[PATCH] [GNU] Implement --enable-new-dtags/--disable-new-dtags

[Rui Ueyama]

On Mon, Apr 6, 2015 at 7:02 PM, Ed Maste <emaste at freebsd.org> wrote:

> On 6 April 2015 at 21:38, Rui Ueyama <ruiu at google.com> wrote:
> > On Mon, Apr 6, 2015 at 6:22 PM, Ed Maste <emaste at freebsd.org> wrote:
> >>
> >> Good point. Do you think we should go further and either just use
> >> DT_RUNPATH when we know the system's rtld supports it, or set both
> >> DT_RUNPATH and DT_RPATH in all cases?
> >>
> >> The "broken" DT_RPATH behaviour is as specified in gabi41.pdf, but I'm
> >> not sure that anyone really wants that. It seems a bit unfortunate to
> >> have the sensible behaviour hidden behind an option.
> >
> > How about other linkers? If gold and/or GNU LD are going to switch
> default,
> > we should follow them.
>
> There's been discussion on enabling it by default in gold, but I
> suspect it'll be hard to overcome the inertia there. I'm not sure
> about GNU ld. Perhaps if we make the change they'll use it to justify
> following suit :)
>

Nice strategy. :) But well, we should stick with what GNU LD or gold do, as
it's hard to justify the (minor) incompatibility as a feature, considering
our position. Of course we can publicly say that if they switch, we are
happy to switch default too.



> If the target supports DT_RUNPATH, and the target implements DT_RPATH
> as specified in gABI, then I think there's no real use in setting only
> DT_RPATH.
>
> FWIW most of the libraries in /usr/local/lib on my FreeBSD 11 laptop
> have both DT_RPATH and DT_RUNPATH, and they are set to the same value.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150406/7c3ba83f/attachment.html>