On Tue, May 10, 2011 at 4:24 PM, Paul Pluzhnikov <ppluzhni...@google.com> wrote:
> On Tue, May 10, 2011 at 7:02 AM, Ian Lance Taylor <i...@google.com> wrote:
>> Richard Guenther <richard.guent...@gmail.com> writes:
>
>>> I wonder why this is a GCC specific patch and not a linker patch. Why
>>> not change the linker(s) to accept such configure option that changes its
>>> default behavior?
>>
>> It is traditionally gcc which tells the linker what to do. E.g., Fedora
>> has patched gcc to pass --hash-style=gnu to the linker.
>
> I think 'traditionally' is the key here.
>
> The same argument applies to e.g. --build-id. We already have GCC
> supply that, but we could have changed the default in
> binutils/{ld,gold}.
I think we patched binutils for hash-style (we use =both) and gcc for build-id.
Richard.
> Also, the argument is discoverable (by observing 'gcc -v' output).
> Changing the default in {ld,gold} makes it hard(er) to understand why
> e.g. binaries linked from the same object files using two different
> versions of GCC are different.
>
>>> Otherwise if people link with ld they suddenly get different hash-style.
>>> That looks wrong to me.
>>
>> That turns out not to be the case.
>
> I think Richard meant "if people link directly with either ld or gold
> (i.e. bypassing 'gcc' driver) get different hash-style". That's kind
> of expected though -- if you want to get the same output you get from
> 'gcc', you need to examine 'gcc -v' very carefully. And, as Jakub
> noted, linking directly with 'ld' is discouraged.
>
> Thanks,
> --
> Paul Pluzhnikov
>
