Re: A symbol version patch for glibc 2.x compatibility

["H . J . Lu" <hjl at valinux dot com>]

On Fri, Nov 17, 2000 at 10:16:04AM -0700, law@redhat.com wrote:
> 
>   In message <20001116165311.E74170@dragon.nuxi.com>you write:
>   > On Mon, Nov 13, 2000 at 11:26:37AM -0800, H . J . Lu wrote:
>   > > I don't think it was written down anywhere. But I believe it has
>   > > come up during discussion. It goes like this, if you see something
>   > > you don't know, don't touch it. One goal for an ABI is ELF tools
>   > > from different vendors can work on the same set of ELF files. But
>   > > an OS vendor is free to add its own extensions. When you do it in
>   > > the ABI incompatible way such that the other ABI compliant ELF tools
>   > > may misprocess it, you should set the EI_OSABI bits.
>   > 
>   > In an effort to get a clear picture on the history of this I'm wondering
>   > if you know about this.  The first non-zero value seems to have been
>   > "ELFOSABI_HPUX".  Do you know how their ELF implementation (the ELF
>   > support in 11.x I assume), is non-standard such that the should be
>   > setting EI_OSABI to "ELFOSABI_HPUX".
> It's part of HP's ELF standard -- I don't immediately recall if it had any
> impact on how HP's tools functioned.

If I am right, HP didn't know what to do with DT_INIT/DT_FINI. That is
why they invented DT_INIT_ARRAY/DT_FINI_ARRAY, which was added to gABI
later. HP's ELF was different from normal ELF at the time.

-- 
H.J. Lu (hjl@valinux.com)