Re: Add support for DT_RUNPATH in ld.elf_so?

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Mon, Nov 27, 2017 at 10:50:37AM +0100, Martin Husemann wrote:
> The update of the ELF spec is like 15 years old (I think). The main point
> is addition of DT_RUNPATH, used to replace DT_RPATH, as the old spec
> for DT_RPATH was seriously broken. However, all BSDs ever implemented
> DT_RPATH differently, with the exact same semantics DT_RUNPATH now has.

The ironic part is that the Linux community can't decide what behavior
it wants. IMO it should only ever be part of the toolchain definition
and random junk should *never* use it. That includes rust.

> Other options would be to make ld(1) ignore the --enable-new-dtags option
> (either silently, with an abort, or by mailing the binutils bugs list
> about their stupid command line option design).

I dislike ignoring it silently, since it is a flag that is supposed to
have a result. I wouldn't mind erroring out on it. At least a good chunk
of users check for it first...

> So we have three choices:
> 
>  - Do nothintg and continue patching 3rd party code and fight upstream
>    to avoid them adding --enable-new-dtags on NetBSD
>  - apply the attached ld.elf_so patch and ignore the politics
>  - patch our ld(1) and try to upstream the change

I still object to the patch to compensate for broken junk. I don't mind
the third option as long as it is an erroring out. That wouldn't really
help with the first point though.

Joerg