not been fixed :-(
>
>     Cheers,
>     Rafael
>